Takeout person identity recognition method based on fusion information

Abstract

The invention discloses a takeout person identity recognition method based on fusion information and provides a multi-target detection network MFC Net based on multi-level features and a spatial attention mechanism. The network structure comprises four parts, namely a feature extraction backbone network module, a feature fusion module, a multi-scale information extraction module and a spatial attention mechanism module. For training of an MFC Net network, firstly, a PASCAL VOC data set is used for pre-training, and a pre-training model with basic target recognition capability is obtained; thena plurality of attributive characters of the takeout person are collected and marked to form a data set; and finally, on the basis of the pre-training model, carrying out further training to obtain afinal MFC Net recognition model so as to carry out test detection. The method provided by the invention not only has better robustness and self-adaptive capability. According to the method, the overall characteristics of the target are collected, the identity of the target can be accurately given through recognition of multiple parts of additional target information, and the applicability is high.

Description

technical field [0001] The invention in this paper relates to a method for identifying the identity of a delivery person, specifically, a method for identifying the identity of a delivery person based on fusion information, which belongs to the technical field of target identification. Background technique [0002] Object recognition using deep learning has been widely used. The identification of personnel identity based on target recognition technology is usually judged by extracting the overall characteristics of the target. This approach still has problems such as missing data sets, low judgment accuracy, and poor robustness. The method of using multiple target information to comprehensively judge the identity of a person is reasonable to a certain extent. It not only collects the overall characteristics of the target, but also can give the target identity more accurately through the identification of multiple additional target information, and has strong applicability. ...

AI Technical Summary

Problems solved by technology

First of all, by fusing the shallow features of the network and the feature information of the middle layer, a fusion feature containing context information is formed, which is used to solve the problem that the semantic information of small objects in the shallow feature map is not rich and the classification ability is weak; then, the atrous convolution is used to increase With the characteristics of large receptive field and no increase in parameters, multi-scale decomposition of fusion features is performed to satisfy the network's positioning and detection of multi-scale targets; finally, the spatial attention module enhances the contribution to key information by fusing global pixel-related spatial attention. features to enhance the network's ability to distinguish between targets and backgrounds

Images