Image semantic segmentation method of guiding feature fusion based on attention mechanism

Abstract

The invention discloses an image semantic segmentation method of guiding feature fusion based on attention mechanism. The image semantic segmentation method comprises the following steps: (10) constructing an encoder basic network: using an improved ResNet-101 to generate a series of features changing from high resolution low semantics to low resolution high semantics; (20) constructing a decoderfeature fusion module: extracting high-level semantics with strong consistency constraint by adopting a pyramid structure module based on three-layer convolution operation, and performing weighted fusion on low-level stage features layer by layer to obtain a preliminary segmentation heat map; and (30) constructing an auxiliary loss function: adding auxiliary supervision to each fusion output in the decoding stage, superposing the auxiliary supervision with main supervision loss after the up-sampling of the heat map, and strengthening the hierarchical training of the model to obtain a semanticsegmentation map. The image semantic segmentation method of guiding feature fusion based on attention mechanism is high in accuracy and clear in boundary contour.

Description

technical field [0001] The invention belongs to the technical field of static image recognition, in particular to an image semantic segmentation method based on an attention mechanism guiding feature fusion with high accuracy and clear boundaries. Background technique [0002] Semantic segmentation, that is, pixel-level image understanding, is one of the important cornerstones in the field of computer vision and has a very wide range of application scenarios. Through fine-grained segmentation, it gives the machine the ability to separate different areas of the visual image at the pixel level. Semantic segmentation divides the pixel regions belonging to the same object in the image together, thereby expanding its application field. [0003] Semantic segmentation combines the two problems of object classification and target positioning while performing pixel-level prediction. How to strike a balance between the two mutually constrained problems of high-level abstract object c...

AI Technical Summary

Problems solved by technology

While semantic segmentation solves the classification problem, it also needs to face the spatial details such as the boundary contour of the positioning object in the segmentation. The simple pixel classification task often has the phenomenon that the boundary contour of the object in the segmentation result is blurred.

Images