Infrared and Laser Engineering, Volume. 53, Issue 12, 20240286(2024)
Embedding spatial position information and Multi-view Feature Extraction for infrared small target detection
Figures & Tables(15)
Fig. 1. SpatialLocation Information and Multi-View Feature Fusion network model
Fig. 10. Detection results of Multi-View Feature Extraction on different datasets
Table 1. Ablation experiments
View tableView in ArticleTable 1. Ablation experiments
Models SLIF MVFE LSK AIFI GFLOPs Size/MB mAP75 mAP50-95 Inference time/ms YOLOv5s 15.8 13.7 82.8% 70.6% 6.9 Ours1 √ 16.0 14.3 86.3% 71.1% 6.6 Ours2 √ √ 16.6 14.4 87.0% 72.2% 6.9 Ours3 √ √ √ 17.5 15.2 88.8% 72.4% 7.1 ESLIMFENet √ √ √ √ 17.4 16.9 90.5% 74.5% 8.5
Table 2. Comparison experiments of different models
View tableView in ArticleTable 2. Comparison experiments of different models
Models GFLOPs Size/MB mAP75 mAP50-95 Inference time/ms YOLOv3-Tiny 12.9 16.6 76.0% 63.4 5.1 YOLOv5s-ghost 8.0 7.4 70.5% 58.4% 7.5 YOLOv5s 15.8 13.7 82.8% 70.6% 6.9 YOLOv5s-bifpn 108.5 89.0 87.0% 73.1% 20.3 YOLOv6 11.8 8.3 78.0% 65.2% 2.6 YOLOv7 103.2 71.3 88.5% 72.3% 9.8 YOLOR-CSP 118.9 100.6 86.2% 71.5% 11.8 YOLOv7-Tiny 13.0 11.7 86.7% 71.5% 4.8 YOLOv8n 8.1 6.0 86.9% 72.6% 2.1 YOLOv8s 28.4 21.5 87.4% 72.9% 4.2 ESPIMFENet 17.4 16.9 90.5% 74.5% 8.5
Table 3. Comparison between different attention
View tableView in ArticleTable 3. Comparison between different attention
YOLOv5s GFLOPs Size/MB mAP75 mAP50-95 Inference time/ms +SPPF 15.8 13.7 82.8% 70.6% 6.9 +SA 15.8 13.7 72.2% 63.2% 7.0 +CBAM 16.8 14.0 84.6% 69.8% 6.9 +C3 ghost 14.8 13.4 83.5% 70.8% 7.2 +CA 15.8 13.7 83.6% 68.6% 6.9 +MHSA 16.4 15.2 83.8% 70.2% 6.8 +SIIF 16.2 15.0 86.3% 71.1% 6.6 +MVFE 16.8 14.4 86.1% 71.4% 7.2 +LSK 16.1 13.9 86.3% 72.2% 7.0 +AIFI 15.7 15.5 87.5% 72.4% 7.9
Table 4. Experimental results of Multi-view Feature Extraction on different datasets
View tableView in ArticleTable 4. Experimental results of Multi-view Feature Extraction on different datasets
Models GFLOPs Size/MB mAP75 mAP50-95 Inference time/ms YOLOv5s (SIRST) 15.8 13.6 77.6% 64.5% 6.9 YOLOv5s (NUDT-SIRST) 15.8 13.7 78.8% 68.3% 6.9 Ours1 (SIRST) 16.0 14.3 86.3% 71.1% 7.0 Ours1 (NUDT-SIRST) 16.0 14.3 87.9% 72.2% 7.0
Tools
Get Citation
Copy Citation Text
Zifen HE, Jinsheng XUE, Yinhui ZHANG, Guangchen CHEN. Embedding spatial position information and Multi-view Feature Extraction for infrared small target detection[J]. Infrared and Laser Engineering, 2024, 53(12): 20240286
Paper Information
Category: 图像处理
Received: Sep. 22, 2024
Accepted: --
Published Online: Jan. 16, 2025
The Author Email:
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20