Multi-Scale Feature Fusion-Based Implementation of Vegetation Segmentation for Remote Sensing Imagery

Vegetation segmentation in remotely sensed images is crucial for ecological monitoring and resource management, but the complex scenarios of vegetation segmentation make the image segmentation task very challenging. In this paper, we proposed VegSegNet, a fine-tuned vegetation segmentation model based on the DeepLabv3+ architecture, multi-scale feature fusion, and a vegetation segmentation dataset, for the problems of complex background interference and category imbalance in remote sensing images. Experiments on a five-species vegetation type segmentation dataset demonstrate the superiority of the model, achieving a mean intersection-over-union (mIoU) of 55.26%, an accuracy of 79.10%, a Kappa coefficient of 71.31%, and a Dice coefficient of 67.61%. The mIoU of the VegSegNet is 13.31% and 15.76% higher than U-Net and SegNet, respectively. This method provides strong technical support for geographic information system applications such as forest resource monitoring and ecological red line delineation.