Investigation on Object Detection Models for Plant Disease Detection Framework

Abstract

Plant diseases are the most common and severe threat to precision agriculture. Therefore, identification and diagnosis of illnesses at a premature stage are vital. In addition, manual observation according to specific selection criteria is difficult and expensive. While various deep learning-based solutions have been proposed for this process, they usually suffer from lengthy training/testing times with massive datasets. In this paper, to address this problem, we explore the potential of computer vision-based object detection methods for early plant disease detection. A comparative study has been performed with three different benchmark object detection models YOLOv4, EfficientDet, Scaled-YOLOV4. The experimental results were evaluated with precision, recall, F1-score, and mean Average Precision (mAP) as performance metrics. All models are trained using the PlantVillage dataset. Empirical results show that the Scaled-YOLOv4 model is a well suitable object detection model providing a real-time solution in detecting even small infected regions of the plant leaves within less time duration. Therefore, detection and diagnosis of diseases at an early stage of infection are essential.