A Coarse-to-Fine Method Based on Saliency Map for Solar Cell Interior Defect Measurement

The internal defects of crystalline silicon solar cells in different production technologies are mainly detected by manual verification or type-specific algorithms based on machine vision currently. To improve the intelligence and universality of inspection algorithms for different types of solar cells, this article proposes a coarse-to-fine method based on a saliency map for solar cell interior defect measurement. First, by introducing singular value decomposition into Fourier transform, the saliency map is generated to make anomalies in the background of heterogeneous textures and nonuniform luminance salient. Then, coarse detection based on global structural information is proposed to extract the potential defects in the saliency map. Finally, fine segmentation using multiscale local statistical analysis is presented to make an accurate classification of the potential defects at the pixel level. The proposed method is applicable to quantitative analysis. To completely evaluate the performance of the presented method, experiments are implemented using five real-world datasets. The experimental results verify that the proposed method performs better than the state-of-the-art methods according to the inspection time and detection results of the solar cells in different production technologies.