Target Detection Algorithm Based on Improved YOLO v3

Qiong Zhao; Baoqing Li; Tangwei Li

doi:10.3788/LOP57.121502

Laser & Optoelectronics Progress, Volume. 57, Issue 12, 121502(2020)

Target Detection Algorithm Based on Improved YOLO v3

Qiong Zhao^1,2, Baoqing Li^1、*, and Tangwei Li^1,2

¹Key Laboratory of Science and Technology on Microsystem, Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences, Shanghai 201800, China

²University of Chinese Academy of Sciences, Beijing 100049, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(10)

Fig. 1. Residual layer structure

Download full size

Fig. 2. Darknet53 structural parameters

Download full size

Fig. 3. Improved network structure

Download full size

Fig. 4. Network structure parameters of increase module

Download full size

Fig. 5. Loss function training curve

Download full size

Fig. 6. Model diagram of different schemes. (a) 1^st kind; (b) 2^nd kind

Download full size

Fig. 7. Experimental renderings on test dataset

Download full size

Table 1. Test results of various algorithms on VOC2007 dataset

View table

Table 1. Test results of various algorithms on VOC2007 dataset

Algorithm	Network	Data	mAP /%
Faster R-CNN	VGG	VOC2007+2012	73.2
Faster R-CNN	Residual-101	VOC2007+2012	76.4
R-FCN	Residual-101	VOC2007+2012	80.5
SSD321	Residual-101	VOC2007+2012	77.1
SSD500	Residual-101	VOC2007+2012	80.6
YOLO v2_416	Darknet19	VOC2007+2012	76.8
YOLO v3_416	Darknet53	VOC2007+2012	79.4
Ours	Darknet53	VOC2007+2012	80.2

Table 2. Impact of data enhancement on model accuracy
View table
Table 2. Impact of data enhancement on model accuracy
Dataset mAP /%
Before After
VOC2007 80.20 80.47

Table 3. Influence of different models on model training time
View table
Table 3. Influence of different models on model training time
Model Solution one Solution two
Time /h 142 134