Research on surface image enhancement algorithms of solar cell by combining depth information

With the sustainable growth in energy demand and the escalation of energy issues, new energy resources such as solar energy, wind energy and others are rapidly developing. Solar energy, a kind of clean and environment-friendly energy,has got development for years.As a carrier of solar energy,the quality of solar cell affects the life of its module,as well as the stability and the power generation efficiency,they also affect the market competitiveness among the solar cell manufacturers.Therefore, the detection of solar cell surface has become the focus of many researches, because of its ability to measure the quality of solar cells.Current detection of solar cell surface indicates that the surface target image is weak,blurred and has low contrast. At the same time，there are problems such as the real-time performance and complexity of the algorithm are not compatible, and over-enhanced contrast exists in existing enhancement algorithms. In this issues,we intend to combine depth image information and use multi-scale Convolutional Neural Networks to locate and detect the surface target image of solar cells,there are three main contents included in this process.Firstly, the automated bias supervised learning method for automatically labeling target samples during data pre-training process is researched.Secondly, the sparse representation of image data model based on depth information is studied.Finally, the effect on the precise detection of small targets based on multi-scale convolution mechanism, combining related solutions to achieve overall optimisation of system performance is discussed. The experimental results demonstrate that the method can effectively improve the contrast of the solar cell surface, avoid the excessive contrast enhancement, and provide strong support for rapid and efficient target detection.