基于CenterNet的光伏阵列遮盖物检测

光伏阵列的局部遮阴会导致光伏功率损失。光伏阵列表面的遮挡物是造成局部阴影的因素之一,因此提出了一种基于CenterNet的光伏阵列遮盖物检测方法。通过迁移学习预训练模型,分步训练获得检测模型。该模型在不同光照方向、不同遮挡程度以及不同远近距离等条件下,能够较精确地识别遮盖物种类并定位。实验结果表明,该遮盖物检测模型的mAP为0.81,检测速度为16 ms,优于所选的对比算法,具有较强的实时性。

Coverings Detection of Photovoltaic Array Based on CenterNet

The partial shading of photovoltaic array may lead to power loss of photovoltaic,the coverings on PV array is one of the factors causing partial shade.Therefore,a CenterNet based PV array covering detection method is proposed.Through the transfer learning pretraining model,step by step training to obtain the detection model.The model can accurately identify and locate the types of covering objects under different illumination directions,different occlusion degrees and different distance and distance.The experimental results show that the mAP of the mask detection model is 0.81 and the detection speed is 16 ms,which is better than the selected contrast algorithm,has strong real-time performance.