Enhance PV Panel Detection Using Drone Equipped With RTK

Abstract

Solar energy is getting a lot of traction due to the reduced cost and friendlier to the environment compared to fossil fuel. It is essential to inspect the PV farms to ensure that the correct capacity produced through early PV fault detection. We proposed a full autonomous solution, where the drone mission is programmed to follow a specific Global Positioning System (GPS) waypoints. The collected videos will undergo various image processing techniques to detect and track the PV panels. In this paper, we tried two different PV panel detection approaches. Both detections gave acceptable results. The first detection relies on various image processing techniques. The second detection relies on deep learning architecture called mask Region-based Convolution Neural Network (R-CNN). After that, we track the PV panels in every frame using camera data alone. The advantage of tracking the PV panels is to ensure unrepeated PV panel through tagging even if the drone flies over the panel again since each PV panel will be associated with a tag. The next step will be to test the PV panel’s proposed detection and tracking algorithm on a larger solar farm.

Research on Visual Recognition Method for Defect Detection in Photovoltaic Cells Based on Deep Learning

“…As a result, precision (P) and mean average precision (mAP) increased by 6.3% and 9.4%, respectively, while the parameter count decreased by 62.65%. Compared to the SSD, Faster R-CNN [27], YOLO V3, and YOLO V4 models, precision (P) improved by 15.62%, 21.47%, 9.54%, and 13.75%, respectively, and mean average precision (mAP) improved by 13.%, 18.12%, 22.19%, and 11.8%, respectively. These comparative results demonstrate that the YOLOv5-BDL model achieves improved detection accuracy while maintaining superior performance in terms of memory and other software costs.…”

GBH-YOLOv5: Ghost Convolution with BottleneckCSP and Tiny Target Prediction Head Incorporating YOLOv5 for PV Panel Defect Detection

“…In addition, the validation set was held-out until the last test. Comparing the methods proposed in this paper with the five techniques discussed in the literature, which are the YOLOv3-based method from Tommaso et al [33], the Faster-RCNN-based method from Girshick et al [54], the SVM-based method from Mantel et al [32], the Mask-RCNN-based method from Almazroue et al [55], and the SSD-based method from Ren et al [56], the results showed that our performance on the Multi-Defect dataset was much better than the other models. The specific mAP performance is listed in Table 3.…”

“…Tommaso et al [33] 57.9 ± 0.07 Girshick et al [54] 69.3 ± 0.06 Mantel et al [32] 45.3 ± 0.07 Almazroue et al [55] 51.2 ± 0.06 Ren et al [56] 30.8 ± 0.06 Proposed GBH-YOLOv5 97.8 ± 0.02…”