This repository consists of recent state-of-the-art deep learning networks for industrial machine vision application. The transformation of manufacturing system towards the intelligent manufacturing focuses on automation and the use of advanced technologies such as AI with robots and advanced machines for greater efficiency and precision. The AI system will allow for an optimized production process, smart decisions, real-time information, preventive maintenance, and self-prognosis of the production processes. With availability of big data and advanced computing equipment, and technologies, the deep learning application has been one of the highly researched areas in the scientific world in the past few years. Promoting this applications, this repositoty presents recent influential works related to deep learning applications on the area of anomaly detection, and other industrial machine vision applications.
Industrial anomaly detection is a critical component of modern industrial processes that involve the monitoring and analysis of data to identify abnormal behavior or deviations from expected patterns within industrial systems. Although various anomalies can be investigated, this repository presents deep learning application for surface anomaly detection for industrial products. Most of the methods presented uses image datasets to identify defective or anomolous parts of the product.
Defect classification, detection, and segmentation are important tasks in various industries, particularly in manufacturing and quality control processes. These tasks involve identifying and categorizing defects in products or materials. Deep learning-based defect classification involves identifying types of defects in a product or simply identifying wether a product is defective or not. Detection involves localization and classification of defects, while defect segmetnation involves identification and localization of defects at a pixel-level. Recent state-of-the-art methods involving these taks are presented in this repository.
Semi-supervised learning and weakly supervised learning are two approaches to machine learning that address scenarios where obtaining fully labeled training data is challenging or expensive. Hence, this repository presents state-of-the-art semi-supervised and weakly-supervised methods proposed for the task of intelligent industrial inspection.
| Year | Title/Source | Journal/Conference | Code |
|---|---|---|---|
| 2023 | Uncertainty-aware and dynamically-mixed pseudo-labels for semi-supervised defect segmentation | Computers in Industry | Pytorch |
| 2022 | A knowledge distillation-based multi-scale relation-prototypical network for cross-domain few-shot defect classification | Journal of Intelligent Manufacturing | Not available |
| 2022 | Semisupervised Defect Segmentation With Pairwise Similarity Map Consistency and Ensemble-Based Cross Pseudolabels | IEEE TII | Pytorch |
| 2019 | Unsupervised weld defect classification in radiographic images using multivariate generalized Gaussian mixture model with exact computation of mean and shape parameters | Computers in Industry | Not available |