How deep learning extracts and learns leaf features for plant classification

Agriculture plays a vital role in the economic growth of any country. With the increase of population, frequent changes in climatic conditions and limited resources, it becomes a challenging task to fulfil the food requirement of the present population. Precision agriculture also known as smart farming have emerged as an innovative tool to address current challenges in agricultural sustainability. The mechanism that drives this cutting edge technology is machine learning (ML). It gives the machine ability to learn without being explicitly programmed. ML together with IoT (Internet of Things) enabled farm machinery are key components of the next agriculture revolution. In this article, authors present a systematic review of ML applications in the field of agriculture. The areas that are focused are prediction of soil parameters such as organic carbon and moisture content, crop yield prediction, disease and weed detection in crops and species detection. ML with computer vision are reviewed for the classification of a different set of crop images in order to monitor the crop quality and yield assessment. This approach can be integrated for enhanced livestock production by predicting fertility patterns, diagnosing eating disorders, cattle behaviour based on ML models using data collected by collar sensors, etc. Intelligent irrigation which includes drip irrigation and intelligent harvesting techniques are also reviewed that reduces human labour to a great extent. This article demonstrates how knowledge-based agriculture can improve the sustainable productivity and quality of the product.

/pdf/machine-learning-applications-for-precision-agriculture-a-595xsgydlw.pdf

Machine Learning Applications for Precision Agriculture: A Comprehensive Review

Robotic weed control has seen increased research of late with its potential for boosting productivity in agriculture. Majority of works focus on developing robotics for croplands, ignoring the weed management problems facing rangeland stock farmers. Perhaps the greatest obstacle to widespread uptake of robotic weed control is the robust classification of weed species in their natural environment. The unparalleled successes of deep learning make it an ideal candidate for recognising various weed species in the complex rangeland environment. This work contributes the first large, public, multiclass image dataset of weed species from the Australian rangelands; allowing for the development of robust classification methods to make robotic weed control viable. The DeepWeeds dataset consists of 17,509 labelled images of eight nationally significant weed species native to eight locations across northern Australia. This paper presents a baseline for classification performance on the dataset using the benchmark deep learning models, Inception-v3 and ResNet-50. These models achieved an average classification accuracy of 95.1% and 95.7%, respectively. We also demonstrate real time performance of the ResNet-50 architecture, with an average inference time of 53.4 ms per image. These strong results bode well for future field implementation of robotic weed control methods in the Australian rangelands.

/pdf/deepweeds-a-multiclass-weed-species-image-dataset-for-deep-1e3dau5ae3.pdf

DeepWeeds: A Multiclass Weed Species Image Dataset for Deep Learning.

The world population growth is increasing the demand for food production. Furthermore, the reduction of the workforce in rural areas and the increase in production costs are challenges for food production nowadays. Smart farming is a farm management concept that may use Internet of Things (IoT) to overcome the current challenges of food production. This work uses the preferred reporting items for systematic reviews (PRISMA) methodology to systematically review the existing literature on smart farming with IoT. The review aims to identify the main devices, platforms, network protocols, processing data technologies and the applicability of smart farming with IoT to agriculture. The review shows an evolution in the way data is processed in recent years. Traditional approaches mostly used data in a reactive manner. In more recent approaches, however, new technological developments allowed the use of data to prevent crop problems and to improve the accuracy of crop diagnosis.

/pdf/a-systematic-review-of-iot-solutions-for-smart-farming-5ct1bbwayq.pdf

A Systematic Review of IoT Solutions for Smart Farming

The digital transformation of agriculture has evolved various aspects of management into artificial intelligent systems for the sake of making value from the ever-increasing data originated from numerous sources. A subset of artificial intelligence, namely machine learning, has a considerable potential to handle numerous challenges in the establishment of knowledge-based farming systems. The present study aims at shedding light on machine learning in agriculture by thoroughly reviewing the recent scholarly literature based on keywords’ combinations of “machine learning” along with “crop management”, “water management”, “soil management”, and “livestock management”, and in accordance with PRISMA guidelines. Only journal papers were considered eligible that were published within 2018–2020. The results indicated that this topic pertains to different disciplines that favour convergence research at the international level. Furthermore, crop management was observed to be at the centre of attention. A plethora of machine learning algorithms were used, with those belonging to Artificial Neural Networks being more efficient. In addition, maize and wheat as well as cattle and sheep were the most investigated crops and animals, respectively. Finally, a variety of sensors, attached on satellites and unmanned ground and aerial vehicles, have been utilized as a means of getting reliable input data for the data analyses. It is anticipated that this study will constitute a beneficial guide to all stakeholders towards enhancing awareness of the potential advantages of using machine learning in agriculture and contributing to a more systematic research on this topic.

Machine Learning in Agriculture: A Comprehensive Updated Review.

Deep neural networks (DNNs) have recently achieved remarkable performance in a myriad of applications, ranging from image recognition to language processing. Training such networks on graphics processing units (GPUs) currently offers unmatched levels of performance; however, GPUs are subject to large-power requirements. With recent advancements in high-level synthesis (HLS) techniques, new methods for accelerating deep networks using field programmable gate arrays (FPGAs) are emerging. FPGA-based DNNs present substantial advantages in energy efficiency over conventional CPU- and GPU-accelerated networks. Using the Intel FPGA software development kit (SDK) for OpenCL development environment, networks described using the high-level OpenCL framework can be accelerated targeting heterogeneous platforms including CPUs, GPUs, and FPGAs. These networks, if properly customized on GPUs and FPGAs, can be ideal candidates for learning and inference in resource-constrained portable devices such as robots and the Internet of Things (IoT) edge devices, where power is limited and performance is critical. Here, we introduce GPU- and FPGA-accelerated deterministically binarized DNNs, tailored toward weed species classification for robotic weed control. Our developed networks are trained and benchmarked using a publicly available weed species dataset, named DeepWeeds, which include close to 18 000 weed images. We demonstrate that our FPGA-accelerated binarized networks significantly outperform their GPU-accelerated counterparts, achieving a>7-fold decrease in power consumption, while performing inference on weed images 2.86 times faster compared to our best performing baseline full-precision GPU implementation. These significant benefits are gained whilst losing only 1.17% of validation accuracy. In this paper, this is a significant step toward enabling deep inference and learning on IoT edge devices, and smart portable machines such as agricultural robots, which is the target application.

Low-Power and High-Speed Deep FPGA Inference Engines for Weed Classification at the Edge

Histograms of Oriented Gradients (HOGs) have proven to be a robust feature set for many visual object recognition applications. In this paper we investigate a simple but powerful approach to make use of the HOG feature set for in situ leaf classification. The contributions of this work are threefold. Firstly, we present a novel method for segmenting leaves from a textured background. Secondly, we investigate a scale and rotation invariant enhancement of the HOG feature set for texture based leaf classification - whose results compare well with a multi-feature probabilistic neural network classifier on a benchmark data set. And finally, we introduce an in situ data set containing 337 images of Lantana camara - a weed of national significance in the Australian landscape - and neighbouring flora, upon which our proposed classifier achieves high accuracy (86.07%) in reasonable time and is thus viable for real-time detection and control of Lantana camara.

In Situ Leaf Classification Using Histograms of Oriented Gradients

Harrisia cactus, Harrisia martinii, is a serious weed affecting hundreds of thousands of hectares of native pasture in the Australian rangelands. Despite the landmark success of past biological control agents for the invasive weed and significant investment in its eradication by the Queensland Government (roughly $156M since 1960), it still takes hold in the cooler rangeland environments of northern New South Wales and southern Queensland. In the past decade, landholders with large infestations in these locations have spent approximately $20,000 to $30,000 per annum on herbicide control measures to reduce the impact of the weed on their grazing operations. Current chemical control requires manual hand spot spraying with high quantities of herbicide for foliar application. These methods are labour intensive and costly, and in some cases inhibit landholders from performing control at all. Robotic spot spraying offers a potential solution to these issues, but existing solutions are not suitable for the rangeland environment. This work presents the methods and results of an in situ field trial of a novel robotic spot spraying solution, AutoWeed, for treating harrisia cactus that (1) more than halves the operation time, (2) can reduce herbicide usage by up to 54% and (3) can reduce the cost of herbicide by up to $18.15 per ha compared to the existing hand spraying approach. The AutoWeed spot spraying system used the MobileNetV2 deep learning architecture to perform real time spot spraying of harrisia cactus with 97.2% average recall accuracy and weed knockdown efficacy of up to 96%. Experimental trials showed that the AutoWeed spot sprayer achieved the same level of knockdown of harrisia cactus as traditional hand spraying in low, medium and high density infestations. This work represents a significant step forward for spot spraying of weeds in the Australian rangelands that will reduce labour and herbicide costs for landholders as the technology sees more uptake in the future.

Robotic Spot Spraying of Harrisia Cactus (Harrisia martinii) in Grazing Pastures of the Australian Rangelands.

/pdf/deepweeds-a-multiclass-weed-species-image-dataset-for-deep-405mdjyfva.pdf

Alex Olsen

Papers

DeepWeeds: A Multiclass Weed Species Image Dataset for Deep Learning.

Low-Power and High-Speed Deep FPGA Inference Engines for Weed Classification at the Edge

In Situ Leaf Classification Using Histograms of Oriented Gradients

Robotic Spot Spraying of Harrisia Cactus (Harrisia martinii) in Grazing Pastures of the Australian Rangelands.

DeepWeeds: A Multiclass Weed Species Image Dataset for Deep Learning