Saliency Fusion of Multispectral Images With Non-uniform Texture and Complex Surface Defects

To suppress the influence of the complex background of polysilicon solar cells on the reliable extraction of surface defects and enhance the defect characteristics, a multi-spectral saliency fusion algorithm based on empirical wavelet was proposed. A structure-texture image decomposition method was used to suppress complex background. Through multi-scale decomposition of two-dimensional tensor empirical wavelet with empirical information, the detail layer containing defect information was obtained. The saliency analysis was carried out to enhance the contrast of the larger weight parts, and the strong defect feature image was finally obtained. The four quality evaluation indexes of average gradient, edge intensity, definition and standard deviation were compared. Results show that these indexes of the proposed algorithm are improved by 0.01, 11.11, 0.48 and 2.33 than those of the classic fusion algorithm, showing better performance.