Deep Learning Based Multi Class Wild Pest Identification and Solving Approach Using Cnn

Abstract

In several nations, specialised pest and disease control has become a high-priority challenge for the agricultural sector. Image processing has become more automated and cost-effective as a result of its cost-effectiveness. In practical crop protection applications, analytic pest recognition systems are commonly used. in the identification and recognition of multi-class pests on a broad scale.This paper proposes a region-wide end-to-end solution called Deeplearning based multi class wild pest monitoring approach using CNN to address this issue. For feature extraction and enhancement, a novel module channel spatial focus (CSA) is proposed to be fused into the convolutional neural network (CNN) backbone. The second is the region proposal network (RPN), which uses derived feature maps from images to include region proposals as possible pest positions. Our 7-year large-scale pest dataset of 88.6K photographs (16 categories of pests) and 582.1K manually labelled pest items is used to test the method. The experimental findings indicate that the proposed System outperforms state-of-the-art approaches in multi-class pest identification, with a mean average precision (mAP) of 87.43 percent.