Research On Remote Sensing Image Change Detection Algorithm Based On Multi-level Channel Fusion

Remote sensing image change detection technology refers to the process of using multiple remote sensing image information obtained from the same place in different time periods to carry out special processing and finally quantitatively analyze the characteristics of land surface changes.Commonly used change detection algorithms have some common problems,such as change vector detection,spectral feature analysis,time series analysis,etc.These algorithms have strict requirements in the pre-processing stage,need to establish a large number of feature projects,and have problems such as long detection time and low accuracy.To solve these issues,this paper applies the deep learning method to the remote sensing image change detection task,and uses the special representation learning ability of convolutional neural network to solve the complicated remote sensing image change detection problem.On the one hand,based on the semantic segmentation idea in deep learning,this paper proposes to construct a change detection model of remote sensing images by using the characteristics of changed and unchanged regions.The model consists of a main module and two auxiliary modules(assimilation and difference module).The main module uses multi-level attention mechanism to extract multi-dimensional information from multi-temporal remote sensing images and refine the edge information of changing areas;Two auxiliary modules respectively carry out weighted detection on the changed and unchanged areas,and provide enough detailed shallow position information.On the other hand,considering the actual engineering requirements,the accuracy and speed of the model are relatively balanced.In this paper,a micro-neural network is proposed to prune the main part of the model to reduce the network parameters and computation,and the accuracy of the algorithm is improved by multi-level channel fusion technology.Finally,the two models proposed in this paper have achieved satisfactory results in large-scale dual-phase remote sensing image change detection data sets.