Showing papers in "Smart agricultural technology in 2022"

The digitization of agricultural industry – a systematic literature review on agriculture 4.0

Abstract: • SLR is conducted using PRISMA approach and148 articles are selected and critically analyzed. • The results show the extent of digital technologies adoption in agriculture. • The potential benefits of digital technologies and roadblocks hindering their implementation in agriculture sector are identified and discussed. • The study will positively impact the research around agriculture 4.0. Agriculture is considered one of the most important sectors that play a strategic role in ensuring food security. However, with the increasing world's population, agri-food demands are growing — posing the need to switch from traditional agricultural methods to smart agriculture practices, also known as agriculture 4.0. To fully benefit from the potential of agriculture 4.0, it is significant to understand and address the problems and challenges associated with it. This study, therefore, aims to contribute to the development of agriculture 4.0 by investigating the emerging trends of digital technologies in the agricultural industry. For this purpose, a systematic literature review based on Protocol of Preferred Reporting Items for Systematic Reviews and Meta-Analyses is conducted to analyse the scientific literature related to crop farming published in the last decade. After applying the protocol, 148 papers were selected and the extent of digital technologies adoption in agriculture was examined in the context of service type, technology readiness level, and farm type. The results have shown that digital technologies such as autonomous robotic systems, internet of things, and machine learning are significantly explored and open-air farms are frequently considered in research studies (69%), contrary to indoor farms (31%). Moreover, it is observed that most use cases are still in the prototypical phase. Finally, potential roadblocks to the digitization of the agriculture sector were identified and classified at technical and socio-economic levels. This comprehensive review results in providing useful information on the current status of digital technologies in agriculture along with prospective future opportunities.

A Survey on Using Deep Learning Techniques for Plant Disease Diagnosis and Recommendations for Development of Appropriate Tools

Abstract: Several factors associated with disease diagnosis in plants using deep learning techniques must be considered to develop a robust system for accurate disease management. A considerable number of studies have investigated the potential of deep learning techniques for precision agriculture in the last decade. However, despite the range of applications, several gaps within plant disease research are yet to be addressed to support disease management on farms. Thus, there is a need to establish a knowledge base of existing applications and identify the challenges and opportunities to help advance the development of tools that address farmers' needs. This study presents a comprehensive overview of 70 studies on deep learning applications and the trends associated with their use for disease diagnosis and management in agriculture. The studies were sourced from four indexing services, namely Scopus, IEEE Xplore, Science Direct, and Google Scholar, and 11 main keywords used were Plant Diseases, Precision Agriculture, Unmanned Aerial System (UAS), Imagery Datasets, Image Processing, Machine Learning, Deep Learning, Transfer Learning, Image Classification, Object Detection, and Semantic Segmentation. The review is focused on providing a detailed assessment and considerations for developing deep learning-based tools for plant disease diagnosis in the form of seven key questions pertaining to (i) dataset requirements, availability, and usability, (ii) imaging sensors and data collection platforms, (iii) deep learning techniques, (iv) generalization of deep learning models, (v) disease severity estimation, (vi) deep learning and human accuracy comparison, and (vii) open research topics. These questions can help address existing research gaps by guiding further development and application of tools to support plant disease diagnosis and provide disease management support to farmers.

Unlocking adoption challenges of IoT in Indian Agricultural and Food Supply Chain

Abstract: • The factors affecting the adoption of IoT in the agri-food domain have been identified. • A MCDM approach has been used for analysing the factors. • The cause-effect relationships amongst the factors were identified. • The most critical factors have been recognised. • This study guides the stake holders to formulate new strategies/policies The Internet of Things (IoT) can play a key role in transforming traditional agricultural sector to smart agricultural domain. However, in developing economies the adoption is in the nascent stage. This paper aims to evaluate the critical factors that impact the adoption decision of IoT technology in the Agricultural and Food Supply Chain (AFSC). Based on a comprehensive literature survey and experts’ opinion 24 critical factors were identified. The identified list of factors was categorized into technological, social, economic, and organizational categories. DEMATEL method was applied to determine the cause-effect relationship of these factors. The results underlined five significant causal factors namely lack of interoperability, environmental sustainability, trust, lack of security, and network challenges which influence the IoT adoption. The results provide unique insights in agro-food sector to improve performance by overcoming the identified key challenges. Also, this study provides a roadmap for the implementation of IoT in emerging economies. Further, this paper guides the agri-food managers, IoT service providers, and the Government to formulate new strategies/policies for the effective adoption of IoT in the agri-food sector

Crops yield prediction based on machine learning models: Case of West African countries

Abstract: Global agricultural production, in particular, is of increasing concern to the major international organizations in charge of nutrition. The rising demand for food globally due to unprecedented population growth has led to food insecurity in some populated regions such as Africa. Another contributing factor to global food insecurity is climate change and its variability. World and African agricultural production in particular are of increasing concern to the major international organizations in charge of nutrition. The World Food Program has reported that high population growth worldwide, especially in Africa in recent years, is leading to increased food security. Moreover, farmers and agricultural decision-makers need advanced tools to help them make quick decisions that will impact the quality of agricultural yields. Climate change has been a major phenomenon in recent decades all over the world. An impact of climate change has been observed on the quality of agricultural production. The arrival of big data technology has led to new powerful analytical tools like machine learning, which have proven themselves in many areas such as medicine, finance, and biology. In this work, we propose a prediction system based on machine learning to predict the yield of six crops, namely: rice, maize, cassava, seed cotton, yams, and bananas, at the country-level in the area of West African countries throughout the year. We combined climatic data, weather data, agricultural yields, and chemical data to help decision-makers and farmers predict the annual crop yields in their country. We used a decision tree, multivariate logistic regression, and k-nearest neighbor models to build our system. We had promising results with both models when using three machine learning models. We applied a hyper-parameter tuning technique throughout cross-validation to get a better model that does not face overfitting. We found that the decision tree model performs well with a coefficient of determination(R2) of 95.3% while the K-Nearest Neighbor model and logistic regression perform respectively with R2=93.15% and R2=89.78%. We also study the correlation between the predicted results and the expected results. We found that the prediction results of the decision tree model and the K-Nearest Neighbor model are correlated to the expected data, which proves the efficacy of the model.

Agricultural Decision System based on Advanced Machine Learning Models for Yield Prediction: Case of East African Countries

Abstract: Food security has become a real challenge for some organizations in charge of the food program and for the majority of countries, especially African countries. The United Nations Organizations’ has recently defined the end of hunger and the improvement of food security in 2030 as its primary goal. Improving food security could also pass through the handling of agricultural yield. Agricultural yield is affected by climate changes since this latest decade. Climate change is considered one of the major threats to agricultural development in Africa. Decision-making level and farmers need efficient analytical tools to help them in decision making. Machine learning has become an impressive predictive analytical tool for large volume of data. It has been used in many domains such as medicine, finance, sport, and recently in agriculture. In this work, we propose three crop prediction models : Crop Random Forest, Crop Gradient Boosting Machine and Crop Support Vector Machine. We combine climate data, crop production data, and pesticides data to develop a decision system based on advanced machine learning models. Despite the poor availability of data related to agriculture in Africa, we were able to propose a decision system able to predict the crop yield at the country level in fourteen East African countries. Our experimental results show that the three proposed machine learning models fit well the crop data with a high accuracy R 2 . The Root Mean Square Error ( R M S E ) and Mean Absolute Percentage Error ( M A P E ) associated to our models are very minimal because the agricultural prediction values are very close to reality. Our proposed models are reliable and generalize well the agricultural predictions in East Africa.

Digital Twins in Agriculture: A State-of-the-art review

Abstract: The Digital Twin enables the distinctions between state sensing, entity understanding and physical automation to be eliminated, through high-fidelity modelling and bi-directional data streams. The concept of real-time virtual representation places the Digital Twin in a unique position to enable digitization in agriculture. The union of data, modelling and what-if simulation can provide an approach to overcome current limitations in decision-making support and automation, across a diverse range of agricultural enterprises. This paper conducts a Systematic Literature Review of Digital Twins in agriculture, identifying current trends and open questions with the goal of increasing awareness and understanding of the Digital Twin and its possibilities.

A standalone photovoltaic/battery energy-powered water quality monitoring system based on narrowband internet of things for aquaculture: Design and implementation

Abstract: This study presents a standalone photovoltaic (PV)/battery energy storage (BES)-powered water quality monitoring system based on the narrowband internet of things (NB-IoT) for aquaculture. (1) A PV/BES system was used as the main energy system of the monitoring system. The PV and BES capacities were optimized to provide uninterrupted electrical energy to the monitoring system, taking into account two techno-economic criteria: a maximum reliability index (RI) and a minimum levelized cost of energy (LCOE). Additionally, sensitivity analyses were conducted to investigate the effects of changes in PV generation and system consumption on the RI to improve the resilience of the PV/BES system. (2) The NB-IoT-based remote monitoring system was developed to aggregate water quality parameters such as dissolved oxygen, potential of hydrogen, temperature, turbidity, and salinity in order to provide early warning of severe water quality. Subsequently, the water quality data were used to calculate the water quality suitability index (WQSI). In addition, electrical measuring devices were installed to measure relevant electrical parameters such as PV power, system consumption, BES power, and state of charge. Grafana was then used to process and visualize these water quality and electrical parameters in real-time for the end-users. The proposed system was tested at an aquaculture pond in Rayong Province, Thailand. From the energy system viewpoint, the optimal techno-economic size of the PV/BES system was determined to be a PV capacity of 50 Wp and a BES capacity of 480 Wh, with an RI of 100% and a minimum LCOE of 0.61 $/kWh. The experimental results revealed that the system could operate continuously and stably without losing power supply. Furthermore, the results demonstrated that the proposed system achieved adequate communication reliability, with a packet loss rate of 0.89%, thereby allowing for reliable near real-time monitoring of the WQSI.

Detection of Fruit Maturity Stage and Yield Estimation in Wild Blueberry Using Deep Learning Convolutional Neural Networks

Abstract: This study looked at the development of six deep learning artificial neural network models for detecting ripeness stage in wild blueberries, along with developing models for yield estimation. The six networks used were YOLOv3, YOLOv3-SPP, YOLOv3-Tiny, YOLOv4, YOLOv4-Small and YOLOv4-Tiny. Both 3-class (green berries, red berries, blue berries) and 2-class (unripe berries, ripe berries) models were developed with YOLOv4 performing the best with mean average precisions of 79.79% and 88.12% respectively. This result was further supported by YOLOv4 achieving the highest F1 score of 0.82. YOLOv4-Tiny performed the best from a computational load perspective having a mean inference time of 7.8 ms and a mean memory usage of 1.63 GB for single 1280 × 736 pixel images. Only minor differences in the accuracy of the nonlinear regression yield prediction models were detected, with YOLOv4-Small performing the best with a mean absolute error of 24.1%. Despite this error, the results are encouraging, and this novel approach to yield estimation in wild blueberries will aid growers in making better, more localized, management decisions, improving yields and ultimately increasing profits by better understanding their fields ripening characteristics.

Applications of data mining and machine learning framework in aquaculture and fisheries: A review

Abstract: • Data mining and machine learning framework offer intelligent decision-making solutions from complex aquaculture and fisheries datasets. • Aquaculture applications such as monitoring and control of production environment, fish biomass and optimization of feed use are discussed with examples. • Fisheries management applications include surveillance of fishing, catch composition and ecosystem-fisheries associations. • Applications related to environment monitoring, fish processing and marketing are also indicated, along with challenges and perspectives. Aquaculture and fisheries sectors are finding ingenious ways to grow and meet the soaring human demand for nutrient-rich fish and seafood by efficiently utilizing the vast water resources and biodiversity of aquatic life on earth. This includes the progressive integration of information technology, data science and artificial intelligence with fishing and fish farming methods to enable intensification of aquaculture production, sustainable exploitation of natural fishery resources and mechanization-automation of allied activities. Exclusive data mining and machine learning systems are being developed to process complex datasets and perform intelligent tasks like analysing cause-effect associations, forecasting problems and providing smart-precision solutions for farming and catching fish. Considering the intensifying research and growing interest of stakeholders, in this review, we have consolidated basic information on the various practical applications of data mining and machine learning in aquaculture and fisheries domains from representative selection of scientific literature. This includes an overview of research and applications in (1) aquaculture activities such as monitoring and control of the production environment, optimization of feed use, fish biomass monitoring and disease prevention; (2) fisheries management aspects such as resource assessment, fishing, catch monitoring and regulation; (3) environment monitoring related to hydrology, primary production and aquatic pollution; (4) automation of fish processing and quality assurance systems; and (5) fish market intelligence, price forecasting and socioeconomics. While aquaculture has been relatively faster in integrating data mining and machine learning tools with advanced farming systems, capture fisheries is finding reliable methods to sort the complexities in data collection and processing. Finally, we have pointed out some of the challenges and future perspectives related to large-scale adoption.

Why the low adoption of robotics in the farms? Challenges for the establishment of commercial agricultural robots

Abstract: In a world that requires sustainable agricultural practices, quantitative plant-by-plant and field status monitoring can be advantageous to all farmers by optimizing their farm management. Autonomous robots, sensing technologies, and automated data analysis will play a key role. In the present article, we discuss three critical aspects of the incorporation of field robots in agriculture: 1) the main design specifications and economic aspects for Commercial Agricultural Robots (CARs) in terms of autonomous navigation (perception, localization, and kinematics); 2) the business models of agricultural robotics companies, which are compared to successful business models of robotic companies on the medical field; 3) the possibilities that CARs could bring, including the generation of big data about local food production and its impact on the local environment. Our analysis highlights the reasons that explain the low adoption of CARs in the current agricultural market. Finally, we reflect on the current distorted costs that consumers pay for food while the ecological footprints of food production and food delivery are neglected, and we encourage the discussion to promote sustainable food policies that support the establishment of robotic agricultural companies.

Assessment of Mechanical Damage to Flaxseeds Using Radiographic Imaging and Tomography

Abstract: The emergence of flax as a nutritional superfood has brought to the forefront the challenges its producers face in the crop's post-harvest quality preservation. Mechanical damage during harvesting, handling, and transportation can severely impair flaxseed quality causing economic losses to producers and processors. Therefore, this study explored the effect of moisture content and impact stress on the breakage susceptibility of this important oilseed. Impact energies (IEs) of 0 mJ, 2 mJ, 4 mJ, and 6 mJ were imparted to flaxseed samples at moisture contents (MCs) of 6%, 8%, and 11.5% and internal as well as external damage to seeds was assessed using 2D X-ray imaging. Furthermore, 3D X-ray tomography was deployed to characterize the nature of damage to the internal seed tissue. Our results indicate that increasing the IE and decreasing MC (specifically under low IE) leads to higher breakage susceptibility in flaxseeds. Furthermore, a regression analysis of the external and internal damage parameters indicated that while external and internal damage to seeds is closely correlated at high IE (6 mJ), the same relationship doesn't always hold at medium to low impact (IE ≤ 4 mJ). To evaluate the extent of internal damage, the gray levels distribution of seeds’ 2D X-ray images were compared, and percentile scores were determined as a promising candidate. Support Vector Machine (SVM) and Linear Discriminant Analysis (LDA) classifiers were employed to classify seeds into two broad groups of nil/low and medium/high damage using the percentile scores. The SVM and LDA classifiers achieved 87.2% and 79.6% classification accuracies, respectively. The developed model confirms that radiographic imaging has the potential to detect mechanical damages of seeds in a rapid, automated, and non-destructive fashion.

A two-stage deep-learning based segmentation model for crop disease quantification based on corn field imagery

Abstract: It is important to develop accurate disease management systems to identify and segment corn disease lesions and estimate their severity under complex field conditions. Although deep learning techniques are becoming increasingly popular to identify singular diseases, access to robust models for identifying multiple diseases and segmenting lesion areas for severity estimation under field conditions remain unsolved. In this study, a custom dataset consisting of handheld images of corn leaves infected with Gray Leaf Spot (GLS), Northern Leaf Blight (NLB), and Northern Leaf Spot (NLS) diseases, acquired under field conditions, was used to develop a novel two-stage semantic segmentation approach for identifying corn diseases and estimate their severity. Three semantic segmentation models were trained for each stage using SegNet, UNet, and DeepLabV3+ network architectures. Stage one used semantic segmentation to extract leaves from complex field backgrounds. In stage two, semantic segmentation was used to locate, identify, and calculate area coverage for disease lesions. After the models were trained, the best performance for stage one was observed from the UNet model, which achieved up to 0.9422 mean weighted intersection over union (mwIoU) and 0.8063 mean boundary F1-score (mBFScore). The best performance for stage two was observed from the DeepLabV3+ model, which could identify the disease lesions with a mwIoU of 0.7379 and mBFScore of 0.5351. Finally, severity was estimated by calculating the percentage of leaf area covered by disease lesions. In the test set, an R2 value (coefficient of determination) of 0.96 was achieved, which denotes that the integrated (UNet-DeepLabV3+) model predicted the severity of three diseases very close to the actual observations. This study developed a novel two-stage deep learning-based approach to accurately identify three targeted corn diseases and estimate their severity to pave the way for developing a field-worthy disease management system.

Efficacy of cotton harvest aids applications with unmanned aerial vehicles (UAV) and ground-based field sprayers – A case study comparison

Abstract: The opportunity of using of Unmanned Aerial Vehicles (UAVs) for applying harvest aids in ultra-low volumes (ULV) in cotton is an interesting perspective to overcome common problems associated with traditional ground-based applications like crop mechanical damage, yield losses and soil compaction. Moreover, as cotton harvest aids induce a prominent effect on the appearance of the crop, their efficacy can be conceptualized in wide scale remote sensing observations to provide general insights on the performance of UAV plant protection applications. The scope of the present study was to evaluate the efficacy of UAV spraying in cotton in comparison with traditional ground-based applications. Two field trials were established for that scope and each field included eight treatments. Two treatments involved ground based, field sprayer applications with and without air assistance on the boom and six treatments, combinations of multirotor UAV applications involving two spray altitudes, (2 m and 3 m above the canopy), two spray volumes (16 l ha −1 and 10 l ha −1 ) and the addition of a spray adjuvant on the mixture. The efficacy of the alternative treatments was evaluated through UAV remote sensing and by ground truth measurements. The evaluated parameters were the differences obtained in the Normalized Difference Vegetation Index - NDVI and the Cotton Fiber Index – CFI, from time of spraying to harvest, the defoliation rate, the boll opening rate and the final yield. The results revealed that UAV sprayings were more efficient from both field sprayer applications (with and without air assistance). The higher efficacy was accompanied by an improvement in cotton yield. The best results were obtained from the low altitude, 2 m UAV operation. There were no clear differences between the high volume and the low volume UAV applications implying that the lower 10 l ha −1 volume might be sufficient and provide the potential to increase field capacity. The addition of the spray adjuvant was beneficial only when spraying was followed by rainfalls. The results of the present study implicate that UAV applications can support a wide range of plant protection and plant care applications in cotton.

ShrimpChain: A Blockchain-Based Transparent and Traceable Framework to Enhance the Export Potentiality of Bangladeshi Shrimp

Abstract: • This study proposed the ShrimpChain , a public-private hybrid blockchain-based conceptual framework. • We proposed a scoring-based certification method for supply chain stakeholders of shrimp. • From the pre-production to the final packaging stage, relevant data will be entered via mobile/web app or Internet of Things (IoT) devices to the blockchain network. • It enables the detection of wrongdoings across the shrimp supply chain and encourages traceability and food safety. Despite substantial progress achieved with shrimp production during the last three and a half decades, growth of shrimp export remains minimal in Bangladesh. In the absence of effective traceability and transparency practices, a wide range of malpractice occurs both at the production and post-harvest stages. Traditional paper-based record-keeping methods for the shrimp supply chain are disparate and, therefore, cannot provide efficient traceability capacity and holistic view of the supply chain. These limit the identification of any issues in earlier juncture to proactively ensure food safety, best practice, and good governance. Addressing this multiplexed challenge, we present here ShrimpChain , a public-private hybrid blockchain-based conceptual framework. Focusing on the export market and utilizing existing technologies, the conceptual framework addresses the traceability, transparency, and certification challenges associated with shrimp export. In this framework, from the post-larva purchasing to the final packaging stage, relevant data for every stage will be entered by the associated actors via mobile/web app or Internet of Things devices to the blockchain network. Data authentication will be achieved by a novel approach of incorporating community consensus in conjunction with the machine-derived data entry timestamping method. Instead of the traditional central and endpoint certification approach, we propose a distributed and accumulative score-based certification approach that will grade packaged shrimps according to the completeness and accuracy of the authenticated data entered during different stages. Such distributed approach of certification will enhance not only food safety but also the quality and compliance to best practices. Most importantly, engaging shrimp farmers in the safety and quality assurance as well as to the certification process will empower them to have better control over the market and incentive to produce high-quality shrimp for high-value market.

A Review of Machine Learning Techniques for Identifying Weeds in Corn

Abstract: Weeds pose a major challenge in achieving high yield production in corn. The use of herbicides although effective can be expensive and their excessive use poses ecological concerns and herbicide resistance. Precise identification of weeds using Machine Learning (ML) models significantly reduces the use of herbicides. In this study, we provide a brief overview of the important ML methods used for identifying weeds in corn i.e., classification and object detection. The various metrics that are used for the evaluation of the performance of ML methods are also discussed. In the end, we identify some important research gaps which warrant future investigation. Most ML methods for the identification of weeds use digital images as input data, however, in some cases, hyperspectral data were used. Most of the current studies employ support vector machines and neural networks for the identification of weeds. Classification accuracy and F1 score are the two most frequently used accuracy metrics to evaluate the performance of ML models used. Future research on the identification of weeds may focus on improving the data volume using data augmentation, transfer learning to benefit from existing models, and interpretability of neural networks to avoid overfitting and make models more transparent.

Digital Twins: A Novel Traceability Concept for Post-Harvest Handling

Abstract: Digital Twins are a novel approach to systems engineering that can help control complex environments and interface humans with them. This is achieved by digitally mirroring a physical asset to provide historical data, monitoring, and predictions of future states. While there are a few applications of Digital Twins to agriculture, none exist for post-harvest grain handling. However, there have been past attempts at integrating computer assistance in grain quality, called expert systems. These systems were largely abandoned due to their inability to keep operators in the loop and the inadequacy of sensors available during the time of expert systems research. By utilizing Digital Twins and modern post-harvest sensors, operators can be provided with a digital representation of inventory and the quality of grain as it moves throughout a facility. This virtual representation also presents a unique opportunity to enhance market traceability. This review focuses on (1) expert systems, their history, and limitations, (2) the history of Digital Twins and their applicability to grain storage and handling, (3) unit operations and the sensors that are common to grain handling facilities, (4) mathematical and computer models to simulate grain handling operations, and (5) a conceptualization of post-harvest Digital Twins, which identifies research gaps where critical questions should be answered if Digital Twins technology is to be considered a logical contender for traceability of commodities post-harvest.

Eden Library: A long-term database for storing agricultural multi-sensor datasets from UAV and proximal platforms

Abstract: In modern agriculture, visual recognition systems based on deep learning are arising to allow autonomous machines to execute field operations in crops. However, for obtaining high performances, these methods need high amounts of data, which are usually scarce in agriculture. The main reason is that building an agricultural dataset covering exhaustively a specific problem is challenging, as visual characteristics of the symptoms may change, and there is a high dependency on environmental factors, such as temperature, humidity and light conditions. Therefore, an efficient methodology is necessary to consistently cover the entire workflow for creating an agricultural dataset, from the image acquisition to its online publication. This paper presents the Eden Library, a platform for contributing to this existing gap in open access crop/plant databases covering proximal and aerial images. The complete workflow on the design and deployment of the platform is also explained and discussed. This workflow is relevant because the provided datasets are thought to be maintained and enriched along the time, and they do not just remain as a static research output covering only specific species, growth stages, and conditions. The image annotations of plants and symptoms are provided, saving users from manually annotating images. Currently, the Eden Library covers 15 different crops, 9 weeds and 30 disorders (pests, diseases and nutrient deficiencies). Eden Library aspires to close this gap by providing a large and diversified image collection of plants, organized in a consistent manner, in order to boost further vision-based and AI-enabled field applications.

Data-driven model predictive control for precision irrigation management

Abstract: • A review of data driven model predictive control for precision irrigation management. • A detailed discussion on model predictive control. • Data driven modelling and system identification. • Applications of model predictive control in precision agriculture. The future of agriculture faces a threat from a changing climate and a rapidly growing population. This has put enormous pressure on water and land resources as more food is expected from less inputs. Advancement in smart agriculture through the use of the Internet of Things and improvement in computational power has enabled extensive data collection from agricultural ecosystems. This review introduces model predictive control and describes its application in precision irrigation. An overview of the application of data-driven modelling and model predictive control for precision irrigation management is presented. Model predictive control has been applied in irrigation canal control, irrigation scheduling, stem water potential regulation, soil moisture regulation and prediction of plant disturbances. Finally, the benefits, challenges, and future perspectives of data-driven model predictive control in the context of irrigation scheduling are presented. This review provides useful information to researchers and agriculturalists to appreciate and use data collected in real-time to learn the dynamics of agricultural systems.

An End-Effector for Robotic Cotton Harvesting

Abstract: Cotton, a major crop worldwide, is harvested in mechanized production systems once at the end of the growing season. To facilitate harvest and maximize fiber quality, the plants are typically defoliated when about 60% of the cotton bolls are open. Due to non-uniform maturation, the bolls that have opened early expose their fiber to weather until harvest, commonly for weeks, degrading fiber quality. Furthermore, high capacity harvesting machines are heavy, potentially compacting the soil that in turn reduces hydraulic conductivity in the wheel tracks and reducing yield. Robotic harvesting with smaller machines brings about the possibility of multiple harvests during the growing season while enabling them to pick the seed cotton soon after the boll opens, preserving fiber quality. Smaller machines would also be less likely to substantially compact the soil. Therefore, research has been conducted to enumerate and address multiple challenges associated with the design of a robotic cotton harvester. The particular focus of the research reported herein was on the design of a robotic end-effector for picking seed cotton from the open boll of a non-defoliated cotton plant. Various design concepts were considered, and some were built as prototypes and experimentally assessed. The design was selected as optimal was: a three-finger, moving pinned belt, underactuated end-effector. A refined prototype of the end-effector was indoor tested on a robotic platform with a computer-controlled three-degree-of-freedom manipulator. The end-effector could pick 66-85% of the seed cotton from a boll with a picking time of 4 s for a simple and less efficient system to 18 s for a controlled-movement and more efficient system. Further implications of this study will include adding a depth sensor on the robot to detect and localize cotton bolls and manipulate arm autonomously.

A Comparison of Machine-Learning Assisted Optical and Thermal Camera Systems for Beehive Activity Counting

Abstract: There is a documented shortage of reliable counting systems for the entrance of beehives. Movement at the entrance of a hive is a measure of hive health and abnormalities, in addition to an indicator of predators. To that end, two camera systems have been designed to provide a comparative analysis for a thermal camera system. The first, a visible spectrum camera, competed directly with the thermal camera. Machine learning is used to address the narrower field of view of the thermal camera, in addition to lost extracted tracks from both cameras. K-nearest-neighbour, support vector machine, random forest, and neural networks are used to classify flights as arriving, departing, or hovering bees. A hierarchical system is used to determine the nature of any flights where a clear label is not feasibly assigned based on the information from either test camera. A third camera at distance from the hive served as the end authority. After three iterations of training and validating, a test case is evaluated between both camera systems. Results from the test are compared to those from a human observer, showing that the thermal camera can perform with the same success as the visual camera despite a smaller field of view, fewer pixels and lower frame-rate, while both systems achieve greater than 96% accuracy and both camera systems are 93% successful at extracting flights. This is advantageous as a thermal camera will work in a wider range of environments, keeping the accuracy of an optical camera, and predicting based on movement characteristics will allow expanded uses such as predicting the presence of predators.

LoRaWAN based Internet of Things (IoT) System for Precision Irrigation in Plasticulture Fresh-market Tomato

Abstract: Precision irrigation with sensors has proven to be effective for water saving in crop production. An Internet of Things (IoT) system is necessary for monitoring real-time data from sensors and automating irrigation systems. Long-range wide-area network (LoRaWAN), a type of low-power wide-area network (LPWAN), is a low-cost and easily implemented IoT system which can be used for precision crop irrigation. In this study, a LoRaWAN based IoT system was developed and evaluated as a precision irrigation tool on fresh-market tomato production in a plasticulture system. Four irrigation scheduling treatments were designed and tested, including irrigation based on crop evapotranspiration (ET), soil matric potential sensors (Watermark 200SS-5) at -60 kPa (MP60) or -40 kPa (MP40), and GesCoN fertigation decision support system (DSS). The treatments were arranged based on a randomized complete block design with four replicates. System feasibility, irrigation water use efficiency (iWUE), and crop yield were evaluated. Throughout the season, the overall water use efficiencies were 22.2, 26.5, 27.9, and 28.4 kg·m−3 for ET, MP60, MP40, and GesCoN, respectively. The results indicated that treatment MP60 and GesCoN had 15.2 and 22.1% higher marketable fruit yield than ET, while, MP40 had 12.5% lower marketable yield compared to ET. Overall, the LoRaWAN-based IoT system performed well in terms of power consumption, communication, sensor reading, and valve control. These results suggested that the IoT system can be implemented for precision and automatic irrigation operations for vegetable and other horticultural crops with enhancing crop's water use efficiency and sustainability.

Detecting olive grove abandonment with Sentinel-2 and machine learning: the development of a web-based tool for land management.

Abstract: The abandonment of rural areas is an important environmental and socio-economic issue in Europe, threatening the stability and profitability of agricultural production. The identification and quantification of abandoned land is key for temporal and spatial monitoring of the process and for applying alternative management measures. Italy is one of the most important European countries for the production of high quality olive oil, accounting for a large slice of the current certificated production (i.e., PDO, PGI). In this study, we present a machine learning model (i.e., Random Forest) for the identification of abandoned olive tree groves using field observations and NDVI time series, tested in a typical agroecosystem in central Italy dominated by olive groves. An application for smartphones able to record the geographic position was developed and used to collect field points, which in turn were utilised to train the model. The data of NDVI from January to December 2020, calculated on Sentinel-2 images, were extracted for each monitoring point and gap-filled to obtain a 10-days interval time series. The Random Forest model used the annual NDVI time series as features and classified the sampling points in the test dataset with an accuracy of 0.85. The model showed a higher capacity of classifying cultivated than abandoned points, sensitivity being equal to 0.88 and specificity equal to 0.82. Results demonstrated the applicability of the combined approach for discriminating cultivated from abandoned olive tree groves, in case that the parcels destined for olive tree cultivation are known. A web-based tool was implemented to support land monitoring and management.

Semantic segmentation of in-field cotton bolls from the sky using deep convolutional neural networks

Abstract: • Performed sematic segmentation of cotton bolls and sky. • Fully convolutional neural networks were used as encoders. • Models successfully discriminate between the pixels of cotton bolls and sky. • Achieved a maximum IoU of 84.5% and 80.67% for cotton bolls and sky, respectively. Manually picking of cotton bolls is a tedious, costly, and labor-intensive task, while harvesting using machines results in higher harvesting losses. By keeping selective picking in mind to maintain fiber quality, minimize harvesting losses, and tackle the shortage of farm labor in near future, cotton harvesting robots seem to be a better alternative in coming years in both developing and developed countries. For the cotton harvesting robot, cotton boll recognition with minimum errors is a foremost and challenging task. While recognizing cotton bolls, false-positive errors occur due to sky interference. In present study, convolutional neural networks were used to segment and discriminate the cotton bolls pixels from sky pixels. For that, three fully convolutional neural networks namely VGG16, InceptionV3, and ResNet34 were used as encoders and trained. These trained neural networks models were evaluated using the intersection-over-union (IoU), F1-score, precision, and recall metrics. All proposed models were tested on a cotton-sky dataset and achieved an IoU score of above 81% and 80% for cotton bolls and sky, respectively. InceptionV3 model outperforms with an IoU score of 84.5% and 80.67% for cotton bolls and sky, respectively with a segmentation time of 1.07 s. For the cotton dataset, proposed models achieved an IoU score of above 90% for cotton bolls and the InceptionV3 model outperforms with an IoU score of 93.29%. It can be concluded that the InceptionV3 model segmented cotton bolls and sky with higher accuracy, and low error rates and, hence can be deployed to cotton harvesting robots effectively.

A lightweight network for mummy berry disease recognition

Abstract: Mummy berry disease caused by Monilinia vaccinii-corymbosi (Reade) occurs during the productive season of blueberry plants. In severe cases, it will cause a huge decline in blueberry yield and cause significant economic losses. The correct identification of mummy berry disease helps growers to take timely preventive measures, which can limit the further spread of mummy berry disease and reduce the damage to blueberries. In this paper, aiming at identifying mummy berry disease in a real environment, we propose a lightweight network model that can be deployed on mobile or embedded devices without worrying about the limitation of computational cost and memory storage capacity. This model selects MobileNet V1 as the basic network. The core convolution layer adopts a multi-scale feature extraction module proposed by us, namely MSFE module, which combines the dilated convolution and the depthwise separable convolution in a parallel manner. At the end of the model, the feature filtering module FFM based on channel attention mechanism is used to improve the classification performance of the model. The parameter size of the model is only 2.61 million, which is about 19.2% lower than that of MobileNet V1. Experiments show that the model has achieved a good classification effect on the mummy berry disease data set, with a test accuracy of 99.33%, which is 3.17% higher than the MobileNet V1 test accuracy.

Spatial modeling via geostatistics and infrared thermography of the skin temperature of dairy cows in a compost barn system in the Brazilian semiarid region

Abstract: The aim of the study was i) to determine the spatial variability of the skin temperature of dairy cows in a compost system and ii) to evaluate the thermal comfort conditions of confined animals. Skin temperature (Tskin,°C) was recorded in three animals at 9:00 am (morning) and the measurement was repeated at 3:00 pm (afternoon) using a thermographic camera (model FLIR i60). Tskin data were submitted to descriptive statistics, geostatistics, and later elaborated kriging maps. When comparing the Tskin mean at 9:00 am and 3:00 pm, for animal 3, there was a reduction of 0.30°C in the afternoon shift, as a result of the individual's position concerning the forced ventilation system. The use of the geostatistics technique made it possible to verify the occurrence of a strong spatial dependence between the skin temperatures of cows raised inside a facility with a compost barn system, with the Gaussian model showing the best fit of the data (R2>0.90), for the morning and afternoon recording shifts. The characterization of the spatial variability of the Tskin of dairy cows allowed us to identify the comfort conditions of the animals housed in the compost barn systems.

Performance Evaluation of Deep Learning Object Detectors for Weed Detection for Cotton

Abstract: Alternative non-chemical or chemical-reduced weed control tactics are critical for future integrated weed management, especially for herbicide-resistant weeds. Through weed detection and localization, machine vision technology has the potential to enable site- and species-specific treatments targeting individual weed plants. However, due to unstructured field circumstances and the large biological variability of weeds, robust and accurate weed detection remains a challenging endeavor. Deep learning (DL) algorithms, powered by large-scale image data, promise to achieve the weed detection performance required for precision weeding. In this study, a three-class weed dataset with bounding box annotations was curated, consisting of 848 color images collected in cotton fields under variable field conditions. A set of 13 weed detection models were built using DL-based one-stage and two-stage object detectors, including YOLOv5, RetinaNet, EfficientDet, Fast RCNN and Faster RCNN, by transferring pretrained the object detection models to the weed dataset. RetinaNet (R101-FPN), despite its longer inference time, achieved the highest overall detection accuracy with a mean average precision ([email protected]) of 79.98%. YOLOv5n showed the potential for real-time deployment in resource-constraint devices because of the smallest number of model parameters (1.8 million) and the fastest inference (17 ms on the Google Colab) while achieving comparable detection accuracy (76.58% [email protected]). Data augmentation through geometric and color transformations could improve the accuracy of the weed detection models by a maximum of 4.2%. The software programs and the weed dataset used in this study are made publicly available (https://github.com/abdurrahman1828/DNNs-for-Weed-Detections; www.kaggle.com/yuzhenlu/cottonweeddet3).

Nutrients Deficiency Diagnosis of Rice Crop by Weighted Average Ensemble Learning

Abstract: Rice is one of the most extensively cultivated food crops on the planet, especially in Bangladesh, China, and India. However, rice production is frequently hampered by nutrient imbalances. The leaves of rice plants often show signs of nutritional shortages. As a result, rice leaf color and shape can be utilized to detect nutrient deficits. Computer vision-based automatic nutrient deficiency detection by image processing has become prevalent in agriculture. In this research, we have proposed a robust Deep Ensemble Convolutional Neural Network (DECNN) model that can diagnose rice nutrient deficiency with high accuracy. Different pre-trained models named InceptionV3, InceptionResNetV2, DenseNet121, DenseNet169, and DenseNet201 are reformed by adding various layers, and their diagnostic accuracy is observed on the Kaggle dataset. Using appropriate data augmentation, a proper dense layer, a pooling layer, and a dropout layer, each of the models improves its prediction accuracy, precision, recall, and F1 score. Among the five modified pretrained models, the modified DensNet169 model provides the highest test accuracy, which has improved from 92% to 96.66%. Finally, we ensembled the modified DenseNet169, DenseNet201, and InceptionV3 models based on their performance in detecting rice nutrient deficiency diagnosis via weighted averaging. This proposed DECNN model improves testing accuracy by 98.33%.

Field Evaluation of a Deep Learning-based Smart Variable-Rate Sprayer for Targeted Application of Agrochemicals

Abstract: The field performance of a newly developed novel smart variable-rate sprayer was evaluated. The sprayer uses convolutional neural networks (CNNs) for target detection and spot-applications of agrochemicals within potato (Solanum tuberosum L.) fields attacked by lamb's quarters (Chenopodium album L.) and corn spurry (Spergula arvensis L.) weeds and the early blight potato disease caused by Alternaria solani Sorauer. There was a non-significant effect of treatment conditions (i.e., cloudy, partly cloudy, and sunny) on spray volume during weed and diseased plant detection experiments (p-value = 0.93 and 0.75, respectively) showing that the smart sprayer performed well during all treatment conditions. There was a significant effect of spraying application techniques on the use of spray volume (p-value ≤ 0.05) reflecting a significant saving of spraying liquid during variable-rate application (VA). On average, the sprayer reduced spray volume by 47 and 51% for weeds and diseased plant detection experiments as compared to the values of chemicals applied at constant-rate application (CA), respectively, under all treatment conditions. The analysis of water-sensitive papers (WSP) data resulted in non-significant differences between CA and VA under all field conditions. These results suggest that this sprayer has a great potential to get a suitable spot application of agrochemicals and reduce the use of plant protection products thereby ensuring farm profits and environmental stewardship.

Time-series analysis of Sentinel-2 satellite images for sunflower yield estimation

Abstract: • Crop yield estimation at field scale for food security, farmers and decision-makers using remote sensing and GIS technologies based Random Forest Regression, a machine learning technique. • Sunflower yield can be predicted 85–105 d into the growing season, at the flowering stage which was found more sensitive time for a prediction. • RFR can accurately predict sunflower yield with high accuracy using Sentinel-2 spectral bands. Accurate estimates and predictions of sunflower crop yields at the pixel and field level are critically important for farmers, service dealers, and policymakers. Several models based on remote sensing data have been developed in yield assessment, but their robustness—especially in small field scale areas—needs to be examined. Here we aim to develop a robust methodology for estimation/prediction of sunflower yield at pilot field scale using Sentinel-2 remote sensing satellite imagery. We conducted the study in Mezőhegyes, south-eastern Hungary. The Random Forest Regression (RFR), a machine learning technique was used in this research to translate the Sentinel-2 spectral bands to sunflower yield based on crop yield data provided by a combine harvester equipped with a yield-monitoring system. Sentinel-2 images obtained from April to September were used to find the best image for prediction. The satellite image acquired on June 28 was found best and considered further for prediction sunflower yield. A developed training model was tested and validated in 10 different parcels to evaluate the performance of the prediction. We examined the results of the prediction model (predicted) against the actual yield data (observed) collected by a combine harvester. The results demonstrated that using 10 spectral bands from Sentinel-2 imagery the best time to predict sunflower yields was between 85–105 d into the growing season during the flowering stage. This model achieved high accuracy with low normalized root means square error (RMSE) ranging from 121.9 to 284.5 kg/ha for different test fields. Our results are promising because they prove the possibility of predicting sunflower grain yield at the pixel or field level, 3–4 months before the harvest, which is crucial for planning food policy.

Evaluation of cameras and image distance for CNN-based weed detection in wild blueberry

Abstract: Agricultural herbicide application efficiency can be improved using smart sprayers which provide site-specific, rather than broadcast, applications of agrochemicals. The YOLOv3-Tiny convolutional neural network (CNN) was trained to detect two weeds, hair fescue and sheep sorrel, in images captured from wild blueberry fields throughout Nova Scotia, Canada. An evaluation was performed in three commercial wild blueberry fields in Nova Scotia to examine the effects of camera selection and target distance on detection accuracy. A Canon T6 DSLR camera, an LG G6 smartphone, and a Logitech c920 webcam were used to capture RGB images at varying distances from target weeds. Mean F1-scores for each combination of camera and image height were analysed in a 3 × 3 factorial arrangement for hair fescue and a 3 × 2 factorial arrangement for sheep sorrel. Images captured from 0.98 m with the LG G6 and Canon T6 produced F1-scores of up to 0.97 for detection of at least one hair fescue tuft. Images captured with the LG G6 and Canon T6 DSLR from 0.57 m achieved F1-scores of 0.94 and 0.93, respectively, for detection of at least one sheep sorrel plant per image. Sheep sorrel was undetectable in images from the Logitech c920 under 19 of 27 parameter combinations. Future work will involve using the CNN to control herbicide applications with a real-time smart sprayer. Additionally, the CNN will be used in a web-based application to detect target weeds and provide site-specific information to aid management decisions. Using a CNN to detect weeds will create improvements in management techniques, resulting in cost-savings and greater sustainability for the wild blueberry industry.