Ground Object Change Detection Methods Based On Hyperspectral Image Analysis

The detection of changes in ground objects is closely related to a series of issues such as urban development,global change,and land use change.Hyperspectral change detection is a fast and appropriate technical means by using remote sensing images of the same surface area for a long time.Temporal and spatial resolution of remote sensing images limits the development of hyperspectral change detection.The focus is on solving large-scale,large-scale,high-level changes.According to different processing methods,the detection methods are divided into direct comparison method and classification comparison method.According to the target object,it is divided into pixel level,feature level and target level change detection.Pixel-based change detection refers to whether the registered remote sensing image pixels change and which types change.This pixel-based change detection is highly accurate.However,the performance of the pretreatment has a great influence on the test results.Pixels are subject to interference from radiation,atmosphere,angles,etc.,and may generate a lot of error messages that affect the change detection results.The choice of threshold values in pixel-level change detection plays a key role,ie it is difficult to determine the type of feature change.In order to realize the detection of the change of the features of the pixel-level hyperspectral image,this thesis studies the preprocessing of the one-dimensional spectral domain data of the hyperspectral image,that is,the research of the noise reduction measurement method and the realization of the hyperspectral ground object change detection algorithm.In this paper,the research on the detection method of ground object change based on hyperspectral image is carried out.The main research contents are as follows:(1)Based on wavelet transform and EMD combined noise reduction method and feature enhancement EMD detection method of ground object change: The wavelet transform and EMD based noise reduction method overcomes the shortcomings of traditional Fourier transform which is easy to cause high frequency information loss.Research on feature detection methods for spectral similarity matching.The focus is on the extraction of feature difference feature vectors based on similarity matching and distance metrics.Obtain the final feature change detection map;(2)Research on detection method of hyperspectral imagery combined with ground object change based on deep learning: In order to make full use of spatial and spectral domain information of hyperspectral image,this design proposes a new extraction pixel spatial domain and spectral domain;(3)Research on detection method of hyperspectral ground object transformation based on migration learning and generating confrontation network: Direct push migration learning method is adopted.The low-level parameters of the model are trained using tagged samples of the source data.Use the target data to train the highlevel parameters of the model and fine-tune the low-level parameters.Improve the accuracy of hyperspectral transform detection.On the basis of migration learning,the anti-automatic encoder is used to make full use of the time series information of the sample to generate multiple types of ground object change detection samples.The extended training samples are used to train the model to make the detection accuracy higher.