The Identification Model Of Pests On The Yuluxiang Pear Leaves Based On The TACNN

As a new medium ripe pear variety,the planting pattern of Yuluxiang pear tends to be collective and large-scale.The insect pest of Yuluxiang pear has a rapid reproduction speed,a wide variety of pests and different degrees of harm,which brings a great challenge to the pest control of Yuluxiang pear.The traditional Yuluxiang pear leaf pest recognition method mainly relies on expert diagnosis experience,which has a strong subjectivity and takes a long time so as to cause the spread of insect pests.Therefore,it is of great urgency to realize the automatic identification of Yuluxiang pear leaf pests in natural environment.With the widespread application of machine vision,deep learning and convolutional neural network technology in the agricultural field,it is possible to automatically identify crop diseases and insect pests.In this paper,the convolution neural network technology is adopted to identify the pest of Yuluxiang pear leaves.The pest image is collected from the Yuluxiang experimental field of Shanxi Agricultural University.After preprocessing the collected data,a pest identification model based on the improved alexnet is constructed.The main contents of the thesis are as follows.(1)The leaves of Yuluxiang insects are collected in a unified way.Considering the difference of image sizes,in order to speed up the work efficiency of the model,the over-sized images are cropped and the under-sized areas are filled.The unified processing image pixels are 224×224,and 1013 pictures are collected.By deleting the image,1005 images of Yulu fragrant pear were obtained.Then 803 processed images are randomly selected for model training,and 101 images are extracted from the remaining images for model verification and testing.(2)The Alexnet model is improved.First of all,the structure and operation of alexnet model are studied and analyzed.According to the fact that there are too many convolution kernel parameters in full connection layer,especially in series layer with the last convolution layer,it is easy to produce over fitting phenomenon.Therefore,the fully-connected layer of the model is improved.Secondly,by changing the parameters of neurons in the full connection layer,two models of Mid-Alexnet and TACNN are obtained.Finally,the recognition results of the three models(Alexnet,Mid-Alexnet,and TACNN)are compared by entering different iteration parameters.(3)Experimental comparative analysis.By compared from three dimensions: model stability,recognition accuracy and model prediction,the experimental results show that TACNN-16 has better classification performance,and the model is the most stable.These dedicate that the combination of the changed parameters of the full connection layer and the appropriate image iteration parameters can improve the recognition rate of the model.The TACNN-16 model constructed in this paper has an average accuracy of 81.18%.The builded model could be used to automatically identify the pests of Yuluxiang and provide help for the decision-making of Yuluxiang pest control.