Automated Visual Inspection (AVI) systems for metal surface inspection is increasingly used in industries to aid human visual inspectors for classification of possible anomalies. For classification, the challenge lies in having a small and specific dataset that may easily result in over-fitting. As a solution, we propose to use deep Convolutional Neural Networks (ConvNets) learnt from the large ImageNet dataset [9] for image representations via transfer learning. Since a small dataset cannot be used to fine-tune a ConvNet due to overfitting, we also propose a Majority Voting Mechanism (MVM), which fuses the features extracted from the last three layers of ConvNets using Support Vector Machine (SVM) classifiers. This classification framework is effective where no prior knowledge of the best performing ConvNet layers is needed. This also allows flexibility in the choice of ConvNet used for feature extraction. The proposed method not only outperforms state-of-the-art traditional hand-crafted features in terms of classification but also obtains good results compared to other deep features with selected best layers on several anomaly and texture datasets.
In this paper, we propose a Phase Fourier Reconstruction (PFR) approach for anomaly detection on metal surfaces using salient irregularities. To get salient irregularity with images captured from an automatic visual inspection (AVI) system using different lighting settings, we first trained a classifier for image selection as only dark images are utilized for anomaly detection. By doing so, surface details, part design, and boundaries between foreground/background become indistinct, but anomaly regions are highlighted because of diffuse reflection caused by rough surfaces. Then PFR is applied so that regular patterns and homogeneous regions are further de-emphasized, and simultaneously, anomaly areas are distinct and located. Different from existing phase-based methods which require substantial texture information, our PFR works on both textual and non-textual images. Unlike existing template matching methods which require prior knowledge of defect-free patterns, our PFR is an unsupervised approach which detects anomalies using a single image. Experimental results on anomaly detection clearly demonstrate the effectiveness of the proposed method which outperforms several welldesigned methods [15], [8], [12], [18], [19], [16] with a running time of less than 0.01 seconds per image.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.