Semantic edge detection method based on deep learning

Abstract

The invention discloses a semantic edge detection method based on deep learning. The core is a semantic edge detection deep neural network model based on multi-level feature fusion. The method comprises the steps of multi-level feature learning, multi-level feature extraction and multi-level feature fusion. The training comprises three steps of initializing model parameters, preparing a target data set and training an overall model. Compared with the prior art, the method has the following advantages: 1) the low-level features are gradually fused into the high-level features from bottom to top, the semantic discrimination capability of the fused high-level features is enhanced, and the missing underlying detail information is supplemented, so that the semantic edge detection effect is improved (2) training a multi-level feature learning module on the big data set in advance to enable the big data set to learn richer multi-level features on a target data set, and (3) expanding the dataset by adopting a data enhancement technology and preprocessing true value tagse to enhance the learning ability of the whole model.

Description

technical field [0001] The invention belongs to the technical field of image processing and computer vision, and relates to a semantic edge detection method based on deep learning. Background technique [0002] Image semantic edge extraction is to detect the contours of objects in images and determine the type of objects to which the contours belong. It is an important subject in computer vision research. Image semantic edge extraction can also be beneficial for other vision tasks, such as image segmentation, depth inference, occlusion inference, object detection, 3D reconstruction, etc. However, due to the influence of lighting, noise and other issues, image semantic edge detection is extremely challenging. [0003] In recent years, with the emergence of convolutional neural network, its powerful multi-level feature expression ability has effectively promoted the development of many computer vision tasks, such as image semantic segmentation, human pose estimation, object d...

AI Technical Summary

Problems solved by technology

However, the feature fusion of different levels of this method is not sufficient, the feature expression ability is limited, the extracted edges are discontinuous and not smooth, and the edge classification accuracy is limited

Images