Lightweight DB-YOLO Facemask Intelligent Detection and Android Application Based on Bidirectional Weighted Feature Fusion

• DOI: 10.3390/electronics12244936
• Publication Date: 2023-12-08T10:47:30Z
• Authors: Bin Qin; Ying Zeng; Xin Wang; Junmin Peng; Tao Li; Teng Wang; Yuxin Qin
• Keywords: Testbed

Conventional facemask detection algorithms face challenges of insufficient accuracy, large model size, and slow computation speed, limiting their deployment in real-world scenarios, especially on edge devices. Aiming at addressing these issues, we proposed a DB-YOLO intelligent algorithm, which is lightweight solution that leverages bidirectional weighted feature fusion. Our method built the YOLOv5 algorithm model, replacing original backbone network with ShuffleNetv2 to reduce parameters computational requirements. Additionally, integrated BiFPN as fusion layer, enhancing model’s capability for objects various scales. Furthermore, employed CARAFE upsampling factor improve perception details small-sized EIOU loss function expedite convergence. We validated effectiveness our through experiments conducted Pascal VOC2007+2012 Face_Mask datasets. experimental results demonstrate boasts compact size approximately 1.92 M. It achieves average precision values 70.1% 93.5% datasets, respectively, showcasing 2.3% improvement compared YOLOv5s. reduced by 85.8%. also successfully deployed Android devices using NCNN framework, achieving speed up 33 frames per second. Compared models like YOLOv5n, YOLOv4-Tiny, YOLOv3-Tiny, not only reduces but effectively improves exhibiting excellent practicality promotional value