Deep Learning Based Multi-platform Remote Sensing Method For Pine Wood Nematode Disease Monitoring

Pine wood nematode disease has caused great damage to our country’s pine forest resources,timely and effective control of the spread of pine wood nematode disease has become the focus of leading departments at all levels.With the development and application of remote sensing technology and artificial intelligence,intelligent remote sensing monitoring with remote sensing technology as the platform and artificial intelligence technology as the core has become a practical and high-precision automated pine wood nematode disease monitoring method.In view of the wide spread of pine wood nematodes and high monitoring accuracy requirements,multi-platform remote sensing pine wood nematode disease monitoring methods based on satellite wide-area surveys,unmanned aerial vehicle area surveys,and manual ground verification have become one of the important research directions for pine wood nematode disease prevention and control.In order to improve the monitoring accuracy of pine wood nematode disease and strengthen the monitoring efficiency of pine wood nematode disease,this paper proposes a multi-platform remote sensing intelligent monitoring method.This method utilizes the characteristics of wide satellite remote sensing image monitoring range and stable return visit period,constructs a 7-layer fully connected neural network as the monitoring method of the satellite remote sensing platform to realize the general survey of pine wood nematode disease and determine the pine wood nematode disease infection area.At the same time,using the features of UAV’s flexibility and high temporal and spatial resolution,the Spatial-ContextAttention Network(SCANet)was constructed to realize the precise identification of individual pine wood nematode diseases.In addition,on-site verification and clean-up of the UAV detailed inspection results is conducted by manual.This paper takes part of Huangshan City as a sample area for testing to determine the feasibility of this method for pine wood nematode disease monitoring,and takes part of Jingde County as a generalization ability verification area to verify the effectiveness of this multi-platform remote sensing monitoring method.The specific results of this paper are as follows:1.The general survey of pine wood nematode diseases in satellite remote sensing images was realized by using a fully connected neural network.In this paper,a satellite platform pine wood nematode disease sample library is established based on the domestic Gaofen 2 remote sensing image.The blue band data is removed in the input data and the RGNDI index is added.The satellite remote sensing platform pine wood nematode disease is realized by constructing a 7-layer fully connected neural network.At the same time,we compare and analyze results of traditional methods and deep learning methods.The results show that our method is superior to traditional methods and other deep learning methods in evaluation indicators.The overall accuracy of the monitoring results in this paper is 0.5614,the precision is 0.6158,and the recall is 0.8615.Although the results of this paper are fragmented to a certain extent,the monitoring results meet the general survey requirements of large-scale satellite remote sensing pine wood nematode disease monitoring,determine the pine wood nematode disease hazard area,and provide scientific guarantee for detailed inspection by UAV.2.This paper uses the fully convolutional neural network to realize the detailed investigation and positioning of the pine wood nematode disease in the remote sensing image of the UAV.Based on UAV remote sensing images,the semantic segmentation sample database of pine wood nematode diseases was established.Aiming at the characteristics of small disease targets,high image resolution,and complex background,this paper designed SCANet to realize the precise identification of individual plant diseases.SCANet network mainly obtains the spatial information of pine wood nematode diseases through the spatial information retention module,uses the context information module to expand the receptive field to further enhance the spatial and spectral details of the pine wood nematode disease.The up-sampling module is designed to restore the features and integrate the shallow features which obtain the final monitoring results.The results show that method extracts pine wood nematode diseases with a recall rate(Recall)of 0.9322 and a missing alarm rate(Missing Alarm)of 0.0678.The extraction effect is better than other comparison methods,achieves rapid,high-precision and automated unmanned Intelligent monitoring of pine wood nematode diseases and has good applicability.The SCANet provides reliable technical support for multi-platform remote sensing pine wood nematode disease monitoring.3.In this paper,the multi-platform remote sensing integrated monitoring method for pine wood nematode diseases was verified in Jingde area.The satellite monitoring method is used to determine the disease occurrence area in the study area,and the UAV is used to fly the disease occurrence area and realize the single plant disease monitoring.Finally,the satellite and UAV remote sensing image monitoring results are compared and verified by manual methods.The experimental results show that the multi-platform remote sensing image pine wood nematode disease monitoring method is accurate and effective.The multi-platform remote sensing image monitoring method in this paper realizes the high-precision,high-efficiency,automatic and intelligent monitoring of the pine wood nematode disease and meets the needs of practical applications.Its application prospects and research significance provide theoretical foundation and scientific basis for further in-depth research.