Improved YOLO11 Optical Detection Method for Crystalline Surface Defects

doi:10.16553/j.cnki.issn1000-985x.2025.0178

Abstract

Abstract: Crystal products are prone to scratch defects during production， manufacturing， and processing， and accurate identification of such defects remains a technical challenge in this field. To address the urgent demand for multi-type scratch detection under complex imaging conditions， this paper proposes an improved approach based on YOLO11 by introducing a deformable attention transformer （DAT） and large separable kernel attention （LSKA）. The method improves adaptive modeling and recognition of surface defects by leveraging deformable convolutions and multi-scale receptive field fusion. Compared with traditional models such as YOLOv3， YOLOv5， YOLOv8， and the baseline YOLO11， the improved YOLO11-DAT_LSKA achieves performance improvements of 2.5， 2.3， 1.9 and 1.5 percentage points respectively in terms of mAP@0.5 for defect detection. These results demonstrate the effectiveness of the proposed method in enhancing feature modeling capabilities and improving the perception of complex scratches， thereby improving the detection accuracy for surface defects on crystal materials. In particular， to address low-contrast and irregular crystal surface defects， the DAT and LSKA modules enhance global contextual modeling and multi-scale receptive field adaptation， enabling effective capture of defect shape variations and multi-scale defect structures.

Key words: crystalline surface defect; YOLO11; deformable attention transformer; large separable kernel attention; attention mechanism

CLC Number:

TP391

YIN Chuangye, WANG Huadong, ZHANG Qingli, SUN Guihua, SUN Yu, ZHANG Zhirong. Improved YOLO11 Optical Detection Method for Crystalline Surface Defects[J]. Journal of Synthetic Crystals, 2026, 55(3): 368-377.

Figures/Tables 16

Fig.1 Diagram of experimental system setup for surface defects detection

Fig.2 Photograph of experimental system setup for surface defects detection

Fig.3 Original architecture diagram of YOLO11 network

Fig.4 Improved architecture diagram of YOLO11 network

Fig.5 Deformable attention module

Fig.6 Offset network

Fig.7 Diagram of fixed and variable convolutions.（a） Conventional square convolution；（b），（c） two examples of deformable convolution. The light yellow dots represent the 9 sampling points of the conventional square convolution kernel， the black arrows indicate the offset vector， and the light blue ones are the sampled points after offsetting

Fig.8 Diagram of deformable convolution structure

Fig.9 An illustration of DAT architecture

Fig.10 Structural diagrams of LKA （a） and LSKA （b）

Table 1 Data classification table

缺陷类别	训练集缺陷数量	验证集缺陷数量	测试集缺陷数量	总缺陷数量
Planar_defect	6 543	1 067	1 113	8 723
Linear_defect	7 769	982	890	9 641

Table 2 Comparison results of various models

Network	P/%	Re/%	mAP@0.5/%
YOLOv3	80.9	84.1	85.6
YOLOv5	81.1	84.2	85.8
YOLOv8	79.9	86.5	86.2
YOLO11	82.1	86.3	86.6
YOLO11-DAT_LSKA	88.8	87.2	88.1

Fig.11 Comparison of precision-recall curves for different models

Fig.12 Comparison chart of mAP@0.5 curves for different models

Fig.13 Comparison of different models for crystal surface scratch detection

Table 3 Ablation experiment data

Network	P/%	Re/%	mAP@0.5/%
YOLO11	82.1	86.3	86.6
YOLO11-DAT	82.7	87.1	87.4
YOLO11-LSKA	88.2	87.3	87.8
YOLO11-DAT_LSKA	88.8	87.2	88.1

References 19

[1]	孙　伟，仲佳勇. 在兆焦耳级激光设备上的实验室天体物理研究进展［J］. 天文学进展， 2020， 38（3）： 243-254.
	SUN W， ZHONG J Y. Recent advances in laboratory astrophysics on mega-joule class laser facility［J］. Progress in Astronomy， 2020， 38（3）： 243-254 （in Chinese）.
[2]	杨　舟. 基于深度学习的单晶硅生长缺陷及芯片缺陷检测研究［D］. 合肥：安徽工业大学， 2023.
	YANG Z. Deep learning based monocrystal silicon growth defect and chip defect detection research［D］. Hefei： Anhui University of Technology， 2023 （in Chinese）.
[3]	钱梦雪，张志荣，王华东，等. 基于片状光束扫描的激光晶体内在缺陷表征方法［J］. 人工晶体学报， 2024， 53（2）： 238-245.
	QIAN M X， ZHANG Z R， WANG H D， et al. Characterization method for internal defects in laser crystals based on slice beam scanning［J］. Journal of Synthetic Crystals， 2024， 53（2）： 238-245 （in Chinese）.
[4]	倪开灶，刘世杰，吴周令，等. 激光聚焦线扫描法测量KDP晶体坯片的体缺陷［J］. 光学精密工程， 2016， 24（12）： 3020-3026.
	NI K Z， LIU S J， WU Z L， et al. Measurement of bulk defects for KDP crystal billet with focused line scanning［J］. Optics and Precision Engineering， 2016， 24（12）： 3020-3026 （in Chinese）.
[5]	SHEFER D， NIKIPELOV A， VAN DE KERKHOF M， et al. Laser light scattering （LLS） to observe plasma impact on the adhesion of micrometer-sized particles to a surface［J］. Journal of Physics D： Applied Physics， 2023， 56（45）： 455201.
[6]	YIN J F， BAI Q， HAITJEMA H， et al. Depolarization of surface scattering in polarized laser scattering detection for machined silicon wafers［J］. Precision Engineering， 2022， 73： 203-213.
[7]	杨菲菲，谬　洁，谢雨江，等. 全内反射技术检测大口径光学元件体内缺陷［J］. 中国激光， 2017， 44（6）： 0604005.
	YANG F F， MIU J， XIE Y J， et al. Internal defects of large optics detected by total internal reflection technique［J］. Chinese Journal of Lasers， 2017， 44（6）： 0604005 （in Chinese）.
[8]	张志豪，厉小润，陈淑涵. 基于改进YOLO11的无人机航拍图像小目标检测算法［J］. 液晶与显示， 2025， 40（6）： 915-930.
	ZHANG Z H， LI X R， CHEN S H. Small object detection algorithm in UAV aerial images based on improved YOLO11［J］. Chinese Journal of Liquid Crystals and Displays， 2025， 40（6）： 915-930 （in Chinese）.
[9]	孙佩月，黄　娟，顾寄南，等. 基于YOLO11n的轻量化芯片表面缺陷检测方法［J］. 半导体技术， 2025， 50（7）： 707-713.
	SUN P Y， HUANG J， GU J N， et al. Lightweight chip surface defect detection method based on YOLO11n［J］. Semiconductor Technology， 2025， 50（7）： 707-713 （in Chinese）.
[10]	CHEN S L， SUN H B， WU Y X， et al. A helmet detection algorithm based on transformers with deformable attention module［J］. Chinese Journal of Electronics， 2025， 34（1）： 229-241.
[11]	ZHANG H， XU C， ZHANG S J. Inner-IoU： more effective intersection over union loss with auxiliary bounding box［EB/OL］. 2023： arXiv： 2311.02877. .
[12]	XIA Z F， PAN X R， SONG S J， et al. Vision transformer with deformable attention［C］//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）. June 18-24， 2022. New Orleans， LA， USA. IEEE， 2022： 4784-4793.
[13]	冉险生，刘圣斌. 基于改进的YOLOv8检测网络在无人机航拍图像识别中的应用［J］. 现代电子技术， 2025， 48（7）： 48-56.
	RAN X S， LIU S B. Application of detection network based on improved YOLOv8 in UAV aerial image recognition［J］. Modern Electronics Technique， 2025， 48（7）： 48-56 （in Chinese）.
[14]	谢明宏. 基于改进YOLOv8的无人机视角下小目标检测模型［J］. 信息技术与信息化， 2025（3）： 7-11.
	XIE M H. Small target detection model based on improved YOLOv8 from the perspective of UAV［J］. Information Technology and Informatization， 2025（3）： 7-11 （in Chinese）.
[15]	吴世琛，毛玉明，胡慧中，等. 基于改进YOLO11n的葡萄果叶病害检测方法［J/OL］. 农业工程学报， 1-8［2025-07-30］. .
	WU S C， MAO Y M， HU H Z， et al. Detecting grape leaf diseases using improved YOLO 11n［J/OL］. Transactions of the Chinese Society of Agricultural Engineering， 1-8 ［2025-07-30］. （in Chinese）.
[16]	LAU K W， PO L M， REHMAN Y A U. Large separable kernel attention： rethinking the large kernel attention design in CNN［J］. Expert Systems with Applications， 2024， 236： 121352.
[17]	向厚雪，许贵阳，张玉华，等. 基于改进YOLOv8的钢轨典型伤损智能识别算法［J］. 科学技术与工程， 2025， 25（18）： 7785-7792.
	XIANG H X， XU G Y， ZHANG Y H， et al. Intelligent identification algorithm for typical rail damage based on improved YOLOv8［J］. Science Technology and Engineering， 2025， 25（18）： 7785-7792 （in Chinese）.
[18]	刘子豪，张佳欣，薛　峰，等. 基于改进YOLO-v8的精密管件表面缺陷检测方法［J］. 浙江大学学报（工学版）， 2025， 59（7）： 1514-1522+1546.
	LIU Z H， ZHANG J X， XUE F， et al. Surface defect detection method of precision pipe fittings based on improved YOLO-v8［J］. Journal of Zhejiang University （Engineering Science）， 2025， 59（7）： 1514-1522+1546 （in Chinese）.
[19]	吴永泽，俞建峰，化春键，等. 基于改进YOLOv8n的圆柱电池壳多维缺陷检测技术研究［J］. 电子测量与仪器学报， 2024， 38（12）： 62-71.
	WU Y Z， YU J F， HUA C J， et al. Research on multi-dimensional defect detection technology for cylindrical battery shells based on improved YOLOv8n［J］. Journal of Electronic Measurement and Instrumentation， 2024， 38（12）： 62-71 （in Chinese）.