Research On Real-time Semantic Segmentation Algorithm Based On Feature Reuse

As one of the basic tasks of computer vision,image semantic segmentation occupies an important position in computer vision research,and has a wide range of applications in the fields of automatic driving,scene understanding,and medical image analysis.At present,most researches still focus on improving the accuracy of segmentation.They usually use huge convolutional neural networks as the backbone of the model,which severely slows down the model's inference speed,and has insufficient performance in real-time segmentation and cannot be applied to mobile terminals.However,if a lightweight convolutional network is used as an encoder,although the segmentation speed will be significantly improved,the features extracted by the encoder are not rich enough,resulting in low segmentation accuracy.Therefore,this article takes this as the research point and focuses on solving the trade-off between the accuracy and speed of image semantic segmentation.In order to solve the above problems,this paper proposes a real-time semantic segmentation algorithm based on efficient feature reuse of lightweight backbone networks.The main research work of this paper is as follows:(1)Aiming at the problem of insufficient high-level feature space information in a single lightweight convolutional network,several one-way feature reuse methods with different focuses are proposed,and the improved lightweight backbone network verifies the effectiveness of these methods Effects,analysis of experimental results and reasons.(2)A lightweight backbone network that can extract multi-scale features is proposed,which reduces the amount of convolutional neural network parameters and computational complexity while improving the ability to extract multi-scale features,effectively improving the extraction of lightweight networks Characteristic ability.(3)An efficient two-way feature reuse algorithm is proposed.The two-way feature reuse method is designed by setting up two backbone networks,which solves the problem of insufficient high-level feature space information.The two features complement each other and improve the performance of the algorithm.The algorithm proposed in this paper is verified on two semantic segmentation data sets.Compared with the same type of real-time semantic segmentation algorithms,the method proposed in this paper achieves better segmentation results and achieves a trade-off between accuracy and speed.