Pest Detection and Classification Approach based on YOLOv11-S and Real-ESRGAN

Abstract

The spread of pests and their infestation on plants pose major threats to crop productivity, especially with theincrease in population all over the world. Early detection of pests is crucial in minimizing the undesirable effects on cropproductivity. Traditional pest detection methods require considerable effort and time. Deep learning can help in detectingpests faster and with fewer human interventions. However, the limited agricultural data and its inferior quality impose hugechallenges affecting the usage of deep learning techniques in agricultural environments. As a result, a robust approach isneeded to help in the early detection of pests. Therefore, this research proposes an approach based on the YOLOv11-S modelto be used in the detection and classification of pests. Also, an enhanced algorithm is proposed to improve the quality of pests’images by using Real-ESRGAN to generate super-resolution images. This image improvement technique led to better resultswhen tested. The method was applied to two datasets, namely, IP102 and R2000. After that, the two datasets were merged,forming eighty-two classes to be used in the training of the model. Experimental results demonstrate that the proposed methodimproves pest detection and classification performance and provides a robust and effective solution for agricultural pestmonitoring compared with existing approaches.