Proceedings of the
The Nineteenth International Conference on Computational Intelligence and Security (CIS 2023)
December 1 – 4, 2023, Haikou, China
Efficient Mask Recognition Using Modified Faster R-CNN with Depth-wise Separable Convolutions
1School of Information and Engineering,Hohai University, JiangSu Changzhou, China.
2School of Mechanical and Electrical Engineering, Hohai University, JiangSu Changzhou, China.
ABSTRACT
Airborne transmission plays a significant role in many infectious diseases, making mask-wearing in public places a crucial preventive measure. This necessitates the development of efficient mask-wearing detection technologies. Faster R-CNN is a well-established algorithm for object detection; however, it still faces challenges like high computational complexity and cumbersome candidate box extraction. In this study, we propose enhancements to the Faster R-CNN algorithm by employing a modified architecture based on the ZFNet network for feature extraction. To reduce computational complexity, we replace conventional convolution operations with depth-wise separable convolutions. Experiments show that the improved method achieves efficient mask recognition.
Keywords: Faster R-CNN, Mask recognition, ZF-Net network, Separable convolution, Feature extraction, Anchor box.
Download PDF
1School of Information and Engineering,Hohai University, JiangSu Changzhou, China.
2School of Mechanical and Electrical Engineering, Hohai University, JiangSu Changzhou, China.
ABSTRACT
Airborne transmission plays a significant role in many infectious diseases, making mask-wearing in public places a crucial preventive measure. This necessitates the development of efficient mask-wearing detection technologies. Faster R-CNN is a well-established algorithm for object detection; however, it still faces challenges like high computational complexity and cumbersome candidate box extraction. In this study, we propose enhancements to the Faster R-CNN algorithm by employing a modified architecture based on the ZFNet network for feature extraction. To reduce computational complexity, we replace conventional convolution operations with depth-wise separable convolutions. Experiments show that the improved method achieves efficient mask recognition.
Keywords: Faster R-CNN, Mask recognition, ZF-Net network, Separable convolution, Feature extraction, Anchor box.
Download PDF