Development of ginger plant health monitoring system

Bu, Jia Wei (2025) Development of ginger plant health monitoring system. Final Year Project, UTAR.

Preview

PDF Download (9Mb) | Preview

Abstract

The general intelligent agriculture system is integrated with several other technologies such as sensors, automated irrigation, fertilization, and surveillance systems for increasing efficiency and improving productivity. This project discusses anomaly detection (AD) in images taken from plantations, an important part of such systems. In the related task of plant disease detection, much previous work has relied on various supervised learning approaches; the use of convolutional neural networks and other deep learning models trained on large, annotated datasets of diseased plants has become common. However, this would remain a less feasible approach, considering some practical challenges to acquiring such datasets, particularly in real-world farming scenarios. For example, it is highly impractical and tedious to expect the farmers themselves to take clear, labelled images of every plant. Besides, data collection usually includes flying drones or moving cameras, adding more problems in capturing regular and quality images. This project applies the unsupervised anomaly detection concept, which avoids the use of large-scale pre-labeled datasets. The proposed system trains the model only on healthy ginger plant images to learn normal patterns and detect deviations, if any, as potential anomalies. Such deviations can be because of a disease in the plant or other health problems. This will not only reduce the overhead of manual data labelling but also enhance the practicality of deploying the system in dynamic agricultural environments. The flagged anomalous images can be used later to augment the supervised learning models, thus enabling hybrid supervision for further refinement of the system. The strength of the AD model for real-world images of varied natural environmental conditions considers changes in background, light, and climate. Automating plant health monitoring, reduces manual inspections to a minimum, hence allowing farmers to identify health problems much earlier, take remedial measures, and focus on strategic features of crop management. Overall efficiency increases, labour costs are reduced, and much healthier plantations ensue, hence promoting sustainable agriculture. It signals a very promising step toward the exploitation of artificial intelligence to address challenges in modern farming.

Actions (login required)