Researchers have developed an enhanced DeepLabV3+ model for more accurate classification of urban vegetation from UAV images.

The modified model addresses common issues in remote sensing, such as misclassification and omissions, especially in complex urban areas.

Feature engineering with the ReliefF algorithm has been incorporated to provide the deep learning model with more detailed sample features.

Upgrades to the DeepLabV3+ include a new backbone network (MoblieNetV2), adjustments to pooling rates, and the integration of attention mechanisms.

Experimental results show that the improved method surpasses other networks in overall accuracy, MarcoF1 score, and mean intersection over union.

The technique offers a promising solution for rapid and precise urban vegetation mapping, vital for urban planning and environmental management.