Research On Image Semantic Segmentation Method Based On Deep Learning

The definition of image semantic segmentation is to assign a predefined label representing its semantic category to each pixel in the image.Image semantic segmentation has a wide range of applications in areas such as autonomous driving,medical image recognition,and satellite remote sensing information systems.Because image semantic segmentation applies semantic information when segmenting images,it has become an important research direction of image segmentation.Deep learning method is widely used due to its strong autonomous learning ability.The application of deep learning method to image semantic segmentation is one of the hot research directions in the field of image processing,which has theoretical research value and practical application prospect.In view of the key technical issues of image semantic segmentation,combined with deep learning method to explore.The image semantic segmentation method based on the bilinear interpolation is summarized and analyzed,and the image semantic segmentation method based on the bicubic convolution interpolation is proposed.The image semantic segmentation method based on the bicubic convolution interpolation is an improvement and enhancement of the image semantic segmentation method based on the bilinear interpolation.Aiming at the problem of image semantic segmentation based on the bilinear interpolation in the upsampling process,the bilinear interpolation method can only extract information from four adjacent sampling points,with fewer sampling data points and reduced image segmentation accuracy.The bicubic convolution interpolation method replaces the bilinear interpolation method.The bicubic convolution interpolation method requires 16 sampling points.When the number of data points increases,it also takes into account the grayscale and grayscale change rate of adjacent points.Different weights are used to extract the information of neighboring pixel values,which can better restore the characteristics.The resolution of the category with fewer pixels in the image.In the upsampling operation of the encoder and the decoder,it is necessary to use the bicubic convolution interpolation method to enlarge the image and retain more detailed information.Based on the image semantic segmentation model based on the bicubic convolution interpolation method,an online matting and background changing system based on image semantic segmentation is designed and implemented from the perspective of practical application.In the Windows environment,the experiment uses the Tensorflow framework for environment configuration and training,and uses GPU to accelerate the neural network model with cuda.Using Pascal VOC 2012 enhanced version of the data set,through the experimental comparison between the image semantic segmentation model based on the bicubic convolution interpolation method and the image semantic segmentation model based on the bilinear interpolation method,the experimental effect of the image semantic segmentation model based on the bicubic convolution interpolation method is better,and the experiment verify that the proposed method is effective.