No abstract

Iteratively reweighted multivariate alteration detection algorithm has the phenomena of broken patches, much noise, and small change area that are difficult to detect, and the overall detection rate is low. In order to solve this problem, this paper proposes a multi-spectral image change detection algorithm based on band selection and single-band iterative weighting. Because the change information of the multi-spectral image is concentrated in some bands, the background and noise information of the rest bands are more, which may have a negative effect on the final result. Therefore, the band with more change information is selected first, and the iterative weighting of a single band can better suppress the noise and background information, so as to obtain a higher band correlation and facilitate the extraction of change information. This method is used to obtain the characteristic difference graph of the selected band with more change information. After Gaussian denoising of each characteristic difference graph, the Euclidean distance formula is used to fuse the difference graph of each band into a change intensity graph. Finally, the unsupervised k-means clustering algorithm is used to perform binary-valued clustering on the fused difference graph to obtain the change detection results. As a practical application, the superior performance of our proposed method was demonstrated through a large number of comparative tests. INDEX TERMS Multi-spectral change detection, IR-MAD, band selection.

In the present study, an improved iteratively reweighted multivariate alteration detection (IR-MAD) algorithm was proposed to improve the contribution of weakly correlated bands in multi-spectral image change detection. In the proposed algorithm, each image band was given a different weight through single-band iterative weighting, improving the correlation between each pair of bands. This method was used to obtain the characteristic difference in the diagrams of the band that contain more variation information. After removing Gaussian noise from each feature-difference graph, the difference graphs of each band were fused into a change-intensity graph using the Euclidean distance formula. Finally, unsupervised fuzzy C-means (FCM) clustering was used to perform binary clustering on the fused difference graphs to obtain the change detection results. By comparing the original multivariate alteration detection (MAD) algorithm, the IR-MAD algorithm and the proposed IR-MAD algorithm, which used a mask to eliminate strong changes, the experimental results revealed that the multi-spectral change detection results of the proposed algorithm are closer to the actual value and had higher detection accuracy than the other algorithms.

Aim at the disadvantage of too large step and error in steady state of VSLMS algorithm, the algorithm is improved to constitute a new filter. The simulation results show that the new algorithm in time domain and frequency domain is better than the original algorithm, which proves that the denoising performance of the new algorithm is excellent. The convergence speed of the new algorithm is faster than the old algorithm. The steady error is smaller and the algorithm is steady at the convergence stage. Finally, the relation between error and parameters is analyzed by comparison.

Aim at the inadequate of traditional adaptive line enhancement algorithm, a new LMS algorithm with output feedback is put forward in this paper. The new filter output is constituted by the current output and the past output， then constitute a new filter. The simulation result shows that the new algorithm is better than the original algorithm in time domain waveforms, and the performance of the noise reduction is verified.Keywords-adaptive filtering, adaptive line enhancement, denoising, fix step LMS.