Detection Method for Field-of-view Defect of Ultraviolet Image Intensifier Based on Improved SSD Algorithm

doi:10.12382/bgxb.2023.1091

Abstract

Abstract:

The field-of-view defect of UV image intensifier is acknowledged as a crucial element impacting the imaging performance of such device. The data enhancement procedures are used to address the issues of few field-of-view defect samples and the significant variances of images in field-of-view. A feature pyramid network (FPN) is added to the single shot multibox detector (SSD) algorithm for the successful detection and fusion of multiscale features. A convolutional block attention module (CBAM) is also introduced to improve the network’s focus on small defect targets and minimize noise interference. The experimental results show that, on the self-constructed dataset, the feature pyramid network-convolutional block attention module-single shot multibox detector (FPN-CBAM-SSD) algorithm outperforms the SSD algorithm significantly in the actual detection of field-of-view defects. For 5 categories of defects including bright spots, dark patches, striped defects, bright patches, and dark spots, the average detection accuracy is improved by 19.76%, 22.84%, 29.56%, 34.55%, and 38.14%, respectively. FPN-CBAM-SSD algorithm is capable of meeting the practical application requirements and adapting to more complex field-of-view conditions, making it an effective method for detecting field-of-view defects in ultraviolet image intensifiers.

Key words: ultraviolet image intensifier, field-of-view defect detection, machine vision, deep learning

CLC Number:

TN23，TP391.4

DING Xiwen, CHENG Hongchang, YUAN Yuan, SU Yue. Detection Method for Field-of-view Defect of Ultraviolet Image Intensifier Based on Improved SSD Algorithm[J]. Acta Armamentarii, 2024, 45(12): 4350-4363.

Figures/Tables 22

Fig.1 Overall flowchart of the UV image intensifier field-of-view defect detection experiment

Fig.2 Differences in the display effects of images from different acquisition batches

Fig.3 Example effect of image data enhancement

Fig.4 Main types of defects in the field of view of UV image intensifier

Fig.5 Labeling of field of view defects of UV image intensifier

Fig.6 Approximate proportion of the number of major types of defects in the field of view

Fig.7 Structure diagram of SSD detection algorithm

Fig.8 Size and distribution of labeled boxes in the dataset

Fig.9 Detection results for different anchor-sizes

Fig.10 Schematic diagram of feature pyramid network

Fig.11 Schematic diagram of convolutional attention mechanisms

Fig.12 Schematic diagram of the structure of FPN-CBAM-SSD algorithm

Table 1 Configuration of hardware environment of the experimental platform

实验硬件配置		型号	生产厂家
光源		BM-Ⅰ No.1608	上海灯泡三厂
外接电源		GWLNSTEK GPP-4323	固纬电子(苏州)有限公司
数码相机		Canon DS126151 400D	佳能(中国)有限公司
计算机	处理器	Intel(R) Core(TM) i5-12400 4.40GHz	英特尔(中国)有限公司
	显卡	Nvidia GTX 3060 12GB	技嘉科技股份有限公司

Table 2 Configuration of software environment of the experimental platform

实验软件环境	具体版本号
Python	3.6
CUDA	11.3
CuDnn	8.4.1

Fig.13 Loss trends of different improved SSD algorithms

Table 3 Comparison of ablation test performance indexes

算法	准确率/ %	召回率/ %	F1分数/ %	mAP/%					mAP/ %	FPS/ (帧·s^-1)
算法	准确率/ %	召回率/ %	F1分数/ %	暗点	亮点	暗斑	亮斑	条纹状	mAP/ %	FPS/ (帧·s^-1)
SSD算法	90.0	14.3	23.2	22.97	43.05	48.57	42.12	27.07	36.76	70.90
CBAM-SSD算法	87.6	26.4	40.5	32.80	46.12	60.58	53.42	36.89	45.96	63.53
FPN-SSD算法	74.7	32.9	43.4	44.82	40.13	66.29	57.56	38.49	49.46	51.47
FPN-CBAM-SSD算法	81.9	51.9	63.2	61.11	62.81	71.41	76.67	56.63	65.73	15.57

Table 4 Performance comparison of different detection algorithms on the self-built UV imageintensifier field of view dataset

算法	平均精准度/%					mAP/ %	参数量/ 10⁶	GFLOPS/ G	FPS/ (帧·s^-1)
算法	暗点	亮点	暗斑	亮斑	条纹状	mAP/ %	参数量/ 10⁶	GFLOPS/ G	FPS/ (帧·s^-1)
SSD算法	22.97	43.05	48.57	42.12	27.07	36.76	24.15	61.14	70.90
FPN-CBAM-SSD算法	61.11	62.81	71.41	76.67	56.63	65.73	30.83	66.13	15.57
Faster-RCNN算法	18.54	23.53	72.85	61.16	42.84	43.78	136.77	369.82	11.92
YOLOv5算法	31.41	23.40	32.04	32.79	25.46	29.02	7.07	16.51	102.19
YOLOv8算法	31.84	25.25	36.72	34.62	27.79	31.25	11.14	28.66	96.11

Table 5 Performance comparison of different detection a lgorithms on VOC 07+12 dataset

算法	mAP/ %	参数量/ 10⁶	GFLOPS/ 10⁹	FPS/ (帧·s^-1)
SSD算法	79.24	26.15	62.65	46.38
FPN-CBAM-SSD算法	83.63	31.77	66.83	13.96
Faster-RCNN算法	78.46	137.08	370.19	7.19
YOLOv5算法	77.12	7.12	16.64	101.41
YOLOv8算法	81.61	11.14	28.69	93.94

Table 6 Typical case test results

Fig.14 Statistical analysis of data for detected results of field-of-view defects in typical cases

Table 7 Complex field-of-view test results

Fig.15 Statistical analysis of data for detected results of field-of-view defects in complex cases

References 25

[1]	石峰, 程宏昌, 闫磊, 等. 紫外探测技术[M]. 北京: 国防工业出版社, 2017:12.
	SHI F, CHENG H C, YAN L, et al. Ultraviolet detection technology[M]. Beijing: National Defense Industry Press, 2017:12. (in Chinese)
[2]	向世明, 倪国强. 光电子成像器件原理[M]. 北京: 国防工业出版社,1999.
	XIANG S M, NI G Q. The principles of photoelectronic imaging devices[M]. Beijing: National Defense Industry Press, 1999. (in Chinese)
[3]	许强. 军用紫外探测技术及应用[M]. 北京: 北京航空航天大学出版社, 2010.
	XU Q. Military ultraviolet detection technology and applications[M]. Beijing: Beijing University of Aeronautics and Astronautics Press, 2010. (in Chinese)
[4]	张忠廉, 刘榴娣. 紫外线技术在军事上的应用研究[J]. 光学技术, 2000, 26(4):289-293, 296.
	ZHANG Z L, LIU L D. Study on application of ultraviolet technology in military affairs[J]. Optical Technology, 2000, 26(4):289-293,296. (in Chinese)
[5]	杨杰. 紫外探测技术的应用与进展[J]. 光电子技术, 2011, 31(4):274-278.
	YANG J. The application and development of UV detection technology[J]. Optoelectronics Technology, 2011, 31(4):274-278. (in Chinese)
[6]	张简屏, 常乐, 陈超, 等. 日盲紫外像增强器的研究现状及应用前景[J]. 云光技术, 2021, 53(2):68-74.
	ZHANG J P, CHANG L, CHEN C, et al. Research status and application prospect of ultraviolet solar-blind image intensifier[J]. Cloud Optical Technology, 2021, 53(2):68-74. (in Chinese)
[7]	唐光华, 宋晓峰, 汪述猛, 等. 日盲型紫外像增强管的应用及研制进展[J]. 光电子技术, 2016, 36(3):164.
	TANG G H, SONG X F, WANG S M, et al. Application and development of ultraviolet solar-blind image intensifiers[J]. Optoelectronics Technology, 2016, 36(3):164. (in Chinese)
[8]	赵清波. 宽光谱像增强器辐射增益和视场缺陷测试技术研究[D]. 南京: 南京理工大学, 2008.
	ZHAO Q B. Research on radiation gain and field defect test technology of wide spectrum image intensifier[D]. Nanjing: Nanjing University of Science and Technology, 2008. (in Chinese)
[9]	FU R G, WEI Y F, YANG Q, et al. The analysis of the defects of the view field of the UV image intensifier[C]// Proceedings of the Sensors and Systems for Space Applications X. Anaheim, CA, US: SPIE, 2017, 10196: 19-26. (in Chinese)
[10]	杨琦. 紫外像增强器视场缺陷检测技术研究[D]. 南京: 南京理工大学, 2011.
	YANG Q. Research on defect detection technology of ultraviolet image intensifier[D]. Nanjing: Nanjing University of Science and Technology, 2011. (in Chinese)
[11]	汤勃, 孔建益, 伍世虔. 机器视觉表面缺陷检测综述[J]. 中国图象图形学报, 2017, 22(12):1640-1663.
	TANG B, KONG J Y, WU S Q. Review of machine vision surface defect detection[J]. Chinese Journal of Image and Graphics, 2017, 22(12): 1640-1663. (in Chinese)
[12]	张涛, 刘玉婷, 杨亚宁, 等. 基于机器视觉的表面缺陷检测研究综述[J]. 科学技术与工程, 2020, 20(35):14366-14376.
	ZHANG T, LIU Y T, YANG Y N, et al. Review of surface defect detection based on machine vision[J]. Science, Technology and Engineering, 2020, 20(35):14366-14376. (in Chinese)
[13]	顾佳晨, 高雷, 刘路硌. 基于深度学习的目标检测算法在冷轧表面缺陷检测中的应用[J]. 冶金自动化, 2019, 43(6): 19-22.
	GU J C, GAO L, LIU L G. Application of object detection algorithm based on deep learning for inspection of surface defect of cold rolled strips[J]. Metallurgical Automation, 2019, 43(6): 19-22. (in Chinese)
[14]	张良安, 陈洋, 谢胜龙, 等. 基于机器视觉与深度学习的飞机防护栅裂纹检测系统[J]. 兵工学报, 2023, 44(2):507-516. doi: 10.12382/bgxb.2021.0674
	ZHANG L A, CHEN Y, XIE S L, et al. Crack detection system for aircraft protective grill based on machine vision and deep learning[J]. Acta Armamentarii, 2023, 44(2): 507-516. (in Chinese) doi: 10.12382/bgxb.2021.0674
[15]	郭永红, 牛海涛, 史超, 等. 基于卷积和注意力机制的小样本目标检测[J]. 兵工学报, 2023, 44(11):3508-3515. doi: 10.12382/bgxb.2022.1167
	GUO Y H, NIU H T, SHI C, et al. Few-shot object detection based on convolution network and attention mechanism[J]. Acta Armamentarii, 2023, 44(11): 3508-3515. (in Chinese) doi: 10.12382/bgxb.2022.1167
[16]	景军锋, 刘娆. 基于卷积神经网络的织物表面缺陷分类方法[J]. 测控技术, 2018, 37(9): 20-25.
	JING J F, LIU R. Classification method of fabric surface defects based on convolution neural network[J]. Measurement and control technology, 2018, 37(9): 20-25. (in Chinese)
[17]	李文俊, 陈斌, 李建明, 等. 基于深度神经网络的表面划痕识别方法[J]. 计算机应用, 2019, 39(7):2103-2108.
	LI W J, CHEN B, LI J M, et al. Surface scratch recognition method based on deep neural network[J]. Computer application, 2019, 39(7): 2103-2108. (in Chinese)
[18]	LIM J S, ASTRID M, YOON H J, et al. Small object detection using context and attention[C]// Proceedings of the 3rd international Conference on Artificial intelligence in information and Communication (ICAIIC). Jeju Island, Korea:IEEE, 2021: 181-186.
[19]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector[C]// Proceedings of the 14th European Conference Computer Vision (ECCV). Amsterdam, The Netherlands. Springer, 2016: 21-37.
[20]	AHMED M, SERAJ R, ISLAM S M S. The k-means algorithm: a comprehensive survey and performance evaluation[J]. Electronics, 2020, 9(8): 1295.
[21]	SHI N, LIU X M, GUAN Y. Research on k-means clustering algorithm: an improved k-means clustering algorithm[C]// Proceedings of the 3rd International Symposium on Intelligent Information Technology and Security Informatics. Jiangxi, China: IEEE, 2010: 63-67.
[22]	GONG Y Q, YU X H, DING Y, et al. Effective fusi-on factor in FPN for tiny object detection[C]// Proceedings of the 21st Winter Conference on Applications of Computer Vision (WACV). Waikoloa, HI, US: IEEE, 2021: 1160-1168.
[23]	LI Z H, ZHANG S, ZHANG J G, et al. MVP-Net: multi-view FPN with position-aware attention for deep universal lesion detection[C]// Proceedings of the 22nd Medical Image Computing and Computer Assisted Intervention (MICCAI). Shenzhen, China: Springer, 2019: 13-21.
[24]	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]// Proceedings of the 15th European conference on computer vision (ECCV). Munich, Germany: Springer, 2018: 3-19.
[25]	CHRISTOU C, ELIOPHOTOU M M, PHILIPPOU G. Teachers’ concerns regarding the adoption of a new mathematics curriculum: an application of CBAM[J]. Educational studies in mathematics, 2004, 57: 157-176.