skip to main content

Title: Visual coating inspection framework via self-labeling and multi-stage deep learning strategies

An instantaneous and precise coating inspection method is imperative to mitigate the risk of flaws, defects, and discrepancies on coated surfaces. While many studies have demonstrated the effectiveness of automated visual inspection (AVI) approaches enhanced by computer vision and deep learning, critical challenges exist for practical applications in the manufacturing domain. Computer vision has proven to be inflexible, demanding sophisticated algorithms for diverse feature extraction. In deep learning, supervised approaches are constrained by the need for annotated datasets, whereas unsupervised methods often result in lower performance. Addressing these challenges, this paper proposes a novel deep learning-based automated visual inspection (AVI) framework designed to minimize the necessity for extensive feature engineering, programming, and manual data annotation in classifying fuel injection nozzles and discerning their coating interfaces from scratch. This proposed framework comprises six integral components: It begins by distinguishing between coated and uncoated nozzles through gray level co-occurrence matrix (GLCM)-based texture analysis and autoencoder (AE)-based classification. This is followed by cropping surface images from uncoated nozzles, and then building an AE model to estimate the coating interface locations on coated nozzles. The next step involves generating autonomously annotated datasets derived from these estimated coating interface locations. Subsequently, a convolutional neural network (CNN)-based detection model is trained to accurately localize the coating interface locations. The final component focuses on enhancing model performance and trustworthiness. This framework demonstrated over 95% accuracy in pinpointing the coating interfaces within the error range of ± 6 pixels and processed at a rate of 7.18 images per second. Additionally, explainable artificial intelligence (XAI) techniques such as t-distributed stochastic neighbor embedding (t-SNE) and the integrated gradient substantiated the reliability of the models.

more » « less
Author(s) / Creator(s):
; ; ; ;
Publisher / Repository:
Springer Science + Business Media
Date Published:
Journal Name:
Journal of Intelligent Manufacturing
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Missions to small celestial bodies rely heavily on optical feature tracking for characterization of and relative navigation around the target body. While deep learning has led to great advancements in feature detection and description, training and validating data-driven models for space applications is challenging due to the limited availability of large-scale, annotated datasets. This paper introduces AstroVision, a large-scale dataset comprised of 115,970 densely annotated, real images of 16 different small bodies captured during past and ongoing missions. We leverage AstroVision to develop a set of standardized benchmarks and conduct an exhaustive evaluation of both handcrafted and data-driven feature detection and description methods. Next, we employ AstroVision for end-to-end training of a state-of-the-art, deep feature detection and description network and demonstrate improved performance on multiple benchmarks. The full benchmarking pipeline and the dataset will be made publicly available to facilitate the advancement of computer vision algorithms for space applications. 
    more » « less
  2. Abstract

    Vision science, particularly machine vision, has been revolutionized by introducing large-scale image datasets and statistical learning approaches. Yet, human neuroimaging studies of visual perception still rely on small numbers of images (around 100) due to time-constrained experimental procedures. To apply statistical learning approaches that include neuroscience, the number of images used in neuroimaging must be significantly increased. We present BOLD5000, a human functional MRI (fMRI) study that includes almost 5,000 distinct images depicting real-world scenes. Beyond dramatically increasing image dataset size relative to prior fMRI studies, BOLD5000 also accounts for image diversity, overlapping with standard computer vision datasets by incorporating images from the Scene UNderstanding (SUN), Common Objects in Context (COCO), and ImageNet datasets. The scale and diversity of these image datasets, combined with a slow event-related fMRI design, enables fine-grained exploration into the neural representation of a wide range of visual features, categories, and semantics. Concurrently, BOLD5000 brings us closer to realizing Marr’s dream of a singular vision science–the intertwined study of biological and computer vision.

    more » « less
  3. A major challenge for online learning is the inability of systems to support student emotion and to maintain student engagement. In response to this challenge, computer vision has become an embedded feature in some instructional applications. In this paper, we propose a video dataset of college students solving math problems on the educational platform with a front facing camera collecting visual feedback of student gestures. The video dataset is annotated to indicate whether students’ attention at specific frames is engaged or wandering. In addition, we train baselines for a computer vision module that determines the extent of student engagement during remote learning. Baselines include state-of-the-art deep learning image classifiers and traditional conditional and logistic regression for head pose estimation. We then incorporate a gaze baseline into the MathSpring learning platform, and we are evaluating its performance with the currently implemented approach. 
    more » « less
  4. Deep neural networks are efficient learning machines which leverage upon a large amount of manually labeled data for learning discriminative features. However, acquiring substantial amount of supervised data, especially for videos can be a tedious job across various computer vision tasks. This necessitates learning of visual features from videos in an unsupervised setting. In this paper, we propose a computationally simple, yet effective, framework to learn spatio-temporal feature embedding from unlabeled videos. We train a Convolutional 3D Siamese network using positive and negative pairs mined from videos under certain probabilistic assumptions. Experimental results on three datasets demonstrate that our proposed framework is able to learn weights which can be used for same as well as cross dataset and tasks. 
    more » « less
  5. This study describes a hybrid framework for post-hazard building performance assessments. The framework relies upon rapid imaging data collected by regional scout teams being integrated into broader data platforms that are parsed by virtual teams of hazards engineers to efficiently create robust performance assessment datasets. The study also pilots a machine-in-the-loop approach whereby deep learning and computer vision-based models are used to automatically define common building attributes, enabling hazard engineers to focus more of their efforts on precise damage quantification and other more nuanced elements of performance assessments. The framework shows promise, but to achieve optimal accuracy of the automated methods requires regional tuning. 
    more » « less