| Pine wood nematode disease,also known as pine wilt disease,is a very devastating major disease of the pine tree.Once it becomes diseased,it will spread rapidly and quickly kill the diseased plant.The disease is a pine tree devastating epidemic that is currently extremely harmful to pine forests in China.Slightly infected pines are yellowbrown,severely diseased pines are reddish-brown and gradually wither until they die.Pine wood nematode disease is highly pathogenic and has a lethal effect on the host.Infected pine trees quickly die and can quickly spread to other pine trees.In addition,the disease is often unpredictable,and once it appears,it will spread quickly and cause a large number of pine trees to die.Therefore,if the pine trees with pine nematode disease cannot be found and treated in time,it will bring huge economic losses.Pine wood nematode disease is easy to cause huge economic losses because of its destructiveness and rapid spread.At present,the control measures for pine wood nematode disease in China are mainly through quarantine,chemical control,clearing dead wood,and forest management which could prevent the spread of pine wood nematode disease,however,no effective progress has been made in accurate detection and positioning.In this paper,Faster-RCNN deep learning network based on Tensorflow deep learning framework and RPN region generation network is used to analyze the orthophoto image data of pine wood nematode dead tree.According to the detection needs and detection results,this paper first replaces the VGG16 convolutional neural network in the Faster-RCNN network with the Restnet101 residual neural network,secondly optimizes the loss function of the convolutional neural network,improves the RPN region generation network,and uses Open CV computer vision tools combined with MATLAB programming software to enhance the original data.Based on the establishment of the pine wood nematode dead tree model,this study further combined with Python for secondary development,through the Pythonbased Gdal library and Osr library to mark the location of the dead tree on the corresponding position of the image,and generate the geographic information text of the dead tree.To achieve rapid detection and accurate positioning of dead trees of pine wood nematode disease.In the process of analyzing the experimental results,compared with the original network and the original data without data enhancement,the detection effect and accuracy of the detection model trained by optimizing the improved network and the enhanced data have been significantly improved.The accuracy of the model reaches 89.1%,the detected target confidence is above 90%,and the detection speed is 1.069 s.In terms of pine wood nematode dead tree location,the method of this study accurately outputs the latitude and longitude information of the center point of each disease tree to meet the requirements of rapid detection and location of pine wood nematode dead tree. |
