Research On Blind Track Segmentation Method Based On Lightweight Convolutional Neural Network

As an important road facility to assist the visually impaired to travel,the blind track has not played its due role due to the occupation of the road by bicycles and cars.Therefore,it is of great significance to identify the blind track.Blind track segmentation is an important part of the blind track recognition system.The existing blind track segmentation algorithms have problems such as poor segmentation rate and single processing method.With the rapid development of deep learning technology,it has important practical value to develop an effective method for segmentation of blind track pictures.Based on the convolutional neural network,this paper proposes a blind track segmentation method based on lightweight convolutional neural network to address the existing problems of the existing blind track segmentation method and network structure,combining the attention mechanism and the multi-branch weight sharing method.The network uses an asymmetric encoder decoder architecture.The encoder network is used to output the convolutional feature map of the image,and the decoder network is used to perform multiple upsampling of the feature map and output the semantic segmentation map of the image.Using different channel attention modules to adjust the lightweight convolutional neural network structure to improve network performance without significantly increasing network parameters;Using the multi-branch weight sharing method to improve the blind track segmentation algorithm based on lightweight convolutional neural networks,due to the small parameters and calculations involved in the network,the structure is lighter,and based on the Cityscapes data set,a comparative experiment was carried out,and it was found that the network performance with the channel attention mechanism and the multi-branch weight sharing method was the best.Taking blind track pictures as the research object,a large-scale blind track picture data set is constructed,and the data set is preprocessed,including operations such as video conversion to picture and picture cropping.Afterwards,1528 pictures of blind track were made into pictures of blind track with category tags.A lightweight network with channel attention mechanism and multi-branch weight sharing method is used to simulate the blind track picture collection.The experimental results show that the algorithm can effectively improve the real-time performance and accuracy of the blind track picture segmentation.