Study On Change Detection Method For High-resolution Remote Sensing Image

The surface of the earth is an open dynamic system with multiple levels,space and time scales.It is of great significance for the protection of ecological environment,the management of natural resources and the scientific decision-making of the government to timely and accurately grasp the dynamic change information of the surface.In recent years,a series of high-resolution remote sensing satellites have been launched one after another,and the acquisition of high spatial resolution remote sensing image is more convenient.Because highresolution image has higher spatial resolution while less spectral information,the traditional change detection method based on spectral has great defects in the application of highresolution image.Based on the high-resolution optical remote sensing satellite images,this study carried out the research on the change detection algorithm to provide support for the dynamic monitoring of land cover.In this paper,based on high-resolution optical image,the method of generating difference image is designed according to the principle of image similarity;the research of automatic sample selection based on D-S evidence theory and SLIC super-pixel segmentation is carried out,and the change detection results are obtained by using high confidence training;the lightweight deep learning network is designed to extract the change information directly,and the end-to-end change detection is realized.The main work of this paper is as follows:(1)The construction of difference image based on structural similarity is studied.Considering the serious noise interference of high-resolution image,Gaussian filter and Mean filter are introduced to filter and smooth the local window first,and the difference image is obtained by calculating the structural similarity of the temporal image before and after the partition,and compared with the traditional difference image generation method,the results show that the difference image generation method based on structural similarity is more effective than the traditional difference image generation method.It has obvious advantages in chemical detection.(2)A change detection method based on D-S evidence theory and SLIC super-pixel segmentation is developed.Taking full account of the spatial texture and spectral features of high-resolution remote sensing image,and using D-S evidence theory difference feature fusion,combined with SLIC super-pixel segmentation algorithm,the automatic selection of changed and unchanged samples with high confidence is realized,and the change information is extracted based on machine learning.The results show that the spatial texture features play an important role in the change detection of high-resolution remote sensing images,and the spectral features can be a powerful supplement to the texture features.The proposed method can effectively reduce the workload of manual sample selection in the change detection of largescale remote sensing images.(3)A change detection network model based on deep level features is constructed.In this paper,we use the deep residual classification network model trained based on the large-scale gaofen-2 satellite image data set to extract the deep-seated features of the experimental image,and design a lightweight deep learning network model with smaller parameters,and preliminarily try the application of deep learning in the field of change detection.The experimental results show that the network can effectively extract the change information under the condition of sufficient training samples,and achieve higher detection accuracy than traditional methods.