J. Eur. Opt. Society-Rapid Publ. 21, 31( 2025) 11
Fig. 7. Detection results on the San Diego-II dataset.
Fig
. 8. Detection results on the Cement Street dataset.
3.4 The impact on spatial information
This section will discuss the effectiveness of the proposed method in preserving spatial information and compare it with some unsupervised dimensionality reduction methods. Including PCA [ 8 ], MNF [ 14 ], Superpixelwise Unsupervised Linear Discriminant Analysis( SuperULDA) [ 40 ], Laplacian Regularized Collaborative Representation Projection( LRCRP) [ 41 ], Superpixelwise Principal Component Analysis( SuperPCA) [ 42 ], and Dual Graph Autoencoder( DGAE) [ 43 ]. All experiments were conducted on the eight data points mentioned earlier, and the experimental results were averaged. The results of the proposed method are based on the average values of the four noise matrices mentioned above as inputs.
Table 3 demonstrates the effectiveness of various dimensionality reduction methods in preserving spatial information. Overall, the proposed method performs well in all indicators and has a leading advantage. Among them, the proposed method achieved the best performance in SSIM, GLCM Contrast, and GLCM Correlation. In terms of PSNR, the proposed method is second only to SuperPCA, indicating that the proposed method has good spatial preservation effects on the overall structure and some prominent textures. An obvious phenomenon is that MNF performs the worst in SSIM, GLCM Contrast, and