Research on Semantic Segmentation Based on Improved PSPNet

Semantic segmentation covers the fields of machine learning, computer vision, image processing and human-computer interaction, has broad application prospects and great application value. But the existing semantic segmantation models still have complex structure, too mang parameters and missing context semantic features problem. To tackle this problems,we propose an improved method based on PSPNet. Specifically, we replace the traditional ResNet with MobileNetV2. Based on this, we extract and preserve the context features effectively, termed contextual semantic features supplement module. Finally, we regard the introduction of the traditional level set method as after processing to the model. Compared with the traditional method, our method has a significant improvement in speed and accuracy on Baidu people segmentation dataset, Adobe's portal segmentation dataset and Sebastien Marcel static hand posture database dataset. To verify the robustness of our method, we make correlation test of multi objects, complex background and low resolution. Result demonstrate the superiority of the proposed approach.