Real-time semantic segmentation method based on multi-scale segmentation fusion

Abstract

The invention provides a real-time semantic segmentation method based on multi-scale segmentation fusion, and relates to the technical field of machine vision, and the method comprises the steps: building and training a self-adaptive multi-scale segmentation fusion network model, wherein the adaptive multi-scale segmentation fusion network comprises a backbone network, a classification layer, an alignment module and a fusion module; sending the to-be-processed image into the backbone network for feature extraction, and outputting feature maps of multiple scales; performing pixel-level classification on the output feature maps of each scale by using a classification layer to obtain segmented maps of different scales; unifying the segmented images of different scales to the same resolution by using an alignment module to obtain segmented images of the same size; sending the segmented images with the same size into a fusion module, and fusing segmented context information of different levels based on a specific target category to obtain a fused segmented image; and carrying out primary refinement on the fused segmentation image by using convolution operation to obtain a final segmentation result, so that efficient and rapid real-time image semantic segmentation is realized.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a real-time semantic segmentation method based on multi-scale segmentation fusion. Background technique [0002] Image semantic segmentation is one of the basic tasks in the field of computer vision. In recent years, it has been widely developed due to the expansion of deep learning. However, the existing high-precision methods are based on deep network design and complex feature reuse, which are difficult to achieve the purpose of real-time application. Real-time semantic segmentation is expected to achieve excellent performance in both speed and performance, and realize the application in real-time scenarios. [0003] At present, the real-time image semantic segmentation method mainly achieves a fast segmentation framework by designing a lightweight backbone network and simplifying the structure of the decoder to reduce the complexity of the model. These methods expec...

AI Technical Summary

Problems solved by technology

[0004] However, it is difficult to restore the spatial detail information lost in the downsampling process, so these methods often cannot obtain accurate segmentation results.

On the other hand, excessive downsampling has become a common paradigm to improve the reasoning speed of real-time semantic segmentation algorithms, but this approach makes the resolution of the final feature map too low, which further increases the difficulty of the model to restore spatial information

Images