Laser & Optoelectronics Progress, Volume. 59, Issue 24, 2410003(2022)
Improved YOLOv3 Flame Detection Algorithm Based on Dynamic Shape Feature Extraction and Enhancement
Figures & Tables(8)
Fig. 3. Comparison between 3×3 standard convolution and deformable convolution. (a) Standard convolution; (b)-(d) deformable convolution
Fig. 4. Diagram of standard convolution fixed receptive field and deformable convolution adaptive receptive field
Fig. 6. Comparison of detection effects of different algorithms. (a) Proposed algorithm; (b) YOLOv4; (c) Cascade R-CNN
Table 1. Ablation experiment
View tableTable 1. Ablation experiment
Model YOLOv3 and its improvement AP /% Speed /（frame·s-1） Model volume /MB A DarkNet53 89.06 70.41 246 B ResNet50_vd-DCN（stage 5） 90.30 83.19 181 C ResNet50_vd-DCN（stage 4） 90.10 81.52 181 D ResNet50_vd-DCN（stage 4，stage 5） 93.81 77.38 182 E D+Drop Block 94.02 77.38 182 F E+IoU Aware（ours） 94.11 73.52 184
Table 2. Comparison of different target detection algorithms
View tableTable 2. Comparison of different target detection algorithms
Model Algorithm AP /% Speed /（frame·s-1） Model volume /MB 1 Faster R-CNN 90.43 5.28 301 2 Cascade R-CNN 90.87 12.96 355 3 YOLOv3 89.06 70.41 246 4 YOLOv3_IoULoss 89.12 70.55 247 5 YOLOv4 90.84 60.41 266 6 RetinaNet_R50 90.10 61.92 229 7 Propose algorithm 94.11 73.52 184
Tools
Get Citation
Copy Citation Text
Hao Ding, Huiqin Wang, Ke Wang. Improved YOLOv3 Flame Detection Algorithm Based on Dynamic Shape Feature Extraction and Enhancement[J]. Laser & Optoelectronics Progress, 2022, 59(24): 2410003
Paper Information
Category: Image Processing
Received: Sep. 9, 2021
Accepted: Oct. 27, 2021
Published Online: Oct. 31, 2022
The Author Email: Wang Huiqin (hqwang@xauat.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20