Research On Road Scene Semantic Segmentation Algorithm Based On Deep Learning

Image segmentation technology divides images into different regions based on semantic features in the image.With the development of deep learning techniques,semantic information features can be extracted from images by combining them with image segmentation.However,in the process of semantic segmentation,interference from factors such as uneven lighting and background noise often affects the segmentation effect.In the field of autonomous driving,there is a high demand for the inference speed of the semantic segmentation model of road scenes to ensure safe driving in high-speed scenarios,while also ensuring segmentation accuracy.However,most existing models are complex and require high-performance hardware devices to support,making it difficult to apply them on mobile devices.Therefore,solving the balance problem between segmentation accuracy and inference speed in the semantic segmentation task in the field of autonomous driving and promoting related research on road scene semantic segmentation is of great value.The main research content of this paper is as follows:(1)This article have constructed a residual-based bilateral road scene image semantic segmentation model called Bi Se Net-Res18 to address the issues of segmentation accuracy and speed in road scene image semantic segmentation models.The model adopts the bilateral idea of Bi Se Net,dividing it into spatial and contextual branches.The lightweight Res Net18 model is chosen as the context feature extraction network.(2)This article have designed a spatial attention mechanism-based residual bilateral image semantic segmentation model called SANet.This model incorporates a spatial attention module(SA)to generate spatial attention weights for the image,allowing the model to focus on key areas and avoid interference from irrelevant regions.Additionally,the feature fusion module is used to solve the problem of simple feature fusion in image segmentation models,which often results in issues with combining semantic and spatial features.(3)To further improve the performance of semantic segmentation models for road scene images,a residual bilateral image semantic segmentation model called CSTNet is proposed based on an interactive fusion mechanism.This model incorporates an interactive fusion attention module(CSTM)and a lightweight spatial attention module(CSSE)to capture rich semantic and spatial information,respectively.The interactive fusion attention module is designed to capture rich semantic information,while the spatial attention module captures rich spatial information.This allows the model to enhance the extraction of important features,reduce the impact of non-key features,and avoid interference from irrelevant features,thereby improving the feature extraction ability of the original network and ultimately enhancing the segmentation performance of the model.