Deep Learning Approaches for Precise Pest Identification in Agriculture

Abstract

One of the most important pest management strategies in agriculture is the precise and accurate identification of pests, which significantly reduces the crop losses and improves yield through timely interventions. This research focuses on the development of a novel approach for identification of pod bug species, which is an important pest in cowpea using the YOLOv8 deep learning model. In this study, major focus was given to the classification of four major species of pod bug viz. Clavigralla horrens, Coptosoma cribraria, Nezara viridula, and Riptortus pedestris. For this, ten different sets of models were created by varying the image sizes, preprocessing steps, number of epochs and training/testing ratios to identify the most optimal model configuration. It was found that the most effective model achieved a mean Average Precision (mAP50) of 0.989 on the validation set with 276 epochs. The ability of developed models to accurately detect and classify the pests in a complex agricultural environment were validated through prediction graphs. The use of this model in crop protection practices imparts significant improvement in managing pests by providing real time insight to farmers. Further, the model developed through this study lays the groundwork for expansion, to include other pest species that could potentially enhance pest management strategies in agriculture.