Semantic segmentation method introducing feature cross attention mechanism

Abstract

The method aims at solving the problems that a Deeplabv3+ model is inaccurate in picture target edge segmentation, image feature fitting is slow, and attention information cannot be effectively utilized. A feature cross attention module is added to the model, and the cross attention network is composed of two branches and a feature cross attention module. The shallow branches are used for extracting low-level space information, and the deep branches are used for extracting high-level context features, so that important features are extracted more finely. According to the invention, the connection between the feature cross attention mechanism and the Deeplabv3+ coding module is designed and realized, and the output features of the Deeplabv3+ coding module are input into the feature cross attention module for convolution operation to realize re-calibration of the original features. A Deeplabv3+ decoding module acquires spatial features and channel features from two branches respectively,and then fuses the acquired features to acquire more important features. The improved model is verified through a Pascal Voc2012 data set, and results show that the model added with the feature crossattention mechanism can effectively improve the defects of an original model, can segment a target more finely, and better solves the problems of rough segmentation boundary and the like.

Description

technical field [0001] The invention belongs to the field of semantic segmentation, relates to a semantic segmentation model that introduces an attention mechanism, and specifically relates to introducing a double attention mechanism method into a Deeplabv3+ model. Background technique [0002] At present, the convolutional neural network has greatly promoted the execution of visual tasks with its rich representation capabilities, and image semantic segmentation is one of the key tasks to promote computer vision. As a classic computer vision problem (image classification, object recognition detection, semantic segmentation), image semantic segmentation has always been an important research direction. Its essence is to classify pixels in pictures. Image semantic segmentation is widely used in autonomous driving, object classification, cloud detection, medical detection and other related fields. In the image semantic segmentation method based on the PascalVoc2012 data set, th...

AI Technical Summary

Problems solved by technology

In the image semantic segmentation method based on the PascalVoc2012 data set, the currently popular models include FCN, U-Net, and Deeplab series. They all have less edge modification on the segmentation results, resulting in rough boundaries for some image segmentation and long-distance pixel categories. The relationship between the relationship can not be fully utilized and other issues

Images