Research On Shadow Detection Method Of Photovoltaic Panel Based On Deep Learning

Under the background of serious environmental pollution and energy shortage,photovoltaic power generation,as a new type of clean and renewable energy,has attracted more and more attention.However,the failure of photovoltaic system has a huge impact on the photovoltaic power generation system.The shadow of photovoltaic panel is a common hidden danger,which can reduce the power generation efficiency,and produce hot spot effect and even damage the photovoltaic cell components.Therefore,the research on shadow detection of photovoltaic panels has important practical significance for improving the efficiency of photovoltaic power generation.Aiming at the problems of high target density,large overlap,high cost and poor realtime performance in photovoltaic panel shadow detection,a CRC-Retina Net photovoltaic panel shadow detection algorithm based on Retina Net is proposed in this paper.The algorithm adopts CSP(cross stage partial)structure to redesign the feature extraction network,which improves the detection accuracy and speed;The feature maps extracted in backbone adopt cyclic feature fusion structure to enhance the feature information of all targets;The robustness of the network is enhanced by replacing the activation function;CIo U Loss is used to improve the positioning accuracy of boundary regression.The experimental results show that the average detection accuracy of the algorithm can reach up to 99.24%,which is 4.02% higher than the original Retina Net algorithm.So,the CRCRetina Net provides an algorithm basis for designing lightweight model in subsequent practical applications.Although CRC-Retina Net algorithm can meet the basic requirements of photovoltaic panel shadow detection,there are still some problems such as large model size and slow detection speed,which need to be further optimized.Therefore,Ghost-Retina Net,a lightweight photovoltaic panel shadow fast detection algorithm based on CRC-Retina Net is proposed.The feature extraction network of the algorithm adopts the idea of Ghost to improve CSP to form Ghost CSP network,at the same time,the Ghost idea is used in feature fusion,classification and regression network to optimize and adjust the network parameters,and the Si LU function is chosen in activation layers.These improvement measures achieve the effects of optimizing the size of the model,improving the detection speed and enhancing the generalization ability.According to the experiments results,the average accuracy of Ghost-Retina Net algorithm can reach 97.17%,the model size is only8.75 MB,and the detection speed attains 50.8 FPS.The Ghost-Retina Net achieves the effect of lightweight under the condition of meeting the detection accuracy,and provides the core algorithm basis for the implementation of photovoltaic panel shadow detection system.On the basis of the above algorithm research,a photovoltaic panel shadow detection system based on Ghost-Retina Net algorithm is constructed and developed by using C++language and Qt5.14.0 visual development tool.The system realizes the detection of shadow pictures and videos of local photovoltaic panels,as well as the real-time detection of online video streams.To a certain extent,it reduces the difficulty of photovoltaic system fault detection and the cost of system maintenance,which has a certain practical application value.