Improved Residual Network Recognition Of Pinewood Nematode Disease Based On GF-1 Image

Pinewood nematode disease is an infectious disease.The infection range mainly occurs between pine trees.Its pathogenic characteristics make pinewood nematode disease have the characteristics of strong pathogenicity and wide influence range.Given this disease,it is very important to determine the treatment plan as soon as possible,quickly count the disaster situation and settle insurance claims.Artificial recognition of trees is slow.the inconsistent color of diseased trees will reduce the artificial recognition rate.If the detection area of trees is wide,it will increase the cost of manual detection and miss the opportunity for treatment of pinewood nematode disease.Therefore,rapid and efficient identification of infected trees is very important for the control of pinewood nematode disease.Given the above problems,based on summarizing and analyzing the existing research results of tree identification with pinewood nematode disease and the related research status at home and abroad,this thesis proposes a method combining deep learning and remote sensing images to identify and predict the infected trees quickly and efficiently.The main research contents and conclusions are as follows:(1)Establish the sample data set of pinewood nematode disease.Leizu Town,Yuan’an County,Yichang City,Hubei Province is taken as the study area.the sample data of pinewood nematode disease suitable for this study were collected.Combined with the GF-1 image,the image data set for pinewood nematode disease was established.The data set contains two kinds of image data of infected and non-infected trees,which solves the problem that there is no public sample set of pinewood nematode disease at present.To meet the requirements of the network model for the amount of data,the image data is preprocessed to expand the number of data sets to meet the training standards of the network model and provide data support for experimental research.(2)The residual network model is used to identify pinewood nematode infected trees.In order to select the appropriate identification model of pinewood nematode disease,five residual network models were designed.Experiments were carried out and the experimental results were analyzed,which proved that the data set can be used as the data basis of this study.Res Net18 was selected as the basic model to identify pinewood nematode-infected trees.(3)The residual network model is improved.Based on the Res Net18 model,an effective residual network model is designed,and the optimization method is improved.The model is evaluated according to the accuracy and other evaluation indicators.The model has strong robustness,and the final recognition accuracy reaches 87%.Compared with the unimproved residual model,the improved model is more accurate in the diagnosis of residual problems.