To address the accuracy and efficiency challenges of deep learning models across various data sources,this study employs the U-Net framework,combining attention modules and lightweight DenseNet to introduce DA-UNet.Through training on a multi-source remote sensing image ground photovoltaic dataset,DA-UNet autonomously extracts ground photovoltaic features from diverse high-resolution remote sensing images.In comparison to conventional networks,DA-UNet demonstrates superior performance in visual quality and precision evaluation,with an overall accuracy(OA),F1 score,and intersection over union(IoU)of 95.92％,94.68％,and 90.91％,respectively.Furthermore,DA-UNet exhibits robust generalization and applicability to high-resolution remote sensing images in other regions,offering substantial practical value for ground photovoltaic extraction.
关键词
深度学习/密集网络/注意力机制/地面光伏/跨数据
Key words
deep learning/DenseNet/attention mechanisms/ground photovoltaic/across data
维普
万方数据