Intelligent Identification of Taratan Photovoltaic Power Plant Using High-Resolution Remote Sensing Imagery

Photovoltaic (PV) power generation is becoming increasingly important as a clean, renewable form of energy, and accurately identifying and calculating the area of PV panels is critical to the planning, monitoring, and optimization of PV systems. Remote sensing technology, due to its large-scale, cyclical, and non-contact data acquisition capabilities, can efficiently obtain imagery of PV power plants across extensive areas, thereby establishing a comprehensive database for PV panel identification. However, in practical applications, intelligent identification of PV power plants still faces several challenges. Individual elongated PV panels with low spatial distinguishability are difficult to recognize, while planned undeveloped areas in planning stages exhibit spatial signatures closely resembling built-up regions, posing challenges for effective discrimination based solely on spatial attributes. Therefore, to address the above challenges, this paper proposes the Spectral-High Resolution Network (S-HRNet), based on HRNet and incorporating a spectral feature extraction module. The model is validated using the Taratan PV power plant case study. The model extracts spectral information from images to enhance existing feature representation while optimizing the feature fusion mechanism, enabling accurate PV panel recognition through the synergy of spatial-spectral features. The model demonstrated effectiveness in the Taratan PV plant dataset, outperforming HRNet by achieving improvements of 1.47 % and 0.79 % in IoU and F1, respectively, and notably reaching 95.28 % IoU and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$97.34 \% ~\mathrm{F} 1$</tex>. When applied to the Taratan PV plant, the model identified a photovoltaic panel area of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$251.42 ~\text{km}^{2}$</tex>, outperforming HRNet's <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$252.52 ~\text{km}^{2}$</tex> and closely matching the visually interpreted <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$251.77 ~\text{km}^{2}$</tex>. This result verifies the model's practical value in large-scale PV plant monitoring and provides reliable technical support for applications such as dynamic PV resource assessment and power generation efficiency prediction.