Research On Real-time Identification Technology Of Stored Grain Pests Based On Image Processing

The food issue is related to the national economy and the people's livelihood.The annual food loss caused by pest damage in the country can reach up to 6 million tons.Therefore,the research on the identification of stored grain pests is the main research direction to prevent the loss of storage grain output.At present,the real-time automatic identification of pest technology is still in the research stage,and it has not been widely used in domestic granaries.Aiming at this issue,the paper probes into the superiority of image processing technology and convolutional neural network(CNN)algorithm in pest image recognition and classification technology.It adopts the current popular Keras framework,which uses the mainstream neural network API.Seven-layer and nine-layer convolutional neural network models have been built.And the parameter settings of the nine-layer convolutional neural network model have deeply studied.The pre-processed pest image data set was sent to the seven-layer convolutional neural network model and the improved nine-layer convolutional neural network model.Through the experimental comparison and analysis,the best accuracy rate of pest identification was obtained.A comparison of good accuracy rates validates the effectiveness of the improved nine-layer convolutional neural network model.And made a certain contribution to the realtime identification of stored grain pests.The main completions are as follows:(1)Obtain the pest picture,filter and sort the picture,perform grayscale,image enhancement and image segmentation processing on the finished picture,and organize the pest identification image data set;(2)Construct a convolutional neural network model.Using the Sequential model,Softmax classifier,adjust the different activation functions,Tanh function,Softsign function and Relu function,build a seven-layer convolutional neural network model and a nine-layer convolutional neural network model in the Keras framework.And optimize the parameters of the nine-layer convolutional neural network model;(3)The accuracy of recognition under different convolutional neural network models is compared and analyzed.The pre-processed pest image data set was sent to the seven-layer convolutional neural network model and the improved nine-layer convolutional neural network for identification and classification.Experiments show that the recognition accuracy is 87%,79%,80%,under the seven-layer convolutional neural network,different activation functions of different iteration times.Under the improved nine-layer convolutional neural network,the recognition success rates of different activation functions are 83%,83%,and 98.6%,respectively.The effectiveness of the improved nine-layer convolutional neural network model is obtained by comparison.