Research On Anomaly Detection And Classification Algorithm Of Hyperspectral Images In Remote Sensing Scenes

With the rapid development of big data technology,computer technology,and remote sensing spectral imaging technology,hyperspectral image(HSI)processing in remote sensing scenes has become a research focus in various related fields at home and abroad,among which HSI anomaly detection and classification research has very important research significance and application value.Remote sensing spectral imaging technology can produce abundant spectral information of HSI,and because of the low spatial resolution of HSI high spectral dimension,problems such as imbalance,lack of data,as well as sample,and susceptible to noise in processing and the influence of such factors as mixed pixels,so HSI anomaly detection and classification research still face many challenges.In view of the above problems,this paper mainly completes the following work:(1)This paper first describes the background and significance of the research on HSI anomaly detection and classification in remote sensing scenarios,roughly analyzes the research status at home and abroad,summarizes some main technical difficulties existing in the existing research work,and introduces the relevant theoretical basis.It provides sufficient theoretical basis,solid theoretical basis,and clear research direction for this paper.(2)For HSI anomaly detection task,in order to prevent the mix pixels,and the interference of noise for anomaly detection in complex background,and fully extract and using spectral features and spatial features of HSI,and considering the lack of the HSI data sample,this paper proposes an extraction based on components and low-rank HSI anomaly detection algorithm of sparse matrix decomposition,Firstly,the optimal fractional Fourier transform is used to transform the original spectral-domain information into the intermediate domain information with the Fourier domain information and the original spectral-domain information,which can enhance the noise suppression ability of the model while extracting the salient features.Then,the intermediate domain data is extracted and a stable anomaly detection subset is constructed by the GRX algorithm.Finally,the low-rank sparse matrix decomposition is performed on the abundance matrix and Mahalanobis distance is calculated to complete anomaly detection.Experimental results show that the proposed algorithm can effectively separate the anomaly from the background.(3)For HSI classification tasks,this paper proposes a fast dual-branch dense connection network based on a 3D-CNN framework combined with an attention mechanism for HSI classification.In order to prevent dimension disaster,PCA reduction and 3D convolution subsampling are firstly used to double dimensionality reduction of the original HSI.Then,the network structure was decoupling into a spatial branch and spectral branch,and the spectral and spatial characteristics of HSI were fully extracted,and the dense connection structure was adopted in the spatial branch and spectral branch to prevent network degradation.And sample to ease the imbalance,used in the process of model training combined with label smoothing,cross-entropy loss and focal loss of double loss function,at the same time introduced combined with Fast Fourier transform(FFT)mechanism,the attention of the ECA is to ensure that the classification accuracy at the same time,effectively accelerate the network training speed,finally will double branch output characteristics of fusion,Classification is done through a full connection layer.Experimental analysis of the proposed algorithm on several common HSI classification data sets shows that the algorithm has superior HSI classification performance.