Light Guide Plate Defect Detection Combing Light Weight and Cascade Deep Learning Network

Table 1. Dataset setting
View table
Table 1. Dataset setting
Dataset Kind Number Training Validation Test
Dataset1 Positive(normal) 15500 10850 1550 3100
Negative(defective) 11373 7962 1138 2273
Dataset2 Normal 4000 2800 400 800
Area 3706 2595 371 740
Lines 3967 2777 397 793
Dots 4197 2938 420 839

Table 2. Test results of different classification networks on light guide plate dataset

View table

Table 2. Test results of different classification networks on light guide plate dataset

Network	2-classes		4-classes				FLOPs /G	Speed /s^-1
Network	Normal /%	Defects /%	Normal /%	Area /%	Lines /%	Dots /%	FLOPs /G	Speed /s^-1
VGG19^[21]	97.54	97.21	97.25	97.03	97.22	97.74	19.61	232
ResNet18^[23]	99.58	99.52	99.50	98.92	98.87	98.33	1.74	427
MobileNetv2^[25]	98.86	99.33	99.46	97.33	96.68	98.49	0.06	660
LW_CNN	99.65	99.63	98.52	93.32	98.25	95.79	0.02	811
IR_CNN	99.77	99.78	99.63	99.32	98.99	99.17	7.63	320

Table 3. Comparision of algorithm in this paper and method in Ref. [18]

View table

Table 3. Comparision of algorithm in this paper and method in Ref. [18]

Defect type
Normal	150	100.00	96.00	1.95	6.8
Area defect	85	98.82	87.06
Line defect	127	96.85	96.85
Dot defect	138	97.83	95.65
Average	500	98.40	94.60

Tools

Get Citation

Copy Citation Text

Junfeng Li, Yansen He, Wenzhan Dai. Light Guide Plate Defect Detection Combing Light Weight and Cascade Deep Learning Network[J]. Laser & Optoelectronics Progress, 2021, 58(14): 1410009

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites