Detection and Recognition Method of Fast Low-Altitude Unmanned Aerial Vehicle Based on Dual Channel

Qi Ma; Bin Zhu; Zhengdong Cheng; Yang Zhang

doi:10.3788/AOS201939.1210002

Acta Optica Sinica, Volume. 39, Issue 12, 1210002(2019)

Detection and Recognition Method of Fast Low-Altitude Unmanned Aerial Vehicle Based on Dual Channel

Qi Ma^*, Bin Zhu^**, Zhengdong Cheng, and Yang Zhang

State Key Laboratory of Pulsed Power Laser Technology, College of Electronic Engineering, National University of Defense Technology, Hefei, Anhui 230037, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(15)

Fig. 1. Schematic of residual block structure

Download full size

Fig. 2. Residual block structure and convolutional layer structure. (a) Residual block structure; (b) convolutional layer structure

Download full size

Fig. 3. Feature fusion idea of multi-scale prediction network

Download full size

Fig. 4. Schematic of prediction module structure

Download full size

Fig. 5. Overall design scheme of proposed method

Download full size

Fig. 6. Optical axis acquisition system for infrared and visible light

Download full size

Fig. 7. Types of UAVs. (a) Dji-MAVIC Pro; (b) Dji-PHANTOM 3; (c) Dji-S900

Download full size

Fig. 8. Network structure diagram of Dual-YOLOv3-S

Download full size

Fig. 9. Network structure diagram of Dual-YOLOv3-M

Download full size

Fig. 10. Network structure diagram of Dual-YOLOv3-D

Download full size

Fig. 11. Different feature fusion modes. (a) Average fusion-1; (b) concatenated fusion-2; (c) concatenated fusion-3

Download full size

Fig. 12. Anchor boxes schematic of 13×13 scale

Download full size

Fig. 13. Comparison of partial detection results of YOLOv3 and Dual-YOLOv3 on test dataset (first category is Dji-PHANTOM 3, second category is Dji-MAVIC Pro, and third category is Dji-S900). (a) YOLOv3 (visible dataset); (b) YOLOv3 (infrared dataset); (c) Dual-YOLOv3-S; (d) Dual-YOLOv3-M; (e) Dual-YOLOv3-D

Download full size

Table 1. Evaluation results of different methods on test dataset

View table

Table 1. Evaluation results of different methods on test dataset

Method	Image source	Detection speed /s^-1	mAP /%	AP
Method	Image source	Detection speed /s^-1	mAP /%	DJ-3	DJ-Pro	DJ-S900
Faster R-CNN	Visible	0.429	86.79	85.23	87.12	88.01
Faster R-CNN	Infrared	0.422	86.77	85.12	86.99	88.21
SSD300	Visible	22.42	81.51	80.21	81.32	82.99
SSD300	Infrared	22.54	81.43	80.08	81.34	82.87
YOLOv3	Visible	30.12	92.77	92.58	91.50	94.23
YOLOv3	Infrared	30.23	92.78	92.41	91.77	94.17
Dual-YOLOv3-S-1	Visible+infrared	27.94	95.41	96.89	93.21	96.14
Dual-YOLOv3-M-1	Visible+infrared	27.59	97.09	98.12	95.94	97.21
Dual-YOLOv3-D-1	Visible+infrared	27.17	98.85	99.83	97.85	98.88

Table 2. Evaluation results of different fusion modes for Dual-YOLOv3-D on test dataset
View table
Table 2. Evaluation results of different fusion modes for Dual-YOLOv3-D on test dataset
Method Detection speed /s^-1 mAP /% AP
DJ-3 DJ-Pro DJ-S900
Dual-YOLOv3-D-1 27.17 98.85 99.83 97.85 98.88
Dual-YOLOv3-D-2 24.31 95.19 96.41 94.02 95.13
Dual-YOLOv3-D-3 24.47 95.21 96.34 94.24 95.04

Tools

Get Citation

Copy Citation Text

Qi Ma, Bin Zhu, Zhengdong Cheng, Yang Zhang. Detection and Recognition Method of Fast Low-Altitude Unmanned Aerial Vehicle Based on Dual Channel[J]. Acta Optica Sinica, 2019, 39(12): 1210002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites