A Study On Monitoring System Of Pine Nematode Tree Based On Deep Learning And UAV Remote Sensing

China is a big country of forestry.Forests play a vital role in controlling land desertification and reducing soil loss.Today,with advanced science and technology,forest protection is becoming more and more complicated,while reducing the use of human resources,it has become an inevitable trend to develop intelligent forest protection system.As early as 1982,an epidemic of pine wood nematode was discovered in Zhong Shan Ling Forest area in Nanjing,Jiangsu Province.Since then,the epidemic spread across the country,causing a large number of withered and yellow pines and even death,and spreading to national scenic areas and nature reserves.Therefore,relying on machine vision and artificial intelligence technology to monitor the disease situation of pine wood nematode and fundamentally stop the spread of the epidemic situation is the first choice for the protection of pine forest in northern China.Timely detection of pine nematode disease trees is the primary work of epidemic prevention and control.Based on the color image of pine nematode disease trees obtained by aerial photography of UAV,this paper constructed a pine nematode disease tree monitoring system based on deep learning and remote sensing of UAV.The monitoring system realizes low missing detection and high precision detection of pine nematode disease trees by deep learning target detection model and edge computing platform.The specific research contents are as follows:(1)A HOG-SVM based pine wood nematode disease tree detection method was constructedHOG-SVM target detection in traditional machine learning algorithms has been widely used.However,due to the complex content of disease tree image,HOG feature is difficult to be accurately extracted,and the detection model cannot converge,resulting in its inability to effectively complete the disease tree detection task.Therefore,this paper proposes to optimize HOG-SVM disease tree detection method by using selection search algorithm and image mask operation.An improved hogSVM disease tree detection model for pine wood nematode was developed by implementing image over segmentation,hierarchical merging and other operations to construct positive and negative sample sets.The experiment shows that the accuracy of the model on the test set is 74.71% and the omission rate is 25.82%,and the model can complete the task of tree detection.(2)A detection method of pine wood nematode disease tree based on deep learning was established.The target detection algorithm based on deep learning is obviously superior to the target detection algorithm based on traditional machine learning in terms of detection accuracy and speed.By studying the mainstream target detection framework,four deep-learning-based pine nematode disease tree detection methods were constructed.The model was trained and tested on the Pascal VOC data set of pine nematode disease tree detection model,which showed that the accuracy of YOLOv3 pine nematode disease tree detection model for target detection was 93.33%,superior to other models.In view of the model’s defects such as slow loss reduction and empty IoU,this paper proposed the YOLOv3-CIoU pine tree disease detection method in combination with the CIoU algorithm.After training,the model achieved 98.88% accuracy rate on the test set.The detection effect showed that YOLOv3-CIoU could better frame the target edge,and the P-R curve showed that the model could detect the disease tree more accurately and comprehensively.Compared with YOLOv3,YOLOv3-CIoU achieved the expected performance by about 5%.(3)The pine nematode disease tree monitoring system of the airborne edge computing platform was developed.In order to reduce the data redundancy and improve the detection efficiency of ground processing terminal,this paper explored the application of edge calculation in pine wood nematode disease tree monitoring task.In this paper,raspberry PI was selected as the edge computing platform to develop a pine wood nematode disease tree detection method based on lightweight MobileNet V2-SSDLite.After training,the model could be tested to complete a phase of disease tree preliminary detection and image screening tasks.The model was deployed on the raspberry PI and mounted on the DJI M600 UAV.Through the self-developed image transmission module,the communication between the air end and the ground end was established,which realized the cooperation with the YOLOv3-CIoU pine nematode disease tree detection model in the open area,achieving the purpose of two-stage epidemic monitoring.The field test showed that the field coordination effect was good,and the system further improved the monitoring efficiency of pine wood nematode.