Structure Feature-based Crack Defect Detection For Multicrystalline Solar Cells In Near Infrared Images

The crack defect of solar cells is common,and it is a defect that need to be critically inspected in solar cells production.The crack defect will cause the finger interruption and obstruct current transmission,which makes partial even total failure of solar cells and directly affects the stability and reliability of Photovoltaic(PV)power system.The computer visionbased automatic crack inspection technology on solar cells has played a key role in guaranteeing the quality of solar cells and improving the power generation efficiency of the PV modules.However,the existing research on crack defect algorithm for solar cells is limited,and most inspection technologies still have many problems in sample image scale,crack defect detection accuracy and robustness,and so on.In the near infrared images of multicrystalline solar cells,there are randomly distributed crystal grains forming the inhomogeneously textured background,and the intensities of crystal grains and crack defect are similar.It is difficult to extract accurate and complete crack defect only by relying on the intensity distribution information.Therefore,by applying image structure feature,two crack defect detection methods are proposed for multicrystalline solar cells.The specific research contents and contributions are as follows:(1)A steerable evidence filtering(SEF)-based crack defect detection algorithm for multicrystalline solar cells is proposed.Based on the basic steerable filter,the algorithm takes the local information around the point to be detected into account,and adds two oriented filters with a certain offset in the orientation and distance forming the steerable evidence filtering.It not only significantly improves the contrast between crack and surrounding background but also provides evidence for the presence of crack defect.Then,a segmentation-based method including local threshold and minimum spanning tree is applied to extract complete crack defect.Finally,the crack defect can be accurately located in the inspection image by computing the crack skeleton.The proposed scheme can extract complete crack defect including sharp bends,bifurcations and crack defect in low contrast EL images.The proposed method is expected to help defect detection on other textured surface images with crack defect and linear-type defect,such as pavement crack,steel surface crack and scratch,and so on.(2)A structure similarity measure(SSM)-based crack defect detection algorithm is proposed.Firstly,the structure features of crack defect and randomly distributed crystal grains are analyzed,and the identification functions of linear-structure and blob-structure are built according to the Hessian eigenvalues.Then,a structure similarity measure(SSM)function based on Hessian Matrix is designed to highlight crack defect and suppress random crystal grains simultaneously.Then,a tensor voting-based non-maximum suppression(TV-NMS)method is developed to extract complete crack defect.The proposed method can significantly weaken the interference of the inhomogeneously textured background and obtain uniform background.The proposed method is also expected to help defect detection on other textured surface images with crack and linear-type defect.