Crop Pest Identification Network Based on ImprovedGhostNet
Insect infestation is one of the important factors affecting crop yield.Timely and accu-rate identification of pests is an important prerequisite for effective prevention and control,and is of great significance for improving crop yield.It addresses issues such as low accuracy,com-plex models,and low recognition efficiency.This article proposes an improved crop pest identi-fication model based on GhostNet,replacing the capsule network model with the GhostNet av-erage pooling layer to retain more spatial information and feature extraction from the model;In the Ghost bottleneck structure of GhostNet,the RcLU activation function is replaced by the Hardswish activation function to improve the model's generalization ability and reduce the number of network model parameters.The experimental results show that when identifying pests under 14 types of crops,the improved GhostNet model has the highest accuracy of 97.60％,which is superior to classic convolutional neural network models such as ResNet,provi-ding a certain guarantee for the application scenarios of actual crop pest identification.
Pest identificationGhostNetCapsule networkConvolutional neural network
NETL
NSTL
万方数据
