Photovoltaic Panel Hot Spot Recognition Based on Lightweight SSD

Abstract

An intelligent recognition technique of photovoltaic panel hot spot based on UAV and target detection algorithm is proposed in order to address the issues of low efficiency and high cost of manual operation and maintenance of photovoltaic panels. When the light intensity distribution is uneven, the photovoltaic panel may produce hot spot effect, damage the photovoltaic cell module, and cause system failure. In this research, a lightweight SSD-based photovoltaic panel hot spot recognition method is proposed, which addresses the issues of high target detection complexity and difficulties in implementing the algorithm on edge devices. The complexity of the model is decreased by using Resnet50 as the backbone network in place of VGG16. To increase the model's accuracy and complete the identification of solar module hot spots, the pyramid module and the Coordinate Attention module are introduced. Tested on the aerial photovoltaic panel dataset, the mAP of the proposed model reaches 86.28%, which is 2.58% higher than the original SSD target detection algorithm, and the proposed model compresses 35.90% of parameters and 48.47% of the calculation, which can meet the requirements of carrying the model to the UAV.