| Rodent damage in forest area is one of the common forest disasters.Early inspection and control can prevent the occurrence of rodent damage.The pest control work can be carried out according to the statistics of rodent infestation in various regions.The control of rodent damage in forest area is of great significance to the stability of ecological environment.However,the investigation of rodent infestation in forest areas requires a lot of energy of forest managers and the results are not necessarily good.Therefore,the study on the intelligent monitoring and identification system of forest rodent damage based on web system can help forest managers check and prevent forest rodent damage,and lay a solid foundation for forest development.In the field of pest monitoring and identification,most of the following methods have the following methods and have the following shortcomings:(1)Manual observation: This is one of the most traditional methods,but this method not only requires professional knowledge and experience,but also wastes a lot of human resources.(2)Genetic testing: Modern technology has made genetic testing an increasingly popular method for monitoring pests.This method usually provides high-precision detection results,but it consumes a lot of human and financial resources.(3)Video surveillance method: by installing video surveillance equipment,observe the presence of pest rats,so as to realize the identification of pest rats.This method relies on manual monitoring,and the accuracy rate is affected by factors such as the scope of monitoring and the number of personnel,which consumes too much manpower.This study uses deep learning technology to focus on the identification,classification and statistical analysis of pest pictures.In order to solve the problem of a small number of pictures of pest rats,transfer learning technology is adopted.In order to solve the problem of high image similarity between categories of pest rat pictures and the problem of wasting a lot of manpower and financial resources,an image classification network based on attention mechanism and local branching,namely ResNet50 Plus network,is proposed.The network adopts the CA attention mechanism and local branching based on the CSLBP algorithm to extract the features of pest mouse pictures.In the experimental part,ViT,VGG19,MobileNetV3,ResNet18,ResNet34,ResNet50 model and YOLOv7 model were first compared,and the ResNet50 model was selected as the basic model for pest classification.Then,Resnet50-SE,Resnet50-CBAM,and Resnet50-CA were compared to conclude that the attention mechanism selected CA,and finally the ResNet50-CA and ResNet50 Plus models were compared,and finally the conclusion that ResNet50 Plus network classification accuracy was the highest,with a value of 93.26%.The goal of expecting to identify images of pest rats with high similarity was achieved.The innovation of this paper is that on the basis of the global branch Resnet50-CA,a local branch based on the CSLBP algorithm is added to extract local texture features to solve the problem of high image similarity between categories.Comparing the ResNet50-CA and ResNet50 Plus models,the final experimental result shows that the ResNet50 Plus network has the highest classification accuracy,with a value of 93.26%.A forest pest intelligent monitoring and identification system based on B/S architecture has been designed and developed.The system can realize intelligent identification and classification of forest pest images,and provide data statistics and analysis functions.At the same time,it also has intelligent monitoring function,which reduces the work burden of forest managers and provides more accurate monitoring and classification statistics results for forest managers.It provides strong technical support for forest protection and management. |
PDF Full Text Request