Human identification and classification network based on object hierarchy

Abstract

In the real world, objects are categorized hierarchically. However, deep learning (DL)-based computer vision (CV) approaches typically learn to classify objects in a flat manner, which limits their ability to generalize to new objects and scenarios. To address this limitation, a new network called human identification and classification network (HICNet) is proposed for hierarchical-based object identification and classification. HICNet is a hierarchical object identification and classification network with feature extraction layers, an identification branch, and a classification branch. HICNet was evaluated on the labeled faces in the wild (LFW) and labeled faces in the wild-A (LFWA) datasets, where it achieved an identification accuracy of 99.27% and an average classification accuracy of 98.61%. These results demonstrate that HICNet is able to classify objects beyond the training sample(s) by tracing the hierarchy and correlation of the object. HICNet can be used in a variety of CV applications, such as visually impaired aids and robot guidance as it provides a comprehensive insight into the objects.