A Comparative Study on Plant Diseases Using Object Detection Models

Plant diseases are the most widespread and significant hazard to 'Precision Agriculture'. With early detection and analysis of diseases, the successful yield of cultivation can be increased; therefore, this process is regarded as a critical event. Unfortunately, manual observation-based detection method is error-prone, hard, and costly. Automation in identifying plant diseases is extremely beneficial because it saves time and manpower. Applying a neural network-based solution can detect disease symptoms at an early stage and facilitate the process of taking preventive or reactive measures. There have been various deep learning-based solutions, which were developed using lengthy training/testing cycles with large datasets. This study aims to investigate the suitability of computer visionbased approaches for this purpose. A comparative study has been performed using recently proposed object detection models such as YOLOv5, YOLOX, Scaled Yolov4, and SSD. Atailored version of the "PlantVillage" and "PlantDoc" datasets was used in the Indian sub-continent context, which included plant disease classes related to Potato, Corn, and Tomato plants. This study provides a detailed comparison between these object detection models and summarizes the suitability of these models for different cases. This paper can be useful for prospective researchers to decide which object detection models could be used for a specific scenario of Plant Disease Detection.