An Unsupervised-Learning-Based Approach for Automated Defect Inspection on Textured Surfaces

Automated Visual Defect Detection for Flat Steel Surface: A Survey

“…Moreover, AE-based algorithms also demonstrate strong competitiveness in steel surface defect detection, which are reported to be fairly noise-robust. Mei et al [114] utilized convolutional denoising AE network to reconstruct image patches, combined with the reconstruction residual maps, this scheme can reliably learning final detection results, where no manual intervention is needed throughout all the detection process. Youkachen et al [18] inventively applied convolutional auto-encoder (CAE) to reconstructed the defective test images, then the reconstructed images were used to highlight the shape feature by simple post-processing algorithms, providing another good application case on miscellaneous defect detection through unsupervised learning.…”

Visual-Based Defect Detection and Classification Approaches for Industrial Applications—A SURVEY

“…Based on their processing mechanics, these classifiers can be classified in two groups: (1) supervised; and (2) non-supervised or semi-supervised classifiers (see Table 3). [12,13,21,30,31,38,142,148,149,171,[192][193][194][195] Unsupervised/semi-Statistical/Novelty detection [58,65,86,[89][90][91][92]103,115,129,[196][197][198][199][200][201][202] supervised classifiers Gaussian mixture model [80,[203][204][205] Supervised classification methods incorporate the human model-as discussed in Section 3-where the application is searching for features of a predefined class. Detectable features are predefined and the classifier has to be previously trained to recognize them under supervision [40,65,90,103,142,[161][162][163].…”

“…Susan and Sharma [207] proposed a new unsupervised automated texture defect detection method that uses a Gaussian mixture entropy model to determine the optimal window size for feature extraction. Recently, Mei et al [205] developed an unsupervised learning based method by using only defect free samples for model training. The approach was carried out by reconstructing image patches with convolutional denoising autoencoder networks at different Gaussian pyramid levels, and synthesizing detection results from these different resolution channels.…”

A Review and Analysis of Automatic Optical Inspection and Quality Monitoring Methods in Electronics Industry

“…Table 7 summarizes the defects investigated in literature using AOI system for different FPD types. [250] Mura defects [251]- [256] TFT-LCD panel defects [218], [257], [258] TFT-LCD panel micro-defects such as pinholes, scratches, particles and fingerprints [29], [223], [259], [260] Polarising film defects [261]- [263] Glass substates defects in TFT-LCD [264] Defects during photolithography process [231], [232], [265]- [267] Backlight defects [212], [213], [268], [269] GE operation defects during TFT array process [270], [271] SD operation defects during TFT array process [272], [273] Color filter defects [214] TFT array defects such as fibre defect, particle defect, pattern damage, pattern residual and pattern scratch [274]- [277] Anisotropic Conductive Film defects [278] Optical thin film defects [279] TFT-LCD pad area defects [280] LCD surface deformation for smartphones [224], [225], [225]- [227] Polarizer transparent microdefect [230] Liquid resin defects [217] Subpixel (dots) functional defects OLED [235], [236] Directional textured surface defects in OLED and PLED …”

“…Moreover, deep learning-based methods can require considerable computational resources and time to perform training and inferencing [244]. Mei et al in [250] proposed an algorithm to detect texture surface defects in general such as LCD panels, ceramic tiles, and textiles. Where Mura defects is one of the LCD panel defects investigated in this paper.…”