SE-ResNet network-based race identification method and device and computer storage medium

Abstract

The invention discloses a race recognition method based on an SE-ResNet network. The race recognition method comprises the following steps: acquiring real race image data as original data, detecting a face, and performing rotary filling, correction and zooming to a uniform size; removing non-feature images of large-area shielding of side face, head lowering and five-sense-organ areas of the human face; performing diversity enhancement on the image data; carrying out value reduction processing on RGB channels of the image data, and then carrying out classified labeling; adding an SE residual module on the basis of ResNet 50 to establish an SE-ResNet network, and training the SE-ResNet network; and selecting a to-be-identified picture, inputting the picture into the SE-ResNet network obtained by training, and carrying out classification and identification to obtain a result. According to the method, the collected image data is detected and processed, and the SE-ResNet network is trained after the data without obvious characteristics is proposed, so that the method has relatively high recognition speed and accuracy.

Description

technical field [0001] The invention relates to a race identification method, in particular to a SE-ResNet network-based race recognition method, device and computer storage medium. Background technique [0002] Most of the world's ethnic identification is recognized as the four major races. In personnel management, due to the different characteristics of different nations and countries, it is necessary to respect their own living and working habits, so face recognition is required. The existing technology focuses on distinguishing the four major races. For example, the two patents with application numbers of 202010996916.2 and 201811372085.0 solve the problems that are all ready-made face data sets. Since they contain most of the European and American race data, the recognition results of other races are not good. good question. However, there are many types within a race, and the existing recognition methods are not very accurate in identifying this specific subtype. Co...

AI Technical Summary

Problems solved by technology

[0003] Aiming at the defects of the above-mentioned prior art, the present invention provides a race identification method based on the SE-ResNet network, which solves the problem that the accuracy of classification and recognition of a single subtype among the four major races is not high

Images