Generative adversarial network with object detector discriminator for enhanced defect detection on ultrasonic B-scans

Application of Model Data for Training the Classifier of Defects in Rail Bolt Holes in Ultrasonic Diagnostics

“…At the same time, the accumulated data set does not contain the same number of samples of each class of defects, that is, it is characterized by its imbalance, since most of it is made up of frequently occurring images of structural elements with no defects. This imbalance of the data set when using deep learning algorithms as a leading approach in the field of pattern recognition leads to a decrease in the forecast accuracy of synthesized neural networks for classes with defects [14,16,20]. Increasing the data set using field experiments with artificially created physical models of defects in rails is notable for its laboriousness and high cost.…”

Utilizing Generative Adversarial Networks for Stable Structure Generation in Angry Birds

“…The central concept behind applying GANs is to define the given task as a game between two opposing systems, which are then trained in an adversarial manner to reach a zero-sum Nash equilibrium (Moghadam et al 2021). While GANs have been primarily used for image synthesis (Goodfellow et al 2014;Radford, Metz, and Chintala 2016;Liu and Tuzel 2016;Karras et al 2017;Huang, Yu, and Wang 2018), they have also been successfully applied to many other applications including object detection (Prakash and Karam 2021;Posilović et al 2021), natural language processing (Subramanian et al 2017), audio enhancement (Torres-Reyes and Latifi 2019; Biswas and Jia 2020), anomaly detection (Schlegl et al 2017;Xia et al 2020), and, most relevant for this paper, video game level generation.…”

72‐4: Invited Paper: Synthetic Defect Generation for Display Front‐of‐Screen Quality Inspection: A Survey

“…Table 1 provides a summary of those methods and their applications. Ö Wafer [2,4] Ö Assembly and test [5] Ö Ö Conveyor belt [17] Ö Electrical machines [18] Ö Solar cells [19] Ö Non-destructive testing [20,21] Ö…”

“…The defect detection algorithm performance (i) with and without synthetic data; (ii) with synthetic data generated from different methods, are usually compared. In defect detection applications, the indirect quantitative evaluation metrics can be further categorized into three different levels of accuracy, including instance level (detect/classify the defective samples [2,4,19,22,24,25,27,28,31]), localization level (detect the defect location inside each sample [17,21]), and pixel-wise level (extract the defect segmentation mask [29]). For example, Xiong et al…”