Research On Image Semantic Segmentation Based On Convolutional Neural Network

With the development and progress of deep learning technology and the rapid development of convolutional neural networks,segmentation by pixels on the semantic layer has become a hot topic in current research,and image semantic segmentation is more and more widely used in many fields of current life.However,in real life,image semantic segmentation technology still has room for improvement due to the complexity of the network structure,the lack of information in feature transmission,and the difficulty of falling into local optima during the training process.This paper is based on the research of convolutional neural network,analyzes and studies the Deep Lab V3+ model,and optimizes the difficulties and model deficiencies in the current field.Experiments are carried out on a large amount of data from the PASCAL VOC and Cityscapes data sets.The improved network is more The original network has been significantly improved.The research content of this article is as follows:(1)Aiming at the problem of chaos in the feature part extracted by the Deep Lab V3+ model,this paper introduces the dilated residual network as the feature extraction network,and improves the ASPP model.The improved network obtains more contextual information through a larger receptive field,and obtains semantic features with better distinguishability.Experiments verify that the accuracy of the network has been significantly improved.(2)In the Deep Lab V3+ network model,part of the feature information is rough when performing feature extraction.Therefore,in response to this problem,this paper introduces two different hybrid attention mechanisms for improvement.The improved model uses the attention mechanism to obtain the feature parameters that are worthy of attention,so that the network pays attention to the features that are more worthy of attention,and thus obtains more discriminative segmentation results.Experiments have proved that the accuracy of the improved method has been improved.(3)The features extracted by the Deep Lab V3+ network model at different stages through the convolutional neural network have their own advantages and disadvantages.Therefore,in response to this situation,the most appropriate upsampling function is selected through experimental comparison in the process of upsampling,and at the same time the decoder structure is improved,so that more abundant and detailed features can be extracted,which improves the accuracy of the model.Finally,an experimental comparison shows that the improved method in this paper has a better effect on image semantic segmentation.