Research On Defect Detection Technology Of PV Panels In Charging Station Based On Deep Learning

The accumulation of dust and damage of photovoltaic panels will reduce the photoelectric conversion efficiency and reduce the power generation per unit time.When the ash accumulation is serious,it will corrode the surface protection layer of photovoltaic panels,causing short circuit in the internal circuit of photovoltaic panels and thus causing safety problems.Therefore,it is very important to detect the defects or ash accumulation of photovoltaic panels.Based on this,this paper focuses on the application of image recognition technology based on Deep Learning in the field of photovoltaic panel defect detection,aiming at the two pain points of long time consuming and low precision in manual inspection of photovoltaic panel defect detection,based on the realistic demand of photovoltaic power station defect detection.Specifically,the main contributions of this paper are as follows:First,Aiming at the jitter phenomenon that occurs when drones take pictures and the problem of unclear shooting when encountering haze weather,an image distortion correction and haze removal algorithm is proposed.The simulation results show that the clarity of the processed pictures is significantly improved.Aiming at the problem that the defects of photovoltaic panels are highly similar and difficult to distinguish from the integrity of photovoltaic panels,a method of fusing shape and texture features in the feature extraction stage is proposed.The simulation results show that the local features of the defects processed by the fusion feature extraction method are better than those after a single The local features of defects after processing by the pattern feature extraction method are more obvious.Second,yolov4 algorithm for small target object detection accuracy is not high question,SKnet attention mechanism is put forward to the method of module inserted into yolov4 model,using the experiment verified the insert SKnet attention mechanism module respectively yolov4 network backbone layer,neck,head layer influence on test results,the simulation results show that insert SKnet attention mechanism module backbone layer has promotion effect to the detection precision of the network,when m AP@0.5 increased 2.3%,but other parts in the embedded promotion effect is not obvious,but has a tendency to decline.The experiment proves that the confidence of the improved model has been significantly improved,and the detection of small targets is more accurate and effective.The research results of this paper : It has positive theoretical and practical significance to improve the power generation efficiency of photovoltaic system,reduce equipment failures,reduce operation and maintenance costs,and ensure the safe and efficient operation of photovoltaic power generation system.