Search

Agriculture has a crucial role in both the economy and food supply. However, the frequent occurrence of plant diseases can have a significant negative influence on the production of foods. Timely detection of a disease is important for its effective management, e.g., targeted use of pesticides. Conventional Plant Disease Detection (PDD) methods are manual, slow, tedious, and prone to errors, thereby increasing the risk of significant yield losses. Recently, various data-driven approaches, including Deep Learning and Computer Vision-based approaches, have been explored for PDD. However, the scarcity of large and annotated datasets and the limited scalability of these approaches have prompted researchers to turn to the Self-Supervised Learning (SSL) approach. In this review, we provide the very first conceptual grounding for the SSL approach in PDD. By reviewing a large body of recent related works in the literature, we thoroughly analyse and categorise them into generative, predictive, contrastive and hybrid SSL models. Moreover, this review analyses various recent datasets and performance metrics used in these models. Finally, we explain the research challenges and key research directions aimed at advancing PDD through the SSL approach.

Accurately predicting grape yield in vineyards is essential for strategic decision-making in the wine industry. Current methods are labour-intensive, costly, and lack spatial coverage, reducing accuracy and cost-efficiency. Efforts to automate and enhance yield estimation focus on scaling fruit weight assessments. Machine learning, particularly deep learning, shows promise in improving accuracy through automatic feature extraction and hierarchical representation. However, most of these methods have been developed for analyses at a particular time point and solutions able to consider temporal information captured across sequential frames are currently poorly developed. This paper addresses this gap by introducing a system for yield estimation, utilising publicly available data repositories, such as Embrapa WGISD, alongside an in-house dataset collected by a Blackmagic camera at the pre-harvest stage. We introduce a system that utilises bunch weight regression to estimate grape yield. Bunch weight estimates are obtained by summing samples randomly drawn from the grape bunch weight distribution through empirical calibration. Grapevine bunches are identified and segmented using Mask R-CNN with Swin Transformer, and a SiamFC-based tracking mechanism is employed to estimate the number of unique bunches per panel or row. The number of berries for each tracked bunch is determined using a density approach known as multitask point supervision. In our experiments, we demonstrate the effectiveness of the proposed system for yield estimation, achieving harvested weight errors of less than 5% in two of the three vineyard panels. Larger harvest weight errors (around 15%) were observed due to inaccuracies in tracking certain bunches caused by dense concentration of bunches in one panel. However, these errors should be contrasted with the current practice error of up to 30%, highlighting the potential of machine vision for hands-off yield estimation at scale.

Advances in machine learning and contactless sensors have enabled the understanding complex human behaviors in a healthcare setting. In particular, several deep learning systems have been introduced to enable comprehensive analysis of neuro-developmental conditions such as Autism Spectrum Disorder (ASD). This condition affects children from their early developmental stages onwards, and diagnosis relies entirely on observing the child's behavior and detecting behavioral cues. However, the diagnosis process is time-consuming as it requires long-term behavior observation, and the scarce availability of specialists. We demonstrate the effect of a region-based computer vision system to help clinicians and parents analyze a child's behavior. For this purpose, we adopt and enhance a dataset for analyzing autism-related actions using videos of children captured in uncontrolled environments (e.g. videos collected with consumer-grade cameras, in varied environments). The data is pre-processed by detecting the target child in the video to reduce the impact of background noise. Motivated by the effectiveness of temporal convolutional models, we propose both light-weight and conventional models capable of extracting action features from video frames and classifying autism-related behaviors by analyzing the relationships between frames in a video. By extensively evaluating feature extraction and learning strategies, we demonstrate that the highest performance is attained through the use of an Inflated 3D Convnet and Multi-Stage Temporal Convolutional Network. Our model achieved a Weighted F1-score of 0.83 for the classification of the three autism-related actions. We also propose a light-weight solution by employing the ESNet backbone with the same action recognition model, achieving a competitive 0.71 Weighted F1-score, and enabling potential deployment on embedded systems. Experimental results demonstrate the ability of our proposed models to recognize autism-related actions from videos captured in an uncontrolled environment, and thus can assist clinicians in analyzing ASD.

With the advances of data-driven machine learning research, a wide variety of prediction problems have been tackled. It has become critical to explore how machine learning and specifically deep learning methods can be exploited to analyse healthcare data. A major limitation of existing methods has been the focus on grid-like data; however, the structure of physiological recordings are often irregular and unordered, which makes it difficult to conceptualise them as a matrix. As such, graph neural networks have attracted significant attention by exploiting implicit information that resides in a biological system, with interacting nodes connected by edges whose weights can be determined by either temporal associations or anatomical junctions. In this survey, we thoroughly review the different types of graph architectures and their applications in healthcare. We provide an overview of these methods in a systematic manner, organized by their domain of application including functional connectivity, anatomical structure, and electrical-based analysis. We also outline the limitations of existing techniques and discuss potential directions for future research.