基于改进UNet的轻量级实时语义分割方法OA

Semantic segmentation plays a significant role in the field of autonomous driving,enabling autono-mous vehicles to parse rich environmental information from road images.However,traditional semantic segmentation techniques have a relatively large computational burden and a high demand for hardware memory.To solve this prob-lem,a lightweight real-time semantic segmentation method based on improved UNet(Convolutional Networks for Bio-medical Image Segmentation)is proposed.An innovative LDIB(Lightweight Double-branch Interactive Bottleneck)is integrated in the backbone network of the model,thereby significantly improving the accuracy of the model for target segmentation.LDIB integrates depthwise separable convolution technology with the feature fusion strategy of element-by-element multiplication,which not only effectively reduces the number of parameters of the model but also enhances the fusion effect of features.The experimental results show that the proposed model can quickly and accurately seg-ment different targets in road scenarios with dense targets.The mIoU(mean Intersection-over-Union)ratio in the open-source dataset Cityscapes is 71.5%,and the mIoU in CamVid is 69.49%.