Agricultural Pest Detection System Based on Machine Learning

Abstract

This paper designs an agricultural pest detection system based on machine learning. The system consists of pest detection algorithm and PC terminal system. The algorithm uses resnet50 as the backbone network, uses Feature Pyramid Network (FPN) to extract features, and optimizes them by Stochastic Gradient Descent (SGD) and Non-Maximum Suppression (NMS). Finally, the method is implemented by HALCON machine vision software. The PC side uses C # as the development language and C / S three-tier architecture for development, which is realized by visual studio 2015 combined with MySQL database. The system can detect and count the uploaded pest images, and save the detection results to MySQL database. The system constructed 27 agricultural common pest detection data sets, with an average precision of 92.5%. The experimental results show that the system can be effectively applied to the actual detection.