Deep learning – Method overview and review of use for fruit detection and yield estimation

The performance of six existing deep learning architectures were compared for the task of detection of mango fruit in images of tree canopies. Images of trees (n = 1 515) from across five orchards were acquired at night using a 5 Mega-pixel RGB digital camera and 720 W of LED flood lighting in a rig mounted on a farm utility vehicle operating at 6 km/h. The two stage deep learning architectures of Faster R-CNN(VGG) and Faster R-CNN(ZF), and the single stage techniques YOLOv3, YOLOv2, YOLOv2(tiny) and SSD were trained both with original resolution and 512 × 512 pixel versions of 1 300 training tiles, while YOLOv3 was run only with 512 × 512 pixel images, giving a total of eleven models. A new architecture was also developed, based on features of YOLOv3 and YOLOv2(tiny), on the design criteria of accuracy and speed for the current application. This architecture, termed ‘MangoYOLO’, was trained using: (i) the 1 300 tile training set, (ii) the COCO dataset before training on the mango training set, and (iii) a daytime image training set of a previous publication, to create the MangoYOLO models ‘s’, ‘pt’ and ‘bu’, respectively. Average Precision plateaued with use of around 400 training tiles. MangoYOLO(pt) achieved a F1 score of 0.968 and Average Precision of 0.983 on a test set independent of the training set, outperforming other algorithms, with a detection speed of 8 ms per 512 × 512 pixel image tile while using just 833 Mb GPU memory per image (on a NVIDIA GeForce GTX 1070 Ti GPU) used for in-field application. The MangoYOLO model also outperformed other models in processing of full images, requiring just 70 ms per image (2 048 × 2 048 pixels) (i.e., capable of processing ~ 14 fps) with use of 4 417 Mb of GPU memory. The model was robust in use with images of other orchards, cultivars and lighting conditions. MangoYOLO(bu) achieved a F1 score of 0.89 on a day-time mango image dataset. With use of a correction factor estimated from the ratio of human count of fruit in images of the two sides of sample trees per orchard and a hand harvest count of all fruit on those trees, MangoYOLO(pt) achieved orchard fruit load estimates of between 4.6 and 15.2% of packhouse fruit counts for the five orchards considered. The labelled images (1 300 training, 130 validation and 300 test) of this study are available for comparative studies.

Deep learning for real-time fruit detection and orchard fruit load estimation: benchmarking of ‘MangoYOLO’

Path planning optimization of indoor mobile robot based on adaptive ant colony algorithm

Evolutionary computational intelligence algorithm coupled with self-tuning predictive model for water quality index determination

Path planning algorithms are used by mobile robots, unmanned aerial vehicles, and autonomous cars in order to identify safe, efficient, collision-free, and least-cost travel paths from an origin to a destination. Choosing an appropriate path planning algorithm helps to ensure safe and effective point-to-point navigation, and the optimal algorithm depends on the robot geometry as well as the computing constraints, including static/holonomic and dynamic/non-holonomically-constrained systems, and requires a comprehensive understanding of contemporary solutions. The goal of this paper is to help novice practitioners gain an awareness of the classes of path planning algorithms used today and to understand their potential use cases—particularly within automated or unmanned systems. To that end, we provide broad, rather than deep, coverage of key and foundational algorithms, with popular algorithms and variants considered in the context of different robotic systems. The definitions, summaries, and comparisons are relevant to novice robotics engineers and embedded system developers seeking a primer of available algorithms.

A Survey of Path Planning Algorithms for Mobile Robots

The problem of water pollution is increasing every day, due to the industries’ waste product disposal, migration of people from rural to urban areas, crowded population, untreated sewage disposal, wastewater and other harmful chemicals’ discharge from the industries. There is a need to resolve this problem for us to get good water that can be used for domestic purposes. This article proposes a suitable classification model for classifying water quality based on the machine learning algorithms. The paper analyzed and compared performance of various classification models and algorithms in order to identify the significant features that contributed in classifying water quality of Kinta River, Perak Malaysia. Five models with respective algorithms were tested and compared with their performance. In assessing the result, the Lazy model using K Star algorithm was the best classification model among the five models had the most outstanding accuracy of 86.67%. Generally, wastewater is harmful to our lives, and bringing scientific models in solving this problem is obligatory.

Classification Model for Water Quality using Machine Learning Techniques

Dijkstra's algorithm is a classic algorithm for finding the shortest path between two points due to its optimisation capability. The adjacency matrix is the naive storage structure of the algorithm. This storage structure has limited the use of the algorithm as it expands large storage space. A multi-layer dictionary is proposed in this work to enhance the storage structure. Previously, the algorithm was used to optimise single parameter (such as distance, time and fuel) for movement between two places. The path computed using the classic Dijkstra's algorithm is the shortest; however, it may not be the most feasible. The algorithm needs to optimise other factors such as energy consumption and the degree of turns a robot need to take. Junction degree of difficulty function is introduced to further improve the output. It is defined as the degree of difficulty of the path between two points in an open environment. The experimental result shows that the proposed path planning method produced the most optimal path between two points when applied to a map of any indoor terrain.

Robotic Indoor Path Planning Using Dijkstra's Algorithm with Multi-Layer Dictionaries

Martial arts (MAs) are considered as a preserved heritage primarily due to the fact that it promotes certain level of identities of a culture. MA refers to the art of combat and self-defense which normally combines offensive and defensive techniques. Technology advancements have made motion capture (MoCap) to be widely used in MA to capture and evaluate human performance. Nevertheless, researches on extrinsic feedbacks (EFs) of MA through the developed evaluation system are scarce. Furthermore, there is no complete framework of evaluation system suggested for MA. This paper presents the theoretical framework of EF-based automated evaluation system in the context of traditional local MA. The framework contains three modules including MoCap, recognition and evaluation. The MoCap module tracks human body accurately in order to generate skeleton, tune focused target, and record human movements. Recognition module develops a script of motion for templates and classification purposes using Reverse-Gesture Description Language (R-GDL) and GDL respectively. Evaluation module produces the extrinsic feedback in terms of pattern and score for the performed movements. This theoretical framework will be used in the development of the digital tool to measure the accuracy and effectiveness of motions performed by one of the traditional local MAs.

/pdf/a-theoretical-framework-of-extrinsic-feedback-based-9gdzp30lap.pdf

A theoretical framework of extrinsic feedback based-automated evaluation system for martial arts

The number of research papers on Motion Capture technologies published in conferences and journals has been rapidly increasing due to the emerging of new technologies, software and hardware which create new challenges and opportunities for Martial Arts research. Current trend of the Martial Arts using Motion Capture technologies (MAMoCap) researches consists of phases of MoCap-Processing and Post-MoCap-Processing; contexts of algorithms, performance and system development; and feedbacks of intrinsic and extrinsic. The purpose of this paper is to study and explore the potential future trend of research and publications pertaining to MAMoCap researches. A systematic survey of research publications was conducted through the topic of Martial Art (MA) and Motion Capture (MoCap) in order to retrieve the scientific articles published in FOUR (4) established publishers including SPRINGERLINK, SCIENCEDIRECT, IEEE and ACM. Search refinements were done by the inclusions criteria of document types of academic journals and conference proceedings; and by the exceptions criteria of letters, editorials and book reviews. The findings show that only 27% of the publications have been selected while other 73% have been classified as irrelevant contents due to none significance and relevance to the MAMoCap researches. Analysis on the research phases, contexts and feedbacks has been conducted and discussed in detailed for pertaining knowledge gaps and future research agenda. Based on the preliminary study, a framework of EFs-Based Automated Evaluation System for the martial arts should be proposed.

A systematic survey of martial art using motion capture technologies: the importance of extrinsic feedback

This paper presents a method of detecting overlapping mango fruits from the complex background image. This research uses image that is obtained from a digital camera. This method is based on pre-processing the input image using the texture analysis to determine the boundary of each overlapping fruits. The image is processed to determine the actual boundary, converted to binary images, and utilise dilation and erosion to determine the object. We use ellipse fitting by applying Randomized Hough Transform to search the potential area of the mango fruit. Ellipse fitting are chosen because the shape of the mango fruit is similar to ellipse shape. Using these techniques, the fruit is successfully detected including the fruits that are overlapping with each other.

Azrul Amri Jamal

Papers

Classification Model for Water Quality using Machine Learning Techniques

Robotic Indoor Path Planning Using Dijkstra's Algorithm with Multi-Layer Dictionaries

A theoretical framework of extrinsic feedback based-automated evaluation system for martial arts

A systematic survey of martial art using motion capture technologies: the importance of extrinsic feedback

Automatic mango detection using texture analysis and randomised Hough transform