Geographical Localization Algorithm for Photovoltaic Panels Based on Monocular Depth Estimation

doi:10.3778/j.issn.1002-8331.2403-0009

Abstract

Abstract: In the unmanned aerial vehicle (UAV) inspection of photovoltaic power plants, accurately locating the geographic positions of solar panels is crucial. Currently, most positioning methods rely on geographic information data, multi-view images, or LiDAR, but these methods struggle to rapidly locate the geographic positions of targets in unknown and complex environments. With the development of monocular depth estimation (MDE) using deep learning, MDE networks have demonstrated high depth prediction accuracy in road scenes. Based on this, a novel photovoltaic panel geographic localization algorithm is proposed, which utilizes an MDE network to estimate the target distance and transforms the pixel coordinates of solar panels to geographic coordinates based on the camera imaging model. To address the poor depth prediction accuracy of classical MDE in UAV scenes with variable shooting angles and long distances, a Swin transformer-based optimized MDE network (SwinDenseDepth) is designed for this scenario. This network enhances the depth perception capability for UAV scenes by employing an encoder composed of Swin-transformers and integrating dense connection structures with channel-wise spatial attention fusion modules, thereby improving the accuracy of depth estimation using semantic and contextual information. Experimental results show that compared to the current mainstream MDE methods, SwinDenseDepth can more accurately predict distances in UAV images and the positioning algorithm achieves a positioning error within the range of 1~2 meters in images captured at inspection heights of 30~60 meters, meeting the practical requirements of solar panel localization.

Key words: geographical target localization, monocular depth estimation, visual geolocation, unmanned aerial vehicle (UAV), photovoltaic panels component

摘要： 在光伏电站的无人机巡检中，准确定位光伏板的地理位置是关键。目前大多数定位方法依赖于地理信息数据、多视角图像或激光雷达，但这些方法难以在未知复杂环境下快速定位目标的地理位置。随着深度学习的单目深度估计（MDE）发展，MDE网络在道路场景下已展现出较高的深度预测精度。基于此，提出了一种全新的光伏板地理定位算法，采用MDE网络估算目标距离，并根据相机成像模型将光伏板的像素坐标转换至地理坐标。为了解决经典MDE在拍摄视角多变和距离较远的无人机场景下深度预测精度不佳，设计了针对该场景优化的MDE网络（SwinDenseDepth），采用由Swin Transformer组成的编码器增强对无人机场景的深度感知能力，并结合密集连接结构与通道空间注意力融合模块，利用语义和上下文信息提高深度估计的准确性。实验结果表明相比于目前主流的MDE能更为准确地预测无人机图像中距离，并且定位算法在巡检高度30~60 m的图像中定位误差在1~2 m范围内，满足定位光伏板的实际需求。

关键词: 目标地理定位, 单目深度估计, 视觉地理定位, 无人机（UAV）, 光伏板组件

NI Yuansong, HAN Jun, HU Guangyi, WANG Wenshuai. Geographical Localization Algorithm for Photovoltaic Panels Based on Monocular Depth Estimation[J]. Computer Engineering and Applications, 2025, 61(14): 353-361.

倪源松, 韩军, 胡广怡, 王文帅. 基于单目深度估计的光伏板地理定位算法[J]. 计算机工程与应用, 2025, 61(14): 353-361.

References

[1] GALLARDO-SAAVEDRA S, HERNáNDEZ-CALLEJO L, DUQUE-PEREZ O. Technological review of the instrumentation used in aerial thermographic inspection of photovoltaic plants[J]. Renewable and Sustainable Energy Reviews, 2018, 93: 566-579.
[2] PEINADO G A, PLIEGO M A, GARCíA M F P. Survey of maintenance management for photovoltaic power systems[J]. Renewable and Sustainable Energy Reviews, 2020, 134: 110347.
[3] HIJJAWI U, LAKSHMINARAYANA S, XU T H, et al. A review of automated solar photovoltaic defect detection systems: approaches, challenges, and future orientations[J]. Solar Energy, 2023, 266: 112186.
[4] 朱得糠, 李东泽, 郭鸿博, 等. 无人机视觉地理定位研究综述[J]. 导航与控制, 2023, 22(3): 21-33.
ZHU D K, LI D Z, GUO H B, et al. A review of vision-based geolocation using UAVs[J]. Navigation and Control, 2023, 22(3): 21-33.
[5] 于越. 基于无人机的航拍车辆目标视觉检测与地理定位系统[D]. 南京: 东南大学, 2019.
YU Y. Vehicle target vision detection and geo-location system based on UAV[D]. Nanjing: Southeast University, 2019.
[6] ZHAO X Y, PU F L, WANG Z H, et al. Detection, tracking, and geolocation of moving vehicle from UAV using monocular camera[J]. IEEE Access, 2019, 7: 101160-101170.
[7] 陈晨, 关棒磊, 尚洋, 等. 受限观测条件下光电对地定位的全局最优化方法[J]. 光学学报, 2023, 43(12): 144-152.
CHEN C, GUAN B L, SHANG Y, et al. Global optimization method for ground target localization of electrooptical platform under limited observation conditions[J]. Acta Optica Sinica, 2023, 43(12): 144-152.
[8] 李梓豪, 匡海鹏, 张泓, 等. 基于数字高程模型高程快速迭代的航拍图像目标定位方法[J]. 中国光学 (中英文), 2023, 16(4): 777-787.
LI Z H, KUANG H P, ZHANG H, et al. A target location method for aerial images through fast iteration of elevation based on DEM[J]. Chinese Optics, 2023, 16(4): 777-787.
[9] NASSAR A, AMER K, ELHAKIM R, et al. A deep CNN-based framework for enhanced aerial imagery registration with applications to UAV geolocalization[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2018: 1513-1523.
[10] 廖杰, 吴明晖. 一种面向无人机的阵列相机设计及其目标定位算法研究[J]. 农业装备与车辆工程, 2023, 61(3): 95-100.
LIAO J, WU M H. Design of an array camera for UAV and research on target localization algorithm[J]. Agricultural Equipment & Vehicle Engineering, 2023, 61(3): 95-100.
[11] SOHN S, LEE B, KIM J, et al. Vision-based real-time target localization for single-antenna GPS-guided UAV[J]. IEEE Transactions on Aerospace and Electronic Systems, 2008, 44(4): 1391-1401.
[12] TAGHAVI E, SONG D, THARMARASA R, et al. Geo-registration and geo-location using two airborne video sensors[J]. IEEE Transactions on Aerospace and Electronic Systems, 2020, 56(4): 2910-2921.
[13] MING Y, MENG X Y, FAN C X, et al. Deep learning for monocular depth estimation: a review[J]. Neurocomputing, 2021, 438: 14-33.
[14] EIGEN D, PUHRSCH C, FERGUS R. Depth map prediction from a single image using a multi-scale deep network[J]. arXiv:1406,2283, 2014.
[15] GODARD C, MAC AODHA O, FIRMAN M, et al. Digging into self-supervised monocular depth estimation[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019: 3828-3838.
[16] DOSOVITSKIY A, BEYER L, KOLESNIKOV A, et al. An image is worth 16×16 words: transformers for image recognition at scale[J]. arXiv:2010.11929, 2020.
[17] RANFTL R, BOCHKOVSKIY A, KOLTUN V. Vision transformers for dense prediction[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021: 12179-12188.
[18] LIU Z, LIN Y, CAO Y, et al. Swin transformer: hierarchical vision transformer using shifted windows[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021: 10012-10022.
[19] SHAH S, DEY D, LOVETT C, et al. AirSim: high-fidelity visual and physical simulation for autonomous vehicles[C]//Proceedings of the Field and Service Robotics. Cham: Springer International Publishing, 2018: 621-635.
[20] LYU X Y, LIU L, WANG M M, et al. HR-depth: high resolution self-supervised monocular depth estimation[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2021: 2294-2301.
[21] WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]//Proceedings of the European Conference on Computer Vision. Cham: Springer International Publishing, 2018: 3-19.
[22] ZHANG N, NEX F, VOSSELMAN G, et al. Lite-Mono: a lightweight CNN and transformer architecture for self-supervised monocular depth estimation[J]. arXiv:2211. 13202, 2022.
[23] FAROOQ B S, ALHASHIM I, WONKA P. AdaBins: depth estimation using adaptive bins[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2021: 4008-4017.
[24] KIM D, KA W, AHN P, et al. Global-local path networks for monocular depth estimation with vertical cutdepth[J]. arXiv:2201.07436, 2022.