Research On Remote Sensing Change Detection Method Of Photovoltaic Energy Activities

With the aggravation of global warming and energy crisis,people are paying more and more attention to environmentally friendly new energy.As a renewable energy source that uses solar power generation,photovoltaic power generation has the characteristics of safety and cleanliness.The monitoring of the change of photovoltaic land closely affects the energy deployment of photovoltaic facility land,satellite remote sensing is an objective means of large-scale photovoltaic energy survey,and photovoltaic remote sensing information extraction based on deep learning will effectively improve the efficiency of information processing.However,due to the complex background,size and shape of the PV target area,the general algorithm is not ideal for detecting PV changes in various scene categories and different resolutions,and lacks a public dataset of PV targets.Therefore,taking photovoltaic change detection methods as the research direction,this paper analyzes the research status and related technologies at home and abroad,expounds the difficulties faced by the change detection task in remote sensing images,and makes targeted improvements to the current problems,and the key research contents are as follows:1.A photovoltaic remote sensing change detection dataset was constructed.Using the deep learning model combined with manual correction,the deep learning model based on convolutional neural network is used as a semi-automatic annotation model of feature extractor,which performs artificial assisted correction of unsatisfactory data,improves the labeling efficiency,and constructs high-resolution and medium-resolution photovoltaic remote sensing change detection datasets with multiple background categories.2.A photovoltaic change detection model of high-resolution remote sensing images was constructed.Combined with the transfer learning pre-training model and the self-attention network module,the U-net was optimized,and the classification detection and comparison experiments of photovoltaic land in different background categories were carried out,and the connection domain analysis post-processing process was optimized for the problem of small area noise misdetection in the model optimized by the self-attention network module,which improved the detection performance of the model for photovoltaic targets in high-resolution images.3.A photovoltaic change detection method for medium-resolution remote sensing images is proposed.A variety of mixed data augmentation operations were carried out to address the data characteristics of medium resolution photovoltaic remote sensing images.The backbone network Mobilenetv2 in Deeplab V3+neural network was replaced with a deep residual Res Net50 backbone network with deep feature extraction capabilities,which significantly improved the ability to extract photovoltaic features and reduced the probability of false positives,And further integrating the edge refinement Cascade PSP module,the model improves the precision of edge detection for irregular photovoltaic regions.Based on the above research content,a prototype system for photovoltaic remote sensing image change detection is designed and implemented.It includes functions such as data management,model management,target prediction,and result display,which proves that the system can provide support for the research of photovoltaic change detection and has good application value.