Research On Image Semantic Segmentation Based On Convolutional Neural Network

As one of the computer vision tasks,image semantic segmentation essentially divides an image into several different and meaningful regions.With the development of deep learning,image semantic segmentation technology has been widely used in medical images,remote sensing maps,driverless cars and face recognition.With the joint efforts of researchers,image semantic segmentation algorithms based on convolutional neural network have been proposed one after another,but there are still some problems to be solved.Firstly,with the deepening of the encoder network,the decoder network has the problem that it is difficult to integrate multi-scale features;secondly,the continuous down-sampling operation leads to the loss of image edge information;thirdly,in the deep network,the model has poor segmentation results for small objects.To solve these problems,this paper studies the current mainstream semantic segmentation methods and puts forward the corresponding improvement methods.The specific research contents are as follows:(1)In order to solve the problem of image edge information loss caused by the difficulty of multi-scale feature fusion and continuous down-sampling operation,this paper proposes an image semantic segmentation method based on multi-scale feature fusion(EFFNet).EFFNet consists of four basic modules,which are feature fusion module(FFM),spatial information module(SIM),global pooling module(GPM)and boundary refinement module(BRM).FFM adopts attention mechanism and residual structure to improve the efficiency of multi-scale feature fusion.SIM consists of the convolution layer and the average pooling layer,and its purpose is to assist in locating the edge information of the object by providing additional spatial details.GPM extracts the global context information of the image,which enhances the ability to segment large objects,thus significantly improving the performance of the model.BRM takes the residual structure as the core to refine the boundary of the feature map.(2)In order to solve the problem of poor segmentation results for small-scale objects,this paper proposes a semantic segmentation method based on dual attention mechanism(DAMNet).The core idea of DAMNet is a dual attention module,which consists of the self-attention mechanism(SAM)and the channel attention mechanism(CAM).SAM models the interdependence between feature pixels,which makes the pixels of small size targets more closely related,and then produces better segmentation results;CAM highlights the key features in the form of weight vectors,so as to obtain richer and deeper semantic information.Finally,a large number of experiments on the PASCAL VOC 2012 data set demonstrate the effectiveness and efficiency of the proposed method.Figure 24 Table 13 Reference 65...