Scene recognition method based on deep learning and privilege information

Abstract

The invention provides a scene recognition method based on deep learning and privilege information, and belongs to the technical field of image processing. The method comprises the following specificsteps: (1.1) selecting RGB images and depth images of a plurality of main scenes from a scene recognition library, and pairing the RGB images and the depth images with the RGB images; (1.2) constructing an end-to-end trainable deep neural network model combining privilege information and an attention mechanism; (1.3) training the deep neural network model, and avoiding data imbalance by using a weight redistribution mode during training; and (1.4) obtaining a scene classification result of the image. According to the invention, a mapping relation from an RGB image to a depth image and then tohigh-level semantic features of the depth image is established by taking image coding, feature decoding and image coding as a framework. The current situation of depth modal loss is effectively solved, and the effect of multi-modal image fusion is achieved under the condition that only RGB images are used.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a scene recognition method based on deep learning and privileged information. Background technique [0002] In the prior art, scene recognition, as one of the important branches in the field of computer vision, has been widely used in human-computer interaction, intelligent robot, intelligent video surveillance, automatic driving and other fields; and the prerequisite or prior knowledge of object detection. [0003] In recent years, with the introduction of large-scale data sets, scene recognition methods based on deep learning have developed rapidly, achieving better recognition results than traditional algorithms; at the same time, depth image information can provide valuable global layout for scene recognition information; the combination of RGB images and depth images will further improve the effect of scene recognition; however, its disadvantages are: depth images ...

AI Technical Summary

Problems solved by technology

[0003] In recent years, with the introduction of large-scale data sets, scene recognition methods based on deep learning have developed rapidly, achieving better recognition results than traditional algorithms; at the same time, depth image information can provide valuable global layout for scene recognition information; the combination of RGB images and depth images will further improve the effect of scene recognition; however, its disadvantages are: depth images are difficult to obtain and the amount of data is small

Images