To address the challenges of weak hotspot signals and blurred features in infrared images of photovoltaic modules captured by drones
as well as the high miss and false detection rates in automated inspections
this paper presents an optimized version of the YOLOv5 deep learning model to enhance its ability to detect faint infrared hotspots. First
by incorporating a hybrid feature module (HFM)
the network's feature extraction capability is further enhanced
leading to a significant improvement in detection accuracy. Additionally
a hierarchical aggregation Injection (HAI) network is employed after the backbone of YOLOv5
enabling the convolutional kernels to capture global feature information
thereby enforcing high detection precision. Finally
the WIoU loss function is introduced to improve the original bounding box loss and increase the detection speed of difficult samples. The experimental results show that the mAP of the improved model reaches 80.3%
which is an improvement of 5 percentage points compared to the original model.
WANG Y D, ITAKO K, KUDOH T, et al.Voltage-based hot-spot detection method for PV string using projector[C]//2016 IEEE International Conference on Power and Renewable Energy (ICPRE). Shanghai, China, 2016: 570-574.
LI Y L, DING K, ZHANG J W, et al.A fault diagnosis method for photovoltaic arrays based on fault parameters identification[J]. Renewable energy, 2019, 143: 52-63.
SUN T Y, XING H S, CAO S X, et al.A novel detection method for hot spots of photovoltaic (PV) panels using improved anchors and prediction heads of YOLOv5 network[J]. Energy reports, 2022, 8: 1219-1229.