Research On Hyperspectral Moving Target Detection Algorithm Based On Temporal-Spatial-Spectral Fusion

The special spectroscopic and imaging technology of the imaging spectrometer can obtain hyperspectral images with high spectral resolution and rich spectral information.How to make better use of the three-dimensional information of the spatial,temporal and spectral in the target detection process is a principal difficulty in the hyperspectral detection.For multi-scene hyperspectral imageries,temporal,spatial and spectral information is used to suppress the background while effectively highlighting the targets,thereby pbtaining better detection results.Therefore,this paper has carried out research on the detection of small targets in single-scene hyperspectral imageries and the problem of multiscale target detection in multi-scene hyperspectral imageries,and the following progress has been made:Aiming at the problem of small target detection in single-scene hyperspectral imageries,a collaborative representation anomaly detection algorithm based on spectral angle feature constraints is proposed.The algorithm is based on the effectiveness of spectral angle mapping as a measure of spectral similarity,and the reciprocal of the spectral angle is used as the coefficient of the ”sum-to-one” constraint in the collaborative representation process,so that the spectral angle feature is used as the ”sum-to-one” constraint to achieve the target collaborative characterization.Especially in the case of dense targets,when the tested pixels are anomalies,the spectral angle feature constraint can reduce the influence of other anomaly pollution in the dual-window by adjusting the collaborative representation weights,and achieve better detection.The experimental results prove that the spectral angle feature constraint proposed in this paper adjusts the representation coefficients of the dual-window neighborhood background pixels in the collaborative representation process,makes full use of the spectral angle feature,highlights the target’s characterization residual,and obtains a better anomaly detection result.Aiming at the problem of multi-scale moving target detection in multi-scene hyperspectral imageries,multi-scale strategies are applied in this paper for the anomaly detection algorithm based on constrained sparse representation and the temporal variance filter.Simultaneously,a fusion-based hyperspectral moving target detection algorithm is proposed based on the results of spatial,temporal and spectral information fusion.The algorithm adopts a multi-scale strategy in the spatial-spectral domain to adapt to targets of different sizes,and adopts a multi-scale strategy in the temporal profile to estimate the appearance time of the target more accurately.Finally,the multi-scale temporal,spatial and spectral fusion strategy is applied to detect the real multiple target hyperspectral sequence dataset,and the detection result based on the multi-scale temporal,spatial and spectral information is obtained.Experimental results demonstrated that compared with single-scale detection,the use of multi-scale fusion strategy can obtain better overall detection results.And through the fusion of multi-scale spatial-spectral anomaly detection and multi-scale temporal profile filtering algorithms,multiple information can be effectively used to highlight common targets and suppress background clutters,thereby obtaining detection results based on multi-scale temporal-spatial-spectral information.