Application of Improved Faster RCNN Model for Foreign Fiber Identification in Cotton

Yuhong Du; Chaoqun Dong; Di Zhao; Weijia Ren; Wenchao Cai

doi:10.3788/LOP57.121007

Laser & Optoelectronics Progress, Volume. 57, Issue 12, 121007(2020)

Application of Improved Faster RCNN Model for Foreign Fiber Identification in Cotton

Yuhong Du^1,2、*, Chaoqun Dong^1,2, Di Zhao^1,2, Weijia Ren^1,2, and Wenchao Cai³

Author Affiliations

¹College of Mechanical Engineering, Tianjin Polytechnic University, Tianjin 300387, China

²Tianjin Key Laboratory of Advanced Mechatronics Equipment Technology, Tianjin 300387, China

³Beijing Daheng Image Vision Co., Ltd., Beijing 100085, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(10)

Fig. 1. Experimental materials. (a) Experimental raw materials; (b) selected materials

Download full size

Fig. 2. Image acquisition equipment and sample examples. (a) Image acquisition device; (b) small size sample; (c) medium size sample; (d) large size sample; (e) densely distributed sample

Download full size

Fig. 3. Residual module structure

Download full size

Fig. 4. Cluster analysis of mark box size

Download full size

Fig. 5. Comparison of anchors generation size before and after improvement. (a)(b) Before improvement; (c)(d) after improvement

Download full size

Fig. 7. Examples of model detection effect before and after improvement. (a) Missed detection in Faster RCNN; (b) misjudged as two foreign fibers in Faster RCNN; (c) repeat recognition in Faster RCNN; (d) not missed in Faster RCNN Pro; (e) recognized as a foreign fiber in Faster RCNN Pro; (f) not re-identified in Faster RCNN Pro

Download full size

Table 1. Size and aspect ratio of the cluster center of mark box

View table

Table 1. Size and aspect ratio of the cluster center of mark box

Serial number	1	2	3	4	5	6	7	8	9	10	11	12
Length /pixel	78.7	129.2	90.8	83.7	100.3	60.5	116.5	299.3	167.0	52.9	183.4	49.8
Width /pixel	261.8	103.7	134.8	95.4	54.4	178.8	176.5	87.7	70.6	71.1	123.9	118.9
Aspect ratio	1∶3.3	1.2∶1	1∶1.5	1∶1.1	1.8∶1	1∶3.0	1∶1.5	3.4∶1	2.4∶1	1∶1.3	1.5∶1	1∶2.4

Table 2. Identification and evaluation parameters of different sizes of foreign fibers

View table

Table 2. Identification and evaluation parameters of different sizes of foreign fibers

Species	Size code	Number	A /%	P /%	R /%	F₁	F₁ average
	Small	26	88.89	92.31	96.15	0.942
Chemical fiber	Middle	91	94.62	97.78	96.70	0.972	0.975
	Big	87	97.70	100.00	97.70	0.988
	Small	115	91.45	93.04	96.52	0.948
Plastic film	Middle	88	94.51	96.63	97.72	0.972	0.941
	Big	7	100.00	100.00	100.00	1.000
	Small	83	84.27	92.59	90.36	0.915
PP yarn	Middle	70	94.37	98.53	95.71	0.971	0.913
	Big	57	92.98	100.00	92.98	0.964
	Small	20	90.00	100.00	90.00	0.947
Feather	Middle	109	100.00	100.00	100.00	1.000	0.991
	Big	93	100.00	100.00	100.00	1.000

Table 3. Model evaluation parameters before and after improvement
View table
Table 3. Model evaluation parameters before and after improvement
Model A /% P /% R /% F₁
Faster RCNN(before improvement) 91.03 97.26 93.42 0.953
Faster RCNN Pro(after improvement) 94.24 98.16 95.93 0.970

Table 4. Comparison of different model evaluation parameters
View table
Table 4. Comparison of different model evaluation parameters
Model A /% P /% R /% F₁
Faster RCNN Pro 94.24 98.16 95.93 0.970
SSD 90.22 97.11 92.70 0.949
YOLOv3 90.89 97.49 93.06 0.952
HOG+SVM 76.49 69.62 70.45 0.700

Tools

Get Citation

Copy Citation Text

Yuhong Du, Chaoqun Dong, Di Zhao, Weijia Ren, Wenchao Cai. Application of Improved Faster RCNN Model for Foreign Fiber Identification in Cotton[J]. Laser & Optoelectronics Progress, 2020, 57(12): 121007

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites