Laser & Optoelectronics Progress, Volume. 61, Issue 24, 2437005(2024)
Underwater Object Detection Using a Multiscale and Cross-Spatial Information Aggregation Network
Figures & Tables(14)
Fig. 1. Structure of multiscale and cross-spatial information aggregation network
Fig. 8. Comparison of experimental results before and after introducing DPM module. (a) (b) Original images; (c) (d) before introducing the DPM; (e) (f) after introducing the DPM
Fig. 9. Comparison of detection results for different picture types. (a) Background blur; (b) different scale target; (c) target occlusion
Fig. 10. Comparison of detection results for different complex scenes. (a) Color deviation; (b) fog effect; (c) light interference
Table 1. Ablation experiment results on URPC dataset
View tableTable 1. Ablation experiment results on URPC dataset
Experiment number YOLOv7-tiny DPM Conv2former MAPE ASSF P /% R /% mAP@0.5 /% mAP@0.5∶0.95 /% FLOPs /G 1 √ 78.5 71.6 78.0 41.0 13.2 2 √ √ 78.2 72.6 78.4 41.3 10.5 3 √ √ 79.6 71.9 78.8 41.9 12.1 4 √ √ 79.6 72.1 79.4 42.3 13.5 5 √ √ √ 79.4 72.3 79.2 42.1 10.9 6 √ √ √ 80.3 73.1 80.3 42.5 12.4 7 √ √ √ √ 81.4 73.7 80.7 43.1 12.2 8 √ √ √ √ √ 82.1 74.2 81.5 44.3 12.7
Table 2. Comparision of experiment results for integrating different attention mechanisms into PS
View tableTable 2. Comparision of experiment results for integrating different attention mechanisms into PS
Model P /% R /% mAP@0.5 /% mAP@0.5∶0.95 /% YOLOv7-tiny 78.5 71.6 78.0 41.0 YOLOv7-tiny-PS 79.4 71.7 78.5 41.9 YOLOv7-tiny-PS-ECA 79.1 72.3 78.9 41.8 YOLOv7-tiny-PS-SIMAM 79.6 72.1 79.1 42.1 YOLOv7-tiny-PS-SE 79.4 73.1 78.5 41.4 YOLOv7-tiny-PS-CA 78.3 72.0 78.4 41.2 YOLOv7-tiny-PS-EMA （MAPE） 79.6 72.1 79.4 42.3
Table 3. Comparasion of results for different models
View tableTable 3. Comparasion of results for different models
Model P /% R /% mAP@0.5 /% Parameters /M FLOPs /G SSD 64.4 56.5 63.4 28.0 116.2 PANet 68.2 66.1 76.5 4.8 8.7 Cascade R-CNN 80.6 73.2 81.1 98.3 271.2 Underwater-YCC 80.2 71.9 80.3 — — RoIMix 75.2 69.3 77.3 24.4 24.9 YOLOv8 79.1 72.6 79.7 11.2 28.6 Proposed 82.1 74.2 81.5 6.2 12.7
Table 4. Comparasion of migration experiment results in RUOD dataset
View tableTable 4. Comparasion of migration experiment results in RUOD dataset
Model P /% R /% mAP@0.5 /% Parameters /M FLOPs /G Faster R-CNN 59.5 56.7 61.3 115.1 368.5 AWBiFPN 81.9 75.1 83.6 — — SegNext 67.1 63.4 68.9 38.0 15.9 PANet 70.7 65.3 72.4 4.8 8.7 RoIMix 68.0 62.4 69.2 24.4 24.9 YOLOv8 82.7 75.8 83.1 11.2 28.6 Proposed 85.3 79.9 86.1 6.2 12.7
Tools
Get Citation
Copy Citation Text
Jihai Yang, Xiaofang Pei. Underwater Object Detection Using a Multiscale and Cross-Spatial Information Aggregation Network[J]. Laser & Optoelectronics Progress, 2024, 61(24): 2437005
Paper Information
Category: Digital Image Processing
Received: Mar. 25, 2024
Accepted: May. 8, 2024
Published Online: Dec. 12, 2024
The Author Email: Xiaofang Pei (xiaofangpei@163.com)
CSTR:32186.14.LOP240958
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20