A Deep Neural Network-Based Target Recognition Algorithm for Robot Scenes

Table 1

Detection results of different network models for PASCAL VOC2007.

Methods	Basic network	Accuracy	Detection speed	Image size
Faster	VGG16	73.2	7	1000 × 600
Faster	Residual-101 [10]	76.4	2.4	1000 × 600
R-FCN	Residual-101	80.5	9	1000 × 600
DSOD300	DS/64-192-4	77.7	17.4	300 × 300
YOLOv2	Darknet-19	78.6	40	544 × 544
SSD300	VGG16	77.5	46	300 × 300
DSSD32	Residual-101	79.5	9.5	321 × 321
STDN321	DenseNet-169	79.2	41.5	321 × 321
Ours320	VGG16	80.1	31.2	320 × 320
SSD512	VGG16	78.6	19	512 × 512
DSSD513	Residual-101	81.5	5.5	513 × 513
STDN513	DenseNet-169	80.9	28.6	513 × 513
Ours512	VGG16	81.9	18.2	512 × 512

Bold values represent the experimental results of our method.