Research And Application Of Image Semantic Segmentation Based On Encoder-decoder

With the development of the Internet of Things and artificial intelligence technology and its wide application in various fields,massive image data has been generated.How to quickly and accurately understand the image content and automatically segment the target area of the image according to the application scene has become a researcher.The focus of attention.In recent years,image segmentation methods based on deep learning have been increasingly developed.The main task is to detect whether a certain type of target object is included in a given image,and to mark the object class to which each pixel belongs in the image.The boundary of the object finally results in a segmentation graph with pixel semantic annotation.At present,image segmentation has been widely used in autonomous driving and robot engineering,and has become a major research task in computer vision.In this paper,the research results of image segmentation related algorithms at home and abroad are deeply studied and compared.The existing methods have low segmentation precision,low computational efficiency and large storage overhead.Focusing on the problem of low segmentation accuracy,this paper proposes a DeepLab-IRCNet image semantic segmentation algorithm based on encoder-decoder..The main research contents include:(1)By studying the image semantic segmentation algorithm based on encoder-decoder,this paper introduces the feature graph segmentation module based on the Deeplabv3+ model to improve the model's attention to small target objects and improve the segmentation accuracy.(2)For the context information that fails to fully consider the image,the spatial pyramid pooling technology is used to capture the image context information,and the utilization of local features and global features is unbalanced,and the characteristics of each intermediate layer at multiple scales are performed.Fusion improves the segmentation results.(3)For the unmanned field,using the CamVid dataset,an image semantic segmentation system is designed to verify the effectiveness of the proposed algorithm.