Target Detection Based on Faster Region Convolution Neural Network

Fig. 5. Experimental results of different algorithms. (a) Average detection accuracy; (b) detection speed; (c) accuracy comparison result; (d) speed comparison

Download full size

Fig. 6. Detection results of the improved algorithm. (a) Target missed detection map; (b) traditional algorithm; (c) improved algorithm

Download full size

Table 1. Detection accuracy and speed under different candidate regions
View table
Table 1. Detection accuracy and speed under different candidate regions
Proposal 2000 1500 1000 500 50 APR(300--1600)
Detection time /s 0.215 0.194 0.160 0.135 0.095 0.157
mAP /% 73.5 73.5 72.6 71.1 69.8 73.5
Δt /% 0 -11 -24 -37 -56 -27
Δm /percentage point 0 0 -0.9 -2.4 -3.7 0

Table 2. Detection accuracy and speed under different candidate frame fluctuation ranges
View table
Table 2. Detection accuracy and speed under different candidate frame fluctuation ranges
Proposal 200--1400 300--1600 400--1800
Detection time /s 0.143 0.157 0.168
mAP /% 73.1 73.5 73.5

Table 3. Influence of adaptive confidence threshold on algorithm performance
View table
Table 3. Influence of adaptive confidence threshold on algorithm performance
Algorithm S_AP /% mAP /%
Faster-RCNN 13.2 73.5
Faster-RCNN+A1 20.1 74.7

Table 4. Influence of APR on algorithm performance
View table
Table 4. Influence of APR on algorithm performance
Algorithm mAP /% Detection time /s
Faster-RCNN 73.5 0.215
Faster-RCNN+APR+A1 75.4 0.161

Tools

Get Citation

Copy Citation Text

Benyuan Lü, Zhenfu Zhuo, Yongsai Han, Lichao Zhang. Target Detection Based on Faster Region Convolution Neural Network[J]. Laser & Optoelectronics Progress, 2021, 58(22): 2210017

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites