Multi-objective band selection algorithm based on NSGA-II for pattern segmentation of textile hyperspectral images

China is the birthplace of silk weaving, possessing a rich heritage of silk artifacts. Textile patterns exhibit unique artistic value and carry profound historical and cultural significance. Through automated segmentation of the patterns, we can provide technical support for the preservation, inheritance, and innovative application of cultural heritage. Edges are critical for image segmentation. Hyperspectral images enable precise identification of material and dye variations via continuous narrow-band spectral resolution, overcoming the limitations of RGB images in low-contrast edge detection. However, the redundant bands adversely affect results. To address the issue, we propose the edge-oriented multi-objective optimization of the band selection (EOMOBS) algorithm. Furthermore, to overcome texture noise interference, insufficient spectral information utilization, edge breaks, and poor parameter adaptability in existing methods, we propose an improved Canny operator for the selected bands. When applied to textile pattern segmentation, the method achieves 93.74% PA and 73.19% IoU, significantly outperforming sixteen alternative methods.