Sewing Gesture Image Detection Method Based on Improved SSD Model

Weiming Yao; Xiaohua Wang; Nan Wu

doi:10.3788/LOP57.181017

Journals >Laser & Optoelectronics Progress >Volume 57 >Issue 18 >Page 181017 > Article

Laser & Optoelectronics Progress
Vol. 57, Issue 18, 181017 (2020)

Sewing Gesture Image Detection Method Based on Improved SSD Model

Weiming Yao¹, Xiaohua Wang^1、*, and Nan Wu²

Author Affiliations

¹College of Electronics and Information, Xi'an Polytechnic University, Xi'an, Shaanxi 710048, China

²College of Information Engineering, Shaanxi Xueqian Normal University, Xi'an, Shaanxi 710048, China

show less

DOI: 10.3788/LOP57.181017 Cite this Article Set citation alerts

Weiming Yao, Xiaohua Wang, Nan Wu. Sewing Gesture Image Detection Method Based on Improved SSD Model[J]. Laser & Optoelectronics Progress, 2020, 57(18): 181017 Copy Citation Text

show less

Fig. 1. SSD improved model network structure

Download full size

Fig. 2. FPN algorithm structure diagram

Download full size

Fig. 3. Model loss curve

Download full size

Fig. 4. Detection effect under different illumination conditions. (a) Cladding sewing; (b) crimp sewing; (c) fabric cutting; (d) knots-making sewing

Download full size

Fig. 5. Experimental detection effect under different algorithms. (a) Cladding sewing; (b) crimp sewing; (c) fabric utting; (d) knots-making sewing

Download full size

Category	Training	Validation	Test	Total
Cladding sewing	900	290	290	1500
Crimp sewing	840	280	280	1400
Fabric cutting	960	320	320	1600
Knots-makingsewing	780	260	260	1300

Table 1. Number of images in amplification dataset

Mothod	mAP /%
Mothod	VOC	NUS-II
YOLO	0.755	0.842
SSD	0.785	0.883
SSD-modified	0.793	0.903

Table 2. Recognition results comparison of 4 algorithms

Data category	Morningbrightness	Noonbrightness	Nightbrightness
A-F1 score	0.834	0.732	0.763
B-F1 score	0.842	0.825	0.838

Table 3. Experimental results under different lighting conditions

Data augmentation		Backbone		Feature layer fusion
Original	Corrected	VGG16	Resnet50	Original	Corrected
77.23	85.44	81.35	83.34	80.44	83.28

Table 4. Comparison of mAP results in ablation experiments%

Method	AP /%				mAP /%	Small target	Detectionspeed /s^-1
Method	Claddingsewing	Crimpsewing	Fabriccutting	Knots-making sewing	mAP /%	Small target	Detectionspeed /s^-1
Faster R-CNN	88.21	84.56	86.93	83.62	85.83	63.23	13
YOLO	85.22	85.04	77.90	85.48	83.41	57.44	47
SSD	87.34	83.55	84.23	87.48	85.65	55.34	48
RetinaNet	87.39	86.04	87.34	87.96	87.18	65.78	51
SSD-modified	87.21	89.45	88.58	89.52	88.69	67.44	52

Table 5. Comparison of recognition results under 4 algorithms

Weiming Yao, Xiaohua Wang, Nan Wu. Sewing Gesture Image Detection Method Based on Improved SSD Model[J]. Laser & Optoelectronics Progress, 2020, 57(18): 181017

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information