Research On The Change Detection Method Of High-resolution Remote Sensing Image Based On VCA-Net

As a widely used technical tool in the field of remote sensing image processing and analysis,remote sensing image change detection can effectively help detect,analyze and understand the change information of features on the earth’s surface,and p lays an important role in land cover and utilization change analysis,urban sprawl research,ecosystem monitoring and natural disaster assessment.The increasing resolution of remote sensing images enables them to provide richer feature information,but it also brings greater challenges to accurately extract change areas from the complex feature information of high-resolution remote sensing images.Firstly,due to the large differences in seasonal features between remote sensing images in different seasons,it causes the current deep learning-based change detection models to perform poorly in the change detection task of cross-seasonal remote sensing images.Secondly,most of the current deep learning models are easily disturbed by the background information of image features when performing the change detection task,and have weak ability to focus on and characterize the feature information of feature targets.To address these problems,this thesis analyzes and summarizes the image domain conversion methods based on generative adversarial neural networks and the technical means of deep learning for change detection of high-resolution remote sensing images,and conducts an in-depth study based on them.The main research contents of this thesis are as follows:(1)To study the cross-seasonal remote sensing image conversion method.To address the impact of seasonal differences of features in cross-seasonal remote sensing images on the change detection task,a two-way adversarial generation network is introduced to eliminate seasonal differences between cross-seasonal remote sensing images.A cross-seasonal remote sensing image dataset is produced for the study of the image seasonal domain conversion method,and the results are evaluated quantitatively using a neural network-based visual perception model to verify the similarity between the seasonal domain conversion results and the reference image,so as to validate the effectiveness of the model method on cross-seasonal remote sensing image conversion.(2)To construct a VCA-Net-based change detection model for high-resolution remote sensing images.Aiming at the problem that the deep learning model neglects the counteracting elimination of image background interference information as well as the attention and characterization of image feature information of image features in the change detection process,a high-resolution remote sensing image change detection model based on VCA-Net is proposed.The model introduces deformable convolution and attention mechanism modules,adaptively acquires feature information of before and after changes in remote sensing images,filters and suppresses background noise interference in the detection process,and enhances the learning and characterization ability of the network model for key feature information of features,so as to improve the accuracy of change detection.(3)Conducting cross-seasonal remote sensing image change detection experiments.A two-way generative adversarial network is used to transform the seasonal domain of the crossseasonal remote sensing images,and the images after seasonal transformation are put into the VCA-Net model with the corresponding seasonal features for the cross-seasonal remote sensing image change detection experiments.Other model methods are selected as comparison algorithms for experimental comparison to verify the effectiveness of the method in remote sensing image change detection tasks.The experimental results show that the method in this thesis effectively weakens the spectral,spatial and semantic differences between the before and after time-phase remote sensing images due to seasonal differences;The problems of interference by image background noise information and low feature extraction solidification and characterization ability in remote sensing image change detection are solved,and the accuracy of cross-seasonal remote sensing image change detection is improved.