Research On Generation Of Adversarial Network Based On Improved Deep Convolution And Its Application In Pest Identification

With the rapid development of Chinese science and technology and the continuous improvement of deep learning technology,its research and application in the field of agriculture have gradually become popular.While the new generation of information technology is integrated with the agricultural sector,it has also laid a solid foundation for the development of agriculture,and will provide an unprecedented solution to China's resource shortage and environmental pollution.However,the classification and identification of crop diseases and insect pests is a key part of control.However,due to the hidden,dynamic,perverted,and large population of pests,traditional image classification and identification methods are facing great challenges.Due to the imperfect control system of crop diseases and insect pests in our country,and the classification and identification methods of pests are not perfect,every year,due to the continuous increase of crop diseases and insect pests,China's crop products continue to decline,the development of the agricultural sector has been limited,and economic benefits have been damaged.The image classification and recognition technology based on the improved deep convolutional condition generation adversarial network can effectively and accurately identify pests.Therefore,this article is based on the "National Key R & D Technology Project",which selects three original pest image data sets,which are collected from 24 different types of field insect data and forest wheat moth 4 and 18 crops collected on the search engine.The data of roll moth pests and the 20 types of northern pests we actually photographed were taken as the research objects.By improving the deep convolution conditions to generate an adversarial network model algorithm,it provides accurate classification basis for pest identification.The main research work of the text is as follows:(1)An overview of related algorithms for pest recognition and analysis are summarized and researched.The shortcomings of traditional pest recognition algorithms are summarized,and the research ideas of pest recognition algorithms based on deep learning and neural networks are clarified.(2)The generation and optimization of the adversarial network were improved and a pest classification algorithm based on the deep convolutional condition generation adversarial network was proposed.First,the spatial pyramid layer(SPP)is used to embed the discriminator's convolution layer,so that the classified images have a fixed size and the same size,which not only allows the model to obtain the required image features,but also greatly improves the model's convergence speed.Secondly,the crossentropy loss function in the deep convolutional conditional adversarial network is converted into a least squares loss function,so that the least squares loss function reaches a certain point,thereby enhancing the stability of model training.Finally,the RMSprop optimizer is used to iteratively optimize the model to make the model converge faster,enhance the stability of the model,and reduce the overall network fluctuation.The deep convolutional conditional adversarial network has obvious advantages in classification accuracy and experimental stability.The average classification accuracy obtained by cross-validation is as high as 96.8%,which is significantly better than other comparison algorithms.,Can provide a reference for related research work.(3)A sparse coding-based deep convolutional condition generation adversarial network model was established.The advantages of sparse coding technology and deep convolutional condition generation adversarial network were established.The convolutional network in the adversarial network discriminator was generated by replacing the deep convolution condition with a sparse coding network.Experimental results Through experimental comparison and cross-validation methods,the average classification accuracy of the network model is as high as 97%,and the convergence rate of the model is greatly increased,which provides a strong and powerful foundation for future research on pest identification and classification algorithms.