Full
Zhao Lijun, E-mail: 20190005@cqwu.edu.cn
Outline
To address the degradation of power generation efficiency caused by soiling accumulation in photovoltaic (PV) power plants, as well as the localization challenges faced by cleaning unmanned aerial vehicles (UAVs) due to limitations in edge computing power and complex background interference, this paper proposes an edge-computed cleaning UAV solution based on the Jetson Orin Nano platform. Within the YOLO11n framework, a lightweight photovoltaic panel soiling detection algorithm designated as GOP-YOLO11n is developed. By introducing GroupDC structure, the C3k2 modules in the backbone network undergo lightweight reconstruction, minimizing computational overhead while enhancing the feature representation of small targets against complex backgrounds. Furthermore, an OptimizedDC multi-scale feature extraction module is embedded into the feature fusion network to improve the perception of multi-scale targets, such as tiny bird droppings and elongated water stains. In the detection head, an upsampling and feature aggregation block (UP-Block) is implemented to strengthen multi-pixel pattern aggregation. For the experiments, a visible-light image dataset of PV panels captured by UAVs, yielding 4 190 valid imagespartitioned into training, validation, and test sets at an 8：1：1 ratio, with input resolution standardized to 960 pixels × 960 pixels. Under identical training and testing configurations, GOP-YOLO11n achieved a Precision of 91.1%, a Recall of 85.8%, a mAP@50 of 92.5%, and a mAP@50：0.95 of 76.8%, outperforming the baseline YOLO11n model by 2.9, 3.4, 1.8, and 4.4 percentage points, respectively. The model parameters were reduced from 2.582 M to 2.068 M, with a floating-point operation count of 6.7×109. Moreover, hardware deployment tests using TensorRT FP16 on the Jetson Orin Nano Super 8 G platform showed an inference speed of 114.7 f/s, a single-frame latency of 8.7 ms, and a GPU memory footprint of 1 685 MB. The results indicate that the proposed method effectively reconciles the need for high-accuracy multi-scale soiling detection in complex scenes with the constraints of lightweight edge-side real-time inference, thereby providing robust visual perception support for soiling recognition, localized cleaning strategies, and the intelligent operation and maintenance (O&M) of PV cleaning UAVs.
53
19
| 科 Family | 属数 Number of genus | 种数 Number of species | 占总种数比例 Percentage of total species (%) | 属 Genus | 种数 Number of species | 占总种数比例 Percentage of total species (%) |
|---|---|---|---|---|---|---|
| 鹅膏菌科Amanitaceae | 2 | 11 | 5.26 | 鹅膏菌属 Amanita | 10 | 4.78 |
| 小菇科 Mycenaceae | 2 | 12 | 5.74 | 丝盖伞属 Inocybe | 5 | 2.39 |
| 多孔菌科 Polyporaceae | 8 | 14 | 6.70 | 蜡蘑属 Laccaria | 5 | 2.39 |
| 红菇科 Russulaceae | 3 | 23 | 11.00 | 小皮伞属 Marasmius | 6 | 2.87 |
| 小菇属 Mycena | 11 | 5.26 | ||||
| 光柄菇属 Pluteus | 5 | 2.39 | ||||
| 红菇属 Russula | 17 | 8.13 | ||||
| 栓菌属 Trametes | 5 | 2.39 |
关闭全屏
