Research on infrared image target detection technology based on deep learning
, , and
Mar 17, 2025
About this article
Published Online: Mar 17, 2025
Received: Oct 19, 2024
Accepted: Feb 03, 2025
DOI: https://doi.org/10.2478/amns-2025-0325
Keywords
© 2025 Jing Gao et al., published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.
Figure 1.
Figure 2.
Figure 3.
Figure 4.
Figure 5.
The comparison results of the lightweight algorithm model experiment
| Model | Parameter quantity(M) | Model size(MB) | Precision | Recall | mAP | FLOPs(G) |
|---|---|---|---|---|---|---|
| YOLOv3-tiny | 9.35 | 18.08 | 0.800 | 0.620 | 0.706 | 13.58 |
| YOLOv5 | 7.69 | 14.37 | 0.827 | 0.728 | 0.814 | 16.48 |
| YOLOv5-ghost | 4.36 | 8.58 | 0.832 | 0.728 | 0.811 | 8.78 |
| YOLOv5-repvgg | 7.69 | 15.48 | 0.845 | 0.778 | 0.839 | 16.48 |
| YOLOv7-tiny | 6.69 | 12.98 | 0.813 | 0.754 | 0.826 | 13.68 |
| Ours | 9.60 | 16.48 | 0.857 | 0.811 | 0.862 | 22.68 |
The parameter modulus model compares the results
| Model | Parameter quantity(M) | Model size(MB) | Precision | Recall | mAP | FLOPs(G) |
|---|---|---|---|---|---|---|
| YOLOv5m | 21.03 | 42.37 | 0.831 | 0.772 | 0.835 | 48.58 |
| YOLOv6s | 17.37 | 36.47 | 0.833 | 0.786 | 0.840 | 44.88 |
| YOLOv7 | 36.66 | 74.97 | 0.861 | 0.814 | 0.865 | 103.88 |
| YOLOv8s | 11.29 | 22.67 | 0.835 | 0.775 | 0.841 | 29.08 |
| Ours | 9.60 | 16.48 | 0.857 | 0.811 | 0.862 | 22.68 |
The model deployment scheme compares the results
| Index | Accuracy | Running time(ms) | Memory footprint(KB) | CPSA(%) | Parameter quantity |
|---|---|---|---|---|---|
| Scheme | |||||
| Torch | 50.43 | 44.946 | 2815645 | 80.45 | 45 |
| Darknet | 55.91 | 102.757 | 2795642 | 35.42 | 95 |
| TensorRT | 53.32 | 52.854 | 2809325 | 68.45 | 51 |
| Ours | 54.87 | 43.150 | 2584360 | 50.21 | 40 |