Design Of Micro-crack Detection Algorithm For Photovoltaic Modules Based On Image Processing

With the development of clean energy,photovoltaic power generation has become one of the most important energy.Distributed photovoltaic power stations are being built on a large scale all over the world,so the quality,operation and maintenance of photovoltaic modules are particularly important.As a typical defect type of solar cells,crack detection algorithm has important practical significance not only for production scenes,but also for operation and maintenance scenes.Especially for crack detection of poly crystalline photovoltaic modules,due to the complex flocculent background,it is easy to interfere with crack detection.Based on the image data of photovoltaic modules provided by the cooperative unit,this paper conducts an in-depth study on the problem of crack detection.The main work of this paper is as follows:(1)Processing the data set.Firstly,the electroluminescence image of the whole battery module is processed,and the effective part of the battery in the image is obtained by using the methods of filtering and morphology,and then the effective part of the battery is cut equally according to the number of batteries.(2)A grid line detection algorithm for battery images is proposed as the basis of subsequent algorithms.For the poly crystalline image,the sliding window is cut along the gate line,and the result is used as the basic input unit of the detection network.The idea of transforming the target detection problem into a classification problem is proposed.The range of crack is defined according to the classification of basic input units.The proposed method achieves high accuracy of image classification at the expense of framing accuracy in an acceptable range.(3)For single crystal image,combining with its characteristics,this paper proposes a detection algorithm based on traditional digital image processing methods.After the position of the gate line is detected,the image completion algorithm is used to complete the gate line,so that only defects and relatively pure background remain in the image content,so as to avoid the interference of the gate line to the hidden crack detection.Filter,adaptive binarization,open-close operation and other methods are used to screen out all the candidate regions of defects.For all the candidate connected domains,the morphology and location characteristics of single crystal hidden cracks are studied,and a screening strategy is proposed,which can effectively filter the interference connected domains and the connected domains caused by boundary clipping redundancy,and improve the accuracy of algorithm detection.(4)For poly crystalline images,this paper proposes an improved algorithm based on FAST R-CNN.For the algorithm of ROI,combined with the characteristics of polycrystalline images,combined with the area,gray level and shape characteristics of background floccus and hidden crack features,a new ROI screening strategy was proposed,which incorporated the characteristics of photovoltaic module images.The algorithm network was improved to fix the number of regions of interest(ROI)obtained from each image,enhance the attention to effective information,and reduce the possibility of overfitting.Finally,the classification accuracy of unit image reached 89.89%.Using traditional machine learning method as a comparative experiment,it can be proved that the polycrystalline detection algorithm proposed in this paper is more accurate and effective.