Hyperspectral Remote Sensing Image Change Detection

Hyperspectral remote sensing image change detection is an information processing process to determine the surface change based on hyperspectral remote sensing images in different phases of the same area.The collection and analysis of surface change information is of great significance for environmental protection,natural resource management,and research on the relationship between human social development and the natural environment.At present,the hyperspectral remote sensing image change detection technology is applied to earthquake disaster assessment,urban planning,battlefield damage assessment,military reconnaissance and other fields.However,there are still many problems at the current stage that make it difficult to improve the accuracy of change detection,which hinders the development of change detection of hyperspectral remote sensing images.This paper focuses on the problems in hyperspectral remote sensing image change detection.The process of hyperspectral remote sensing image change detection can be divided into two steps: 1.generation of difference map 2.extraction of change information.First,this article summarizes the traditional methods in these two steps,including the difference method,image regression method,correlation coefficient method in generation of difference map,and threshold segmentation method,clustering method and method based on probability and mathematical statistics in extraction of change information.This paper focuses on the use of spectral information,the selection of change thresholds,and the use of spatial information in hyperspectral remote sensing image change detection.The contents are as follows:1.A double-threshold segmentation change detection method based on SDAE is proposed.In the step of generating difference map,aiming at the problem that information redundancy caused by the large number of bands in hyperspectral remote sensing images,unsupervised pretraining stacked denosing autoencoder is used to reduce the dimension of images in different phases,and spectral angle is used to measure the difference information between corresponding pixels.In the step of extracting change information,the double-threshold segmentation method is used to divide the pixels of difference map into three categories: changed pixels,unchanged pixels,and weakly changed pixels by maximizing the variance between classes,then the weakly changed pixels are discriminated based on the neighborhood spatial information of the weakly changed pixels..Experiments on three sets of hyperspectral change detection data sets demonstrate the effectiveness of the proposed method.A change detection method based on multi-path convolutional neural network is proposed.Aiming at the problem that traditional algorithms cannot effectively use spatial information for change detection,a multi-path convolutional neural network is designed to extract change information.This multipath convolutional neural network uses the Inception module,and sets different paths to extract and use spatial information at different scales.Dropout is introduced to prevent the network from overfitting,and auxiliary outputs are set on each path to prevent the gradient vanishing problem.Comparative experiments on the Bay Area dataset,Santa Barbara dataset,and Hermiston dataset demonstrate the effectiveness and applicability of the proposed method in different types of hyperspectral remote sensing images.