Research On Image Semantic Segmentation Based On Deep Convolutional Neural Network

Deep convolutional neural networks are a very effective method in the field of computer vision.Massively growing image data and increasingly popular smart devices require fast and accurate understanding the content of the image.In addition,the target object in the image is required to be automatically segmented and identified.The task of image segmentation is to detect whether a certain type of target object is included in a given image,and to mark the object class of which each pixel in the image belongs in,to draw the boundary of each object,and finally to obtain a pixel semantic annotation.The detection and segmentation of target objects in image is a great significance for the development of computer vision,it also has high practical value in practical engineering application.At present,many excellent research results about image segmentation algorithm have been published at home and abroad,but when applied to the actual operation process,it is found that there are still many problems.For example,under the influence of factors such as partial object overlap occlusion,illumination intensity and background interference,it is difficult to obtain an ideal segmentation accuracy.In this paper,the current existing algorithms are deeply researched and analyzed.According to the analysis results,I propose an algorithm that can improve the segmentation accuracy of target objects and has high adaptability.The main research contents of this paper are briefly described below:(1)Based on the deep neural network semantic segmentation method,the improvement is proposed based on the Deeplabv3-plus network model.The spatial cavity pyramid pooling module which based on the coprime factor is used to mitigate the grid effect and reduce the influence of image space structure information loss on the semantic segmentation precision,thus improving the segmentation accuracy.(2)The effects of sample imbalance on machine learning performance are studied in detail,and the main difficulties and problems that need to be solved are analyzed and summarized.An image semantic segmentation algorithm based on local enhancement is proposed,which improves the semantic segmentation accuracy.