A Semantic Segmentation Method and a Road Waterlogging Detection Method and Device Applying It

Abstract

This application proposes a semantic segmentation method and road water detection method and device, the semantic segmentation method includes: inputting the image to be processed into the semantic segmentation model, the semantic segmentation model includes a backbone network, a multi-scale feature extraction network and a semantic segmentation prediction head In the part, the multi-scale feature extraction network includes multiple global feature extraction networks connected in parallel, wherein each global feature extraction network includes at least one or more global feature extraction modules connected in series, and the global feature extraction module is embedded with multiple stacked global The CNN module of the attention module makes the global feature extraction module not only have the nature of convolution, but also perform global modeling. The road ponding detection method uses the above semantic segmentation method to quickly obtain the semantic segmentation result of the image to be processed, and identify whether there is road ponding, so as to realize the efficient management of road ponding in urban road management.

Description

technical field [0001] The present application relates to the field of computer vision, and in particular, to a semantic segmentation method and a road water detection method and device using the same. Background technique [0002] In the field of computer vision, semantic segmentation technology is to recognize images from the pixel level in the way of human perception. Compared with other image recognition technology, it is a process of linking each pixel in the image to the corresponding class label , so semantic segmentation can be regarded as pixel-level image classification. [0003] The Transformer model abandons the convolutional neural networks and recurrent neural networks used in previous deep learning tasks. It is a model that uses the attention mechanism to improve the speed of model training. The Transformer model is applied to the field of computer vision, which is Vision Transformer ( Vision Transformer), which uses a multi-head self-attention mechanism to e...

AI Technical Summary

Problems solved by technology

[0005] Moreover, the convolutional neural network is modeled locally in space. The convolutional neural network can extract local features, such as edges and corners, by calculating the connection between local adjacent pixels, and can provide rich local features in the shallow layer ( localfeature), but it is difficult to establish long-distance dependencies in deep CNN

Images