Research On Image Semantic Segmentation Method Based On Encoder-Decoder Network And Feature Encoding

Semantic segmentation is one of the most fundamental tasks in computer vision.It plays a key role in image understanding and provides a significant impact on various real-world applications,such as autonomous driving,robotics,and medical image diagnosis.In recent years,although the research on semantic segmentation based on deep convolutional neural networks has made significant progress,there are still many challenges.Based on deep learning technology,this paper studies the semantic segmentation model based on deep convolutional network from two aspects,namely,the research of improving semantic segmentation accuracy by fusing context semantic information and spatial information,and using elastic feature pyramid module reduces the time complexity of the semantic segmentation model and achieves real-time semantic segmentation research.The work done in this article is as follows:Firstly,this paper gives a detailed overview of current deep learning-based semantic segmentation algorithms,and introduces in detail the deep model framework commonly used in image semantic segmentation.Then,according to the four dimensions of the method of improving the receptive area of the network,the method based on feature fusion,the method based on attention mechanism,and the method based on neural network architecture search,the current most effective semantic segmentation methods are discussed.Finally,the most promising real-time semantic segmentation method in the future is elaborated.By reviewing the current semantic segmentation methods,we can clearly understand the development of semantic segmentation technology,and inspire research ideas for related researchers.Secondly,current research on image semantic segmentation basically focuses on how to extract effective semantic context information and restore spatial detail information to design more efficient algorithms.In order to improve the semantic representation ability of the network and simultaneously establish the spatial relationship between pixels,this paper proposes a new semantic segmentation solution based on context and shallow spatial encoder-decoder network.A two-branch strategy is adopted on the encoder.One branch is called contextual branch,which is constructed with a proposed semantic context module to obtain high-quality semantic context information.And the other branch is spatial branch,which is designed as an inverse U-shaped structure with the proposed chain-reverse residual module to enhance semantic information and preserve spatial details.Moreover,a refinement module is proposed to add to the decoder to further refine the fusion features of context information and spatial information.The proposed approach achieves competitive results on the CamVid,SUN RGB-D and Cityscapes benchmarks.Finally,a high-quality semantic segmentation model requires a lot of computing resources.In order to realize real-time applications,a feature encoding fusion network is proposed.In order to encode multi-scale feature information while reducing memory overhead,an elastic feature pyramid module is designed as the basic building block of the feature extraction network.Then,a multi-path semantic module is designed at the end of the feature extraction network to optimize the learning of semantic features and gradient back-propagation.Finally,the dual-attention feature fusion module is used to selectively fuse features at different levels.The proposed method completes the semantic segmentation of urban road scenes with fewer parameters and faster processing speed on the CamVid and Cityscapes datasets.