Image Semantic Segmentation Based On Improved Convolutional Neural Network

Image semantic segmentation is an important branch in the field of computer vision.Traditional methods often use thresholding,clustering and other methods to segment images.The vigorous development of deep learning technology has brought a new form of problem-solving for the field of computer vision.Compared with the original image semantic segmentation method,deep learning technology reduces a lot of intermediate steps and realizes the end-to-end structure.Although the addition of deep learning technology has greatly improved the effect of image semantic segmentation,there are still some problems in the field of image semantic segmentation,such as incomplete and incoherent object segmentation.However,the use of multi-scale information often leads to the increase of model size and reasoning time,which limits the application scenarios of image semantic segmentation model.To solve this problem,this paper proposes a MS-SEGNET model using multi-scale semantic information.The original SEGNET model makes full use of the multi-scale semantic information contained in the model,and increases a small amount of model parameters to ensure the reasoning speed of the model and improve the segmentation effect of the model.For image semantic segmentation task,the ultimate goal of the model is to classify each pixel accurately.Most image semantic segmentation models fit conditional probability distribution.Compared with joint probability distribution,conditional probability distribution can reflect a relatively limited internal pattern of the data,which limits the segmentation effect of the model(such as average intersection ratio).In order to solve this problem,this paper further proposes the image semantic segmentation network(MSSEGNET-GAN)for generating confrontation training,which makes full use of the ability of generating confrontation network to fit the joint probability distribution and adds additional constraints to the model Experiments show that MS-SEGNET can effectively improve the segmentation effect of the model without increasing any parameters.