Extraction of Gansu Photovoltaic Power Station Based on GF-1 Satellite

Aiming at the problem of low extraction accuracy of photovoltaic power plants in large-scale complex scenarios, an improved Mask2Former photovoltaic power plant extraction network is proposed, which introduces a decomable convolutional network as the backbone network and Attention Module, enhances the model's ability to extract irregular photovoltaic power plants and distinguish background similar objects, and improves the network's ability to extract photovoltaic power plants. Using domestically produced GF1 2-meter resolution data to extract photovoltaic power plants in Gansu, the improved Mask2Former model achieved an F1 score of 0.947, an increase of 0.07. According to statistics, in 2022, the deployment area of photovoltaics in Gansu Province was 283.18km<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>, showing a trend of more in the north and less in the south. This method achieves rapid extraction of photovoltaics in complex scenarios, providing strong support for dynamic monitoring of photovoltaic power plants.