Development of Image Recognition System for Tobacco Aphid Based on Resnet-101 Model
Myzus persicae are one of the main pests that harm the growth of tobacco.The accu-rate identification of the occurrence of tobacco aphids and the precise grading of the severity of their damage are crucial for guiding prevention and control.This study collected images of the number of M.persicae on tobacco plants during tobacco growth,and supplemented the images with data augmentation methods such as sharpening,flipping,and brightness changes to con-struct a dataset of crop images of M.persicae infestation.The images of the number of M.per-sicae were classified into three levels:mild occurrence,moderate occurrence,and severe occur-rence.Resnet-101 model was used for image recognition training.According to the model pa-rameter results,the average accuracy of the training set in the Resnet-101 training cycle was 85.49％,with the highest value being 87.33％.The average accuracy of the test set was 80.13％,and the highest value was 89.92％.The average accuracy of the recognition system in identifying the number of M.persicae was 83.00％.This study achievedquantitative image rec-ognition of the number of M.persicae,providing model support for the development of an au-tomated tobacco pest control system.
万方数据
维普
