DBGENet: A Dual-Branch Global-context and Edge-aware Encoder Network for Photovoltaic Panel Segmentation under Complex Remote Sensing Scenes

Engineering notes

Engineering notes will be added by the Power for Solar editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为光伏效率、钙钛矿太阳能电池、储能技术、太阳能热利用、BIPV、并网技术等高价值论文补充中文说明。

Original abstract

Abstract With the rapid expansion of photovoltaic (PV) installations, accurate segmentation of PV panels from remote sensing imagery has become essential for efficient operation and performance assessment of PV power plants. However, complex backgrounds and strong reflective characteristics make it challenging for existing semantic segmentation models to simultaneously capture global structural information and preserve boundary details. To address these issues, this paper proposes DBGENet, a unified PV panel segmentation network that integrates global context modeling, local detail preservation, and edge-aware learning. A dual-branch encoder is designed to collaboratively extract global semantic features and fine-grained spatial information, enhancing the structural understanding of PV arrays. In addition, the global channel-spatial attention (GCSA) module adaptively emphasizes discriminative features while suppressing background interference. The edge-aware lightweight bottleneck (EALB) module is further embedded to incorporate explicit boundary priors, thereby improving the continuity and integrity of the segmentation results. Experiments on multi-resolution PV remote sensing datasets demonstrate that DBGENet consistently outperforms mainstream segmentation models under complex backgrounds. The proposed method achieves an mIoU of 94.48% for rooftop PV at a 0.1 m resolution and 96.63% for ground-mounted PV at a 0.3 m resolution, demonstrating clear advantages in boundary accuracy, small-scale target recognition, and robustness across diverse scenes.

Links and sources

• PDF from original source

Comments