Simultaneous target and attribute detection with an enhanced YOLO model

In the field of computer vision, target attribute recognition has become a critical research area with applications in security monitoring, scene understanding, and behavior analysis. Current methods typically employ separate models for target detection and attribute recognition, leading to inefficiencies in terms of resource utilization and model training. This paper proposes a novel framework for simultaneous target and attribute detection based on an enhanced YOLO architecture. The proposed approach integrates an attribute recognition module into the detection framework, enabling shared feature extraction and multi-scale feature fusion. This design leverages the network’s representational capacity and facilitates end-to- end training, thereby improving model efficiency and resource utilization. Tailored loss functions are designed, and the confusion matrix is adapted to evaluate model performance. Experimental comparisons with combined models show that the proposed model improves the F1 score for clothing attributes by 1.46 percentage points and for gender attributes by 1.45, thereby validating the effectiveness of the model architecture.