Research on fine segmentation method of hyperspectral remote sensing canopy images based on quantum-enhanced U-Net

In canopy coverage monitoring, traditional segmentation models face limitations including limited modeling ability, insufficient generalization performance, and small target missed detection due to challenges such as fragmented canopy boundary fuzziness, spectral confusion from multi-layer shadows, and hyperspectral feature redundancy. We proposed quantum-enhanced U-Net to achieve fine canopy segmentation. The model adopts U-Net as the backbone network and introduces a quantum feature distillation mechanism together with a quantum feature mapping mechanism. These components form a hybrid quantum–classical architecture to capture global relationships and local details. In addition, the convolutional block attention module and a dynamic fusion strategy are incorporated for adaptive feature optimization. Experimental comparisons on the OpenAerialMap-Tree Canopy Dataset demonstrate the model’s superior performance. The unmanned aerial vehicle image dataset results demonstrate its strong cross-scenario generalization capability. Ablation experiments further validate the contributions of the quantum feature distillation module, quantum feature mapping module, and convolutional block attention module. Experimental outcomes indicate that this approach significantly enhances hyperspectral remote sensing canopy segmentation performance across multiple scenarios, demonstrating practical value in ecological monitoring.