Automated Overheated Region Object Detection of Photovoltaic Module With Thermography Image
Yonghe Su, Fei Tao, Jian Jin, Changzhi Zhang
Abstract
The overheated region is an abnormal condition for photovoltaic (PV) module in the routine inspection of PV plant. Many studies have invited thermography images to identify overheated region problems (e.g., hotspot), since they are easy and cheap to be collected. But these studies fail to automatically recognize the specific types and the exact positions of different potential overheated region targets in a single thermography image of PV module. Moreover, some overheated regions of PV module are small in scale, which induces that many traditional approaches fail to identify some overheated regions effectively and efficiently. Accordingly, a deep learning-based framework is proposed to handle these problems. First, multiple large-scale images are transformed from thermography images with overheated regions to precisely detect overheated region targets. Then, regions of interest are extracted from these images to bound potential regions that may exist overheated regions. Finally, a deep joint learning model is used to recognize the overheated region type and position from these regions. To benchmark the proposed framework, categories of experiments are conducted over the collected dataset. It proves that the proposed approach outperforms benchmarked approaches in terms of effectiveness and efficiency.