Bird Detection Algorithm in Natural Scenes Based on Improved YOLOv3

Ziying Song; Kuihe Yang; Yu Zhang

doi:10.3788/LOP202259.1815013

Journals >Laser & Optoelectronics Progress >Volume 59 >Issue 18 >Page 1815013 > Article

Laser & Optoelectronics Progress
Vol. 59, Issue 18, 1815013 (2022)

Bird Detection Algorithm in Natural Scenes Based on Improved YOLOv3

Ziying Song^1、2、*, Kuihe Yang², and Yu Zhang²

Author Affiliations

¹School of Computer and Information Technology, Beijing Jiaotong University, Beijing 100044, China

²School of Information Science and Engineering, Hebei University of Science and Technology, Shijiazhuang ;050018, Hebei , China

show less

DOI: 10.3788/LOP202259.1815013 Cite this Article Set citation alerts

Ziying Song, Kuihe Yang, Yu Zhang. Bird Detection Algorithm in Natural Scenes Based on Improved YOLOv3[J]. Laser & Optoelectronics Progress, 2022, 59(18): 1815013 Copy Citation Text

show less

Fig. 1. Depthwise separable convolution process

Download full size

Fig. 2. Schematic of YOLOBIRDS structure

Download full size

Fig. 3. Influence of positive and negative sample imbalance in natural scene

Download full size

Fig. 4. Statistical results of bird objects. (a) Statistic of number of birds; (b) bird location; (c) bird box size

Download full size

Fig. 5. Comparison of experimental results of different algorithms. (a) Images seriously affected by light; (b) images affected by fog; (c) images taken normally

Download full size

Bird class	SSD300	YOLOv3	Faster RCNN	YOLOBIRDS
baiqueling0	71.85	87.78	88.54	91.15
bailu1	73.78	87.65	90.12	88.57
haiou2	87.82	88.97	86.32	82.12
heizuiou3	69.49	84.32	83.54	92.85
huiqiongniao4	90.56	91.12	92.39	91.32
luzi5	63.89	87.75	89.43	86.48
shanmaque6	61.43	78.57	82.45	79.58
xiaopiti7	79.33	85.67	88.23	91.82
zhuomuniao8	60.64	75.34	83.44	87.67
dae9	53.51	76.93	79.14	79.64

Table 1. AP of different algorithms for different birds

Method	mAP /%	Speed /（frame·s^-1）	Access inventory /MB	Number of parameters	FLOPs /10⁹
SSD300	71.23	45.64	107.51
YOLOv3	84.41	27.23	254.03	61760674	60.68
Faster RCNN	86.36	12.52	588.59
YOLOBIRDS	87.12	32.67	132.72	12425917	24.22

Table 2. Overall model performance comparison of different algorithms

Method

mAP /%

Speed /（frame·s^-1）

YOLOv3（origin，darknet53）

YOLOv3（VGG19）

YOLOv3（Resnet50）

YOLOv3（Inceptionv4）

YOLOv3（DenseNet）

YOLOv3（SENet）

YOLOv3（DualPathNet）

YOLOBIRDS

84.41

78.69

79.58

82.89

84.51

82.54

86.46

87.12

27.23

38.15

24.59

26.25

27.97

30.18

31.65

32.67

Table 3. Comparison between YOLOBIRDS and YOLOv3 series

Method

R_IoU=0.5

R_IoU =0.6

R_IoU＝0.7

SSD300

YOLOv3

Faster RCNN

YOLOBIRDS

71.23

84.41

86.36

87.12

62.39

72.33

77.38

76.27

55.72

67.32

69.33

70.34

Table 4. Comparison of mAP of different algorithms under different IoU thresholds

Method	Precision /%			Recall /%			F1_Score /%
Method	0.25	0.45	0.65	0.25	0.45	0.65	0.25	0.45	0.65
SSD300	69.86	72.76	81.83	65.09	73.28	83.29	67.39	73.02	82.55
YOLOv3	62.47	74.87	82.28	69.12	76.27	84.91	65.63	75.56	83.57
Faster RCNN	63.39	78.63	83.09	66.93	78.92	86.53	65.11	78.77	84.78
YOLOBIRDS	66.34	79.37	85.13	69.42	78.20	85.91	67.85	78.78	85.52

Table 5. Index parameters of different algorithms under different confidence thresholds

Ziying Song, Kuihe Yang, Yu Zhang. Bird Detection Algorithm in Natural Scenes Based on Improved YOLOv3[J]. Laser & Optoelectronics Progress, 2022, 59(18): 1815013

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information