Detection of Plant Stress Condition with Deep Learning Based Detection Models

Abstract

Deep learning has seen significant growth in its use in agriculture over the past decade due to the environmental challenges faced by this sector. While there have been many deep learning-based approaches proposed in the literature, there are only a few that focus on the detection of plant stress symptoms. This research applied deep learning object detection methods to detect plant stress in eggplant crops during the juvenile vegetative phase. The plants were divided into three classes based on their physical condition: healthy, early stress, and severe stress. Water status, specifically drought stress, was identified as a key factor in plant stress as it can alter normal plant equilibrium and molecular changes, negatively impacting growth and productivity. Three deep learning object detection algorithms, You Only Look Once version-3 (YOLOv3), You Only Look Once version-4 (YOLOv4), and Single Shot Detector (SSD), were explored as potential methods for building a plant stress detection model. The results of the quantitative experiments on eggplant plant images showed that YOLOv3 achieved a mean average precision value of 52%, YOLOv4 achieved 83%, and SSD achieved 56%.