Dense control valve parts dataset for industrial object detection

	图像（张）	样本	阀体	上阀盖	套筒	柱状阀杆	杆状阀杆	阀座	阀芯	上套筒	导向套
训练集	408	12 089	1 820	2 381	2 295	1 921	590	2 210	449	195	228
验证集	51	1 523	276	283	305	278	47	260	35	25	14
测试集	51	1 403	232	285	280	240	59	221	47	17	22
总数	510	15 015	2 328	2 949	2 880	2 439	696	2 691	531	237	264

Table 2. Average number of samples per dataset
View table
View in Article
Table 2. Average number of samples per dataset
数据集平均样本数
MS COCO 7.19
PASCAL VOC（07++12） 2.89
Open Images 5.00
ImageNet 1.37
DOTA 3.50
LEVIR 0.50
PD4CV2023 29.00

Table 3. Comparison of relevant datasets

View table

View in Article

Table 3. Comparison of relevant datasets

名称	类别	图像数	用途	样本标记数
MS COCO	80	123 287	一般目标检测	886 266
PASCAL VOC（07++12）	20	21 503	一般目标检测	62 199
ImageNet	200	349 319	一般目标检测	478 806
DOTA	15	2 806	航空目标检测	188 282
LEVIR	3	22 000	遥感目标检测	11 000
KolektorSDD	1	399	工业缺陷检测	52
DeepPCB	6	3 000	电路板缺陷检测	8 853
NEU-CLS	6	1 800	钢板缺陷检测	1 800
PD4CV2023	9	510	控制阀目标检测	15 015

Table 4. Comparison of part detection results of different training sets of models on the PD4CV2023 dataset
View table
View in Article
Table 4. Comparison of part detection results of different training sets of models on the PD4CV2023 dataset
YOLOv5s YOLOv8s
训练集 MS COCO KolektorSDD NEU-CLS PD4CV2023 MS COCO KolektorSDD NEU-CLS PD4CV2023
测试集 PD4CV2023 PD4CV2023 PD4CV2023 PD4CV2023 PD4CV2023 PD4CV2023 PD4CV2023 PD4CV2023
mAP 0 0 0 88.20 0 0 0 85.40

Table 5. Comparison of detection performance of different object detection methods in the PD4CV2023 dataset
View table
View in Article
Table 5. Comparison of detection performance of different object detection methods in the PD4CV2023 dataset
年份模型 mAP FPS
2016 Faster-RCNN 75.38 10
2016 SSD 75.83 23
2017 DETR 55.11 20
2018 YOLOv3 80.67 17
2020 YOLOv5s 88.20 47
2023 YOLOv8s 85.40 42

Table 6. Comparison of different object detection methods on parts of different sizes in the PD4CV2023 dataset

View table

View in Article

Table 6. Comparison of different object detection methods on parts of different sizes in the PD4CV2023 dataset

尺寸	类别	Faster-RCNN	SSD	DETR	YOLOv3	YOLOv5s	YOLOv8s	均值
	导向套	54.19	69.83	55.11	59.49	76.80	66.40	61.86
较小	套筒	64.07	91.57	37.93	95.47	98.80	98.00	80.97
	柱状阀杆	69.17	65.26	48.65	79.92	91.20	91.60	74.30
	均值	62.48	75.55	47.23	78.29	88.93	85.33
	杆状阀杆	68.51	83.32	57.85	82.69	91.70	87.00	78.51
中等	阀座	65.21	82.44	34.00	91.47	95.50	96.30	77.49
	阀芯	75.27	37.90	46.47	66.38	55.20	50.70	55.32
	均值	69.66	67.89	46.11	80.18	80.80	78.00
	阀体	94.32	96.52	90.84	99.02	98.60	98.50	96.30
较大	上阀盖	92.35	96.57	80.46	98.45	96.30	97.10	93.54
	上套筒	95.37	59.09	55.36	53.11	89.50	82.90	72.56
	均值	94.01	84.06	75.55	83.53	94.80	92.83

Table 7. Comparison of different object detection methods on parts of different sizes in the PD4CV2023 dataset

View table

View in Article

Table 7. Comparison of different object detection methods on parts of different sizes in the PD4CV2023 dataset

规模	类别	数量（个）	比例	Faster-RCNN	SSD	DETR	YOLOv3	YOLOv5s	YOLOv8s	均值
	上套筒	237	1.58	95.37	59.09	55.36	53.11	89.50	82.90	72.56
较小	导向套	264	1.76	54.19	69.83	55.11	59.49	76.80	66.40	61.86
较小	阀芯	531	3.54	75.27	37.90	46.47	66.38	55.20	50.70	55.32
	杆状阀杆	696	4.64	68.51	83.32	57.85	82.69	91.70	87.00	78.51
			均值	73.34	62.54	53.70	65.43	78.30	71.75
	阀体	2 328	15.50	94.32	96.52	90.84	99.02	98.60	98.50	96.30
	柱状阀杆	2 439	16.24	69.17	65.26	48.65	79.92	91.20	91.60	74.30
较大	阀座	2 691	17.92	65.21	82.44	34.00	91.47	95.50	96.30	77.49
	套筒	2 880	19.18	64.07	91.57	37.93	95.47	98.80	98.00	80.97
	上阀盖	2 949	19.64	92.35	96.57	80.46	98.45	96.30	97.10	93.54
			均值	77.02	86.47	58.38	92.87	96.08	96.30

Table 8. Comparison of efficiency and error rates of different methods
View table
View in Article
Table 8. Comparison of efficiency and error rates of different methods
方式分拣速度错误率
人工 50个/h 12.00%
YOLOv5s 1 450个/h 11.80%

Tools

Get Citation

Copy Citation Text

Linyi WANG, Jing BAI, Yanmei LI, Wenjing LI. Dense control valve parts dataset for industrial object detection[J]. Optics and Precision Engineering, 2024, 32(8): 1241

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category:

Received: Sep. 22, 2023

Accepted: --

Published Online: May. 29, 2024

The Author Email: BAI Jing (baijing_nun@163. com)

DOI:10.37188/OPE.20243208.1241

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. PD4CV2023 dataset data statistics

Table 1. PD4CV2023 dataset data statistics

Table 2. Average number of samples per dataset

Table 2. Average number of samples per dataset

Table 3. Comparison of relevant datasets

Table 3. Comparison of relevant datasets

Table 4. Comparison of part detection results of different training sets of models on the PD4CV2023 dataset

Table 4. Comparison of part detection results of different training sets of models on the PD4CV2023 dataset

Table 5. Comparison of detection performance of different object detection methods in the PD4CV2023 dataset

Table 5. Comparison of detection performance of different object detection methods in the PD4CV2023 dataset

Table 6. Comparison of different object detection methods on parts of different sizes in the PD4CV2023 dataset

Table 6. Comparison of different object detection methods on parts of different sizes in the PD4CV2023 dataset

Table 7. Comparison of different object detection methods on parts of different sizes in the PD4CV2023 dataset

Table 7. Comparison of different object detection methods on parts of different sizes in the PD4CV2023 dataset

Table 8. Comparison of efficiency and error rates of different methods

Table 8. Comparison of efficiency and error rates of different methods

微信扫一扫：分享