Research On Image Semantic Segmentation Based On Deep Convolutional Neural Network

With the advent of the big data era and the continuous improvement of the computer processing performance of massive data,use deep learning methodsto solve the problem of computer vision has become a research hotspot.Image semantic segmentation as an important research content in computer vision tasks,has been widely used in various fields.However,the existing image semantic segmentation method requires large-scale pixellevel image annotation,and the single segmentation scale model has the problem of low segmentation accuracy.In this paper,an image semantic segmentation method is deeply studied based on deep convolutional neural network,and applied to the actual scene tasks such as automatic driving and terrain survey.A semi-supervised image semantic segmentation method based on deep convolution generator adversarial network is presented.By introducing adversarial ideas into the model,combined deep convolutional neural networks and generative adversarial networks,a semi-supervised semantic segmentation model is built to solve the problem that timeconsuming in pixel-level annotation of the images in the full-supervised model.Finally,the semantic segmentation method is applied to the semantic segmentation task of street scenes.The feasibility of the method is proved by experiments,and the influence of different proportions of labeled samples on the performance of the model is compared and analyzed.A multi-scale fusion image semantic segmentation method based on deep convolutional neural network is proposed.Based on the deep convolutional neural network,a more robust multi-scale feature extraction method is built,and the attention mechanism is introduced in the multi-scale feature fusion stage,the different weights are assigned to different scales.Compared with the single-scale segmentation method,this method solves the problem that the large different scale of different objects in the same scene,and further improves the image semantic segmentation precision.Finally,the method is applied to aerial image segmentation tasks,and proved the effectiveness of the method by comparison experiments.