Identification Of Insect Pests Of Broad-Leaved Crops Foliar Based On Deep Learning

In the process of agricultural modernization,the prevention and control of pests and diseases is the top priority,which has received extensive attention from the country and all sectors of society.In the research,in the traditional identification of insect pests of foliar,the characteristics of different diseases and insect pests are small,and the requirements for professional knowledge are high.Advanced information technology means deep learning method is selected to complete the identification task of insect pests of broad-leaved crops foliar.The Plant Village project builds relevant data sets,and proposes two improved pest identification models to complete related experiments:(1)In the task of insect pest identification,the conventional VGG16 network has a huge amount of parameters and a large amount of calculation to improve,adjust the network structure,use channel shuffling and the point-by-point group convolution operation of the pyramid structure,and at the same time reduce the weight of the network,fully using the channel information of the input image,and using spatial pyramid pooling to extract multi-scale feature information,the strict limitation on the size of the input image in the insect pest identification task is lifted,and the adaptability of the identification model is increased.(2)In order to further improve the insect pest identification efficiency of the model,aiming at the contradictory relationship between high performance and low speed of the multi-branch structure in the deep convolutional neural network,an improved residual network model is proposed,and a multi-scale convolution module is used to enhance the network.Feature extraction ability,and after the training,the training phase and the inference phase of the network are divided by the structural reparameterization method,and the multi-branch convolution and BN layers are combined in the inference network to achieve high performance of multi-branch training and single branch training.High speed of road reasoning.The experimental results show that this method can effectively reduce the identification time of pest images.The paper has 36 pictures,19 tables,and 54 references.