Road Segmentation Algorithm Based on Improved DeepLabv3+
The use of Deep Learning-based remote sensing image segmentation technology is becoming increasingly widespread.In response to the problems of large parameter quantities and poor results in extracting details in existing algorithms,a road image segmentation method based on improved DeepLabv3+is proposed.Introducing the lightweight network MobileNetV2 into an improved pooling pyramid model to extract mid-order feature maps,which enhance the correlation between different receptive fields.A multi-scale concatenation fusion method is adopted to generate high-order feature maps,while introducing attention mechanisms to further enhance the extraction effect of image features.The experimental results show that the proposed method improves mIoU by 5％compared to the DeepLabv3+model,effectively enhancing the segmentation accuracy of remote sensing images.
国家科技期刊平台
NSTL
万方数据
