铝型材是重要的工业基础材料，其表面异常检测是保证产品质量的必要环节。目前表面异常检测方法存在图像背景复杂、主体姿态不统一、纹理分布随机、异常样本数量少、异常类型不确定以及现有重构模型无法生成高质量正常特征图像等问题。针对以上问题，主要工作如下：

提出了一种基于自适应位置编码的分割一切模型（Segmented Everything Model with Adaptive Position Encoding, SAM-APE）的铝型材数据预处理方法，解决了图像背景复杂、主体姿态不统一、纹理分布随机为表面异常检测带来强烈干扰的问题。首先利用单通道提取、二值化、SAM-APE、Sobel算子边缘检测等方法对铝型材主体部位进行目标定位，得到一个最终掩码（Final Mask, F-Mask），利用该掩码替换掉原始图像中的复杂背景部分；然后利用图像的几何运算技术，对图像进行自适应旋转、裁剪操作，提取出铝型材主体部位；最后通过全变分正则化图像去噪的方法去除了铝型材表面欠规则分布的纹理干扰，减少冗余数据。

（2）提出了一种采用傅里叶卷积图像修复模型（Fourier Convolutional Image Restoration Model, FCIRM）的铝型材表面异常检测方法，解决了异常样本数量少、异常类型不确定的问题，并且引入傅里叶卷积，使得模型在网络浅层具有全局感受野，提高了图像修复的质量。首先建立了FCIRM，有效提取出正常的图像特征；然后通过对区域特征图和对应修复图像结构相似性值比对，判断输入图像表面是否为存在异常；最后和现有流行的有监督检测方法对比，实验结果表明：在阿里云天池大赛铝型材数据集上的精确率达96.4%，F1值至少提升了10.3%。

（3）提出了一种采用多头注意力图像修复模型（Multiple Attention Image Restoration Model, MAIRM）的铝型材表面异常检测方法，解决了现有重构模型无法生成高质量正常特征图像的问题。首先建立了MAIRM，并利用通过基于SAM-APE的数据预处理方法得到的铝型材无异常图像对其进行训练，采用多头注意力提高模型对铝型材全局有效数据特征的关注度，使模型能够更准确地提取铝型材表面有效特征；然后利用训练好的MAIRM对区域特征图进行重构，并对重构前后的图像进行结构相似性比对；最后进行了实验，并和几种常见基于重构模型提出的表面异常检测方法进行了对比，结果表明：所提方法精确率和F1值分别达97.6%和96.7%。

（4）设计并实现了铝型材表面异常检测实验系统，验证了所提表面异常检测方法的有效性。利用构建的表面异常检测模型实现了铝型材的自动检测功能，并对表面异常检测功能进行了测试。

综上所述，提出的铝型材表面异常自监督检测方法能够准确检测铝型材表面是否存在异常，具有较高的检测精度，能够满足铝型材表面异常检测的要求。

Aluminum profiles are crucial industrial base materials, and surface anomaly detection is essential to ensure product quality. Current methods for surface anomaly detection face challenges such as complex image backgrounds, inconsistent object postures, random texture distributions, limited anomaly samples, uncertain anomaly types, and the inability of existing reconstruction models to generate high-quality normal feature images. To address these issues, the primary work includes the following:
       （1）An aluminum profile data preprocessing method based on Segmented Everything Model with Adaptive Position Encoding (SAM-APE) is proposed to solve the problems of complex image background, inconsistent subject pose, and random texture distribution that bring strong surface anomaly detection. interference problem for surface anomaly detection. Firstly, single-channel extraction, binarization, SAM-APE, Sobel operator edge detection and other methods are used to target the main part of the aluminum profile, and a final mask (F-Mask) is obtained, which is used to replace the complex background part of the original image; then, using the geometric operation technique of the image, adaptive rotation and cropping operations are performed on the image, and the main part of the aluminum profile is extracted; then, using the geometric operation technique of the image, adaptive rotation and cropping operations are performed on the image, and the main part of the aluminum profile is extracted. The main part of the aluminum profile is extracted; finally, the texture interference which is irregularly distributed on the surface of the aluminum profile is removed by full variational regularization image denoising method to reduce redundant data.

A method of aluminum profile surface anomaly detection using Fourier Convolutional Image Restoration Model (FCIRM) is proposed, which solves the problems of small number of anomaly samples and uncertainty of anomaly type, and introduces Fourier convolution, which makes the model have a global sensing field in the shallow layer of the network, and The quality of image restoration is improved. Firstly, FCIRM is established to effectively extract normal image features; then, by comparing the similarity value of the regional feature map and the corresponding restored image structure, we determine whether the input image surface is the presence of anomalies; finally, comparing with the existing popular supervised detection methods, the experimental results show that the accuracy rate on the aluminum profile dataset of the AliCloud Tianchi Competition is up to 96.4%, and the F1 value has been improved by at least 10.3%.

A surface anomaly detection method for aluminum profiles using Multiple Attention Image Restoration Model (MAIRM) is proposed to solve the problem that the existing reconstruction models cannot generate high-quality normal feature images. Firstly, MAIRM is established and trained with the aluminum profile anomaly-free images obtained through SAM-APE-based data preprocessing method, which adopts multiple attention to improve the model's focus on the global valid data features of aluminum profiles, so that the model is able to extract the valid features of the surface of the aluminum profiles more accurately; and then, the region feature map is reconstructed with the trained MAIRM, and structural similarity comparisons are performed on the images before and after the reconstruction. Then the trained MAIRM is used to reconstruct the area feature map, and the structural similarity comparison of the images before and after the reconstruction is performed; finally, experiments are conducted and compared with several common surface anomaly detection methods proposed based on the reconstruction model, and the results show that: the accuracy rate and the F1 value of the proposed method reach 97.6% and 96.7%, respectively.

An experimental system for surface abnormality detection of aluminum profiles was designed and implemented to verify the effectiveness of the proposed surface abnormality detection method. The automatic detection function of aluminum profiles is realized using the constructed surface abnormality detection model, and the surface abnormality detection function is tested.

In summary, the proposed self-supervised detection method of aluminum profile surface abnormality can accurately detect whether there is abnormality on the surface of aluminum profile, has high detection accuracy, and can meet the requirements of aluminum profile surface abnormality detection.

[1] 侯文庆, 景会成, 周易林. 融合改进小波变换的小型磁瓦表面缺陷检测方法研究[J]. 制造业自动化, 2024, 46(01): 130-133+139.

[2] 左才, 张勇斌, 齐元胜, 等. 基于机器视觉的印刷品表面划痕缺陷检测[J]. 印刷与数字媒体技术研究, 2023, (05): 42-48.

[3] 王正家, 昝傲, 谷峰. 基于 Gabor 滤波的软包电池表面缺陷检测[J]. 组合机床与自动化加工技术, 2023, (11): 146-149.

[4] Xu B, Sun Y, Li J, et al. Accurate Detection for Zirconium Sheet Surface Scratches Based on Visible Light Images[J]. Sensors, 2023, 23(16): 1-16.

[5] Suo X, Zhang J, Liu J, et al. Anomaly Detection in Annular Metal Turning Surfaces Based on a Priori Information and a Multi-Scale Self-Referencing Template[J]. Sensors, 2023, 23(15): 6807-6827.

[6] Xu J, Hu X, Zhang Y, et al. BHE ‐ YOLO: Effective Small Target Detector for Aluminum Surface Defect Detection[J]. Advanced Theory and Simulations, 2024, 7(1): 1-13.

[7] Chen H, Du Y, Fu Y, et al. DCAM-Net: A rapid detection network for strip steel surface defects based on deformable convolution and attention mechanism[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 1-12.

[8] Zhao C, Shu X, Yan X, et al. RDD-YOLO: A modified YOLO for detection of steel surface defects[J]. Measurement, 2023, 214: 1-11.

[9] Lu J, Zhu M, Ma X, et al. Steel Strip Surface Defect Detection Method Based on Improved YOLOv5s[J]. Biomimetics, 2024, 9(1): 28-48.

[10] Xie Y, Yin B, Han X, et al. Improved YOLOv7-based steel surface defect detection algorithm[J]. Mathematical Biosciences and Engineering: MBE, 2022, 21(1): 346-368.

[11] Xu X, Li X. Research on surface defect detection algorithm of pipeline weld based on YOLOv7[J]. Scientific Reports, 2024, 14(1): 1881-1900.

[12] Cumbajin E, Rodrigues N, Costa P, et al. A Real-Time Automated Defect Detection System for Ceramic Pieces Manufacturing Process Based on Computer Vision with DeepLearning[J]. Sensors, 2023, 24(1): 232-253.

[13] Pu Q, Zhang H, Xu X, et al. Casting-DETR: An End-to-End Network for Casting Surface Defect Detection[J]. International Journal of Metalcasting, 2024: 1-14.

[14] Chen S, Zhou F, Gao G, et al. Unleashing the power of AI in detecting metal surface defects: an optimized YOLOv7-tiny model approach[J]. PeerJ Computer Science, 2024, 10: 1-28.

[15] Ye Q, Dong Y, Zhang X, et al. Robustness defect detection: Improving the performance of surface defect detection in interference environment[J]. Optics and Lasers in Engineering, 2024, 175: 1-11.

[16] Zhu J, Pang Q, Li S, et al. ADDet: An Efficient Multiscale Perceptual Enhancement Network for Aluminum Defect Detection[J]. IEEE Transactions on Instrumentation and Measurement, 2023，73：1-14.

[17] Li X, Zheng Y, Chen B, et al. Dual Attention-Based Industrial Surface Defect Detection with Consistency Loss[J]. Sensors, 2022, 22(14): 5141-5159.

[18] 董豪, 李少波, 杨静, 等. 基于特征融合与语义引导的药用胶囊表面缺陷检测[J]. 计算机集成制造系统, 2022, 8: 1-21.

[19] 王延舒, 余建波. 基于自适应全局定位算法的带钢表面缺陷检测[J]. 自动化学报, 2023, 45(1): 1-16.

[20] Zhang C, Dai W, Isoni V, et al. Automated anomaly detection for surface defects by dual generative networks with limited training data[J]. IEEE Transactions on Industrial Informatics, 2024, 20(1): 421- 431.

[21] Yao H, Luo W, Lou J, et al. Scalable Industrial Visual Anomaly Detection with Partial Semantics Aggregation Vision Transformer[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 73: 1-17.

[22] Rui J, Qiang N. Research on textile defects detection based on improved generative adversarial network[J]. Journal of Engineered Fibers and Fabrics, 2022, 17: 1-12.

[23] Guo Y, Zhong L, Qiu Y, et al. Using ISU-GAN for unsupervised small sample defect detection[J]. Scientific Reports, 2022, 12(1): 11604-11616.

[24] Lien C C, Chiu Y D. A Defect-Inspection System Constructed by Applying Autoencoderwith Clustered Latent Vectors and Multi-Thresholding Classification[J]. Applied Sciences, 2022, 12(4): 1883-1904.

[25] Szarski M, Chauhan S. An unsupervised defect detection model for a dry carbon fiber textile[J]. Journal of Intelligent Manufacturing, 2022, 33(7): 2075-2092.

[26] Yeung C C, Lam K M. Efficient fused-attention model for steel surface defect detection[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 1-11.

[27] Zhai H, Ma Z. Detection algorithm of rail surface defects based on multifeature saliency fusion method[J]. Sensor Review, 2022, 42(4): 402-411.

[28] Yang H, Zhu Z, Lin C, et al. Self-supervised Surface Defect Localization via Joint De-anomaly Reconstruction and Saliency-guided Segmentation[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 1-10.

[29] Xie X, Huang Y, Ning W, et al. RDAD: A reconstructive and discriminative anomaly detection model based on transformer[J]. International Journal of Intelligent Systems, 2022, 37(11): 8928-8946.

[30] Zhang R, Wang H, Feng M, et al. JRCC-Net: A Segmentation Network with Joint Representation and Contrast Clustering for Surface Anomaly Detection[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 1-12.

[31] Luo J, Lin J, Yang Z, et al. SMD anomaly detection: a self-supervised texture–structure anomaly detection framework[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 1-11.

[32] Tao X, Adak C, Chun P J, et al. ViTALnet: Anomaly on Industrial Textured Surfaces With Hybrid Transformer[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 1-13.

[33] Huo D, Wang J, Qian Y, et al. Glass segmentation with RGB-thermal image pairs[J]. IEEE Transactions on Image Processing, 2023, 32: 1911-1926.

[34] Song K, Bao Y, Wang H, et al. A Potential Vision-Based Measurements Technology: Information Flow Fusion Detection Method Using RGB-Thermal Infrared Images[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 589-596.

[35] Huo D, Wang J, Qian Y, et al. Glass segmentation with RGB-thermal image pairs[J]. IEEE Transactions on Image Processing, 2023, 32: 1911-1926.

[36] Zhu X, Zhu Y, Wang H, et al. Skeleton sequence and RGB frame based multi-modality feature fusion network for action recognition[J]. ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM), 2022, 18(3): 1-24.

[37] Zhang F, Jiang X, Xia Z, et al. Non-Local Color Compensation Network for Intrinsic Image Decomposition[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2022, 33(1): 132-145.

[38] Liu C, Shu X, Pan L, et al. Multi-Scale Underwater Image Enhancement in RGB and HSV Color Spaces[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 502-534.

[39] Moreira G, Magalhães S A, Pinho T, et al. Benchmark of deep learning and a proposed hsv colour space models for the detection and classification of greenhouse tomato[J]. Agronomy, 2022, 12(2): 356-368.

[40] Zhang Y, Di X, Zhang B, et al. Better than reference in low-light image enhancement: conditional re-enhancement network[J]. IEEE Transactions on Image Processing, 2021, 31: 759-772.

[41] 张大锦, 刘辉, 陈甫刚, 等. 频域多方向C-UNet及动态损失的工业烟尘图像分割[J]. 控制理论与应用, 2023, 1-12.

[42] Luo T, Li S, Li J, et al. Image Fuzzy Edge Information Segmentation Based on Computer Vision and Machine Learning[J]. Journal of Grid Computing, 2023, 21(4): 56-71.

[43] Watanabe Y, Togo R, Maeda K, et al. Text-guided Image Manipulation via Generative Adversarial Network with Referring Image Segmentation-based Guidance[J]. IEEE Access, 2023, 11: 42534-42545.

[44] 彭家磊, 黄成泉, 雷欢, 等. 自适应空间强度约束和 KL 信息的模糊 C 均值彩色噪声图像分割[J]. 控制与决策 2023, 12: 1-9.

[45] Gao P, Song Y, Song M, et al. Extract nanoporous gold ligaments from SEM images by combining fully convolutional network and Sobel operator edge detection algorithm[J]. Scripta Materialia, 2022, 213: 114-127.

[46] Zhang H, Cheng S, Zhao Y, et al. Measurement of yarn apparent evenness based on modified Canny edge detection[J]. The Journal of The Textile Institute, 2023: 8(25) 1-7.

[47] Bakurov I, Buzzelli M, Schettini R, et al. Full-Reference Image Quality Expression via Genetic Programming[J]. IEEE Transactions on Image Processing, 2023, 32: 1458-1473.

[48] 刘建欣, 潘如如, 周建. 基于欠完备字典重构的无监督织物疵点检测方法[J]. 上海交通大学学报, 2023, 7: 1-16.

[49] 马敏, 郭鑫, 于洁. 改进正则化半阈值算法的ECT图像重建[J]. 仪器仪表学报, 2022, 43(05): 110-119.

[50] Rudin L I, Osher S, Fatemi E. Nonlinear total variation based noise removal algorithms[J]. Physica D: nonlinear phenomena, 1992, 60(1-4): 259-268.

[51] Teng S, Liu Z, Luo W, et al. Bridge anomaly detection based on reconstruction error and structural similarity of unsupervised convolutional auto-encoder[J]. Structural Health Monitoring, 2023: 11(04): 1-16.

[52] Khan S, Naseer M, Hayat M, et al. Transformers in vision: A survey[J]. ACM computing surveys (CSUR), 2022, 54(10s): 1-41.

[53] Zhuang X, Liu F, Hou J, et al. Modality attention fusion model with hybrid multi-head self-attention for video understanding[J]. Plos one, 2022, 17(10): 355-368.

[54] Kirillov A, Mintun E, Ravi N, et al. Segment anything[J]. arXiv preprint arXiv:2304.02643, 2023.

[55] Chang Q, Li X, Zhao Y. Reversible data hiding for color images based on adaptive three-dimensional histogram modification[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2022, 32(9): 5725-5735.

[56] Chen R, Cai D, Hu X, et al. Defect detection method of aluminum profile surface using deep self-attention mechanism under hybrid noise conditions[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 70: 1-9.

[57] Suvorov R, Logacheva E, Mashikhin A, et al. Resolution-robust large mask inpainting with fourier convolutions[C]//Proceedings of the IEEE/CVF winter conference on applications of computer vision. 2022: 2149-2159.

[58] Chi L, Jiang B, Mu Y. Fast fourier convolution[J]. Advances in Neural Information Processing Systems, 2020, 33: 4479-4488.

[59] Pukelsheim F. The three sigma rule[J]. The American Statistician, 1994, 48(2): 88-91.

[60] Jiang L, Yuan B, Wang Y, et al. MA-YOLO: A Method for Detecting Surface Defects of Aluminum Profiles with Attention Guidance[J]. IEEE Access, 2023, 11: 71269-71286.

[61] Liu J, Song K, Feng M, et al. Semi-supervised anomaly detection with dual prototypes autoencoder for industrial surface inspection[J]. Optics and Lasers in Engineering, 2021, 136: 1-9.