Attention mechanism feature fusion segmentation method for image

Abstract

The invention discloses an attention mechanism feature fusion segmentation method for an image. The method comprises the following steps: uniformly cutting an input image; carrying out pooling operation on the output feature map; extracting four different convolution stages in the network ResNet101 from the features; aggregating multi-scale context information of the output feature map through spatial pyramid pooling; performing double up-sampling on the output feature map through bilinear interpolation, and performing feature fusion with the output feature map through a channel attention mechanism; performing double up-sampling on the output feature map through bilinear interpolation, performing feature fusion on the output feature map through a channel attention mechanism, and using as an output feature map of a channel attention mechanism feature fusion module; and carrying out quadruple up-sampling on the feature image by using bilinear interpolation, restoring the feature image to an input image resolution, and generating a final prediction result. According to the method, a clearer target segmentation boundary and a fine segmentation result can be formed, detail information in the low-level feature map is fully utilized, and the semantic segmentation precision of the image is improved.

Description

technical field [0001] The invention belongs to the technical field of image semantic segmentation, and in particular relates to an image attention mechanism feature fusion segmentation method. Background technique [0002] With the continuous progress of society, people expect computers to have the same logical reasoning ability and decision-making ability as humans, and rescue people from various complex tasks. Image semantic segmentation divides the image into regions representing different semantic information by assigning a predefined semantic label to the pixels in the image. The image after image semantic segmentation can be used in image semantic recognition, target tracking, and other scene understanding tasks, and it is an important means of image processing. At present, image semantic segmentation technology has produced many important applications in the fields of medical imaging, automatic driving, smart home and image engine search. [0003] Image semantic se...

AI Technical Summary

Problems solved by technology

The encoder of the DeepLabv3+ semantic segmentation network effectively extracts high-level features, but the decoder directly fuses a single low-level feature map with a high-level feature map in the feature extraction network. This feature fusion method is too simple to effectively restore the detailed information of the image, resulting in In the segmentation result, the positioning of the target edge pixels is not accurate, and there are problems of missed segmentation and wrong segmentation

Images