Industrial object and defect recognition utilizing multilevel feature extraction from industrial scenes with Deep Learning approach

Modern industry requires modern solutions for monitoring the automatic production of goods and detecting defected materials. Smart monitoring of the functionality of the mechanical parts of technology systems or machines is a mandatory step towards automatic production. Deep Learning has proven its efficiency in feature extraction from images, videos and text, thereby succeeding in various object detection, recognition, segmentation and classification tasks. Despite its advances, little has been investigated about the effectiveness of specially designed Convolutional Neural Networks (CNNs) for defect detection and industrial object recognition. In the particular study, we employed six publicly available industrial-related image datasets, containing defected materials and industrial tools, or engine parts, aiming to develop a specialized model to classify them. Motivated by the success of the Virtual Geometry Group (VGG) network, we propose a modified version of it, called Multipath VGG19, which allows for extra local and global feature extraction (multi-level feature extraction) by making use of several processing paths. The extra features are fused via concatenation. The experiments verified the effectiveness of MVGG19 over the baseline VGG19. Specifically, top classification performance was achieved in five of the six image datasets, whilst the average classification improvement was 6.95%. MVGG19 also showed better overall stability and robustness to dataset variation, compared to other baseline state-of-the-art CNNs.