Multi-Scale YOLOv5-AFAM Based Infrared Dim Small Target Detection DOI Open Access
Yuexing Wang, Zhao Liu, Yixiang Ma

et al.

Published: June 5, 2023

Infrared detection plays an important role in the military, aerospace, and other fields, which has advantages of all-weather, high stealth, strong anti-interference. However, infrared dim small target suffers from complex backgrounds, low signal-to-noise ratio, blurred targets with area percentages, challenges. In this paper, we proposed a multiscale YOLOv5-AFAM algorithm to realize high-accuracy real-time detection. Aiming at problem intra-class feature difference inter-class similarity, Adaptive Fusion Attention Module - AFAM was generate maps that are calculated weigh features network make focus on targets. This paper fusion structure solve variable scales vehicle addition, downsampling layer is improved by combining Maxpool convolutional reduce number model parameters retain texture information. For multiple scenarios, constructed dataset, ISVD. The conducted ISVD compared YOLOv7, [email protected] achieves improvement while only 17.98% it. By contrast YOLOv5s model, 4.3% 6.6% reduction parameters. Experiments results demonstrate higher accuracy speed vehicles.

Language: Английский

Multiscale YOLOv5-AFAM-Based Infrared Dim-Small-Target Detection DOI Creative Commons
Yuexing Wang, Zhao Liu, Yixiang Ma

et al.

Applied Sciences, Journal Year: 2023, Volume and Issue: 13(13), P. 7779 - 7779

Published: June 30, 2023

Infrared detection plays an important role in the military, aerospace, and other fields, which has advantages of all-weather, high stealth, strong anti-interference. However, infrared dim-small-target suffers from complex backgrounds, low signal-to-noise ratio, blurred targets with small area percentages, challenges. In this paper, we proposed a multiscale YOLOv5-AFAM algorithm to realize high-accuracy real-time detection. Aiming at problem target intra-class feature difference inter-class similarity, Adaptive Fusion Attention Module (AFAM) was generate maps that are calculated weigh features network make focus on targets. This paper fusion structure solve variable scales vehicle addition, downsampling layer is improved by combining Maxpool convolutional reduce number model parameters retain texture information. For multiple scenarios, constructed dim dataset, ISVD. The conducted ISVD dataset. Compared YOLOv7, [email protected] achieves improvement while only 17.98% it. contrast, YOLOv5s model, 81.4% 85.7% parameter reduction 7.0 M 6.6 M. experimental results demonstrate higher accuracy speed vehicles.

Language: Английский

Citations

4
Multi-Scale YOLOv5-AFAM Based Infrared Dim Small Target Detection DOI Open Access
Yuexing Wang, Zhao Liu, Yixiang Ma

et al.

Published: June 5, 2023

Infrared detection plays an important role in the military, aerospace, and other fields, which has advantages of all-weather, high stealth, strong anti-interference. However, infrared dim small target suffers from complex backgrounds, low signal-to-noise ratio, blurred targets with area percentages, challenges. In this paper, we proposed a multiscale YOLOv5-AFAM algorithm to realize high-accuracy real-time detection. Aiming at problem intra-class feature difference inter-class similarity, Adaptive Fusion Attention Module - AFAM was generate maps that are calculated weigh features network make focus on targets. This paper fusion structure solve variable scales vehicle addition, downsampling layer is improved by combining Maxpool convolutional reduce number model parameters retain texture information. For multiple scenarios, constructed dataset, ISVD. The conducted ISVD compared YOLOv7, [email protected] achieves improvement while only 17.98% it. By contrast YOLOv5s model, 4.3% 6.6% reduction parameters. Experiments results demonstrate higher accuracy speed vehicles.

Language: Английский

Citations

3