Solutions for the quality control of metallic surfaces are proposed. Specifically, we study a deflectometric apparatus based on coaxial structured light and the related algorithmic procedure, which is able to detect the faulty surface of a sample captured by a video sequence. First, by considering the metallic surface a dynamic scene illuminated under different light conditions, we develop the descriptor residuals of linear evolution of light (RLEL) that extracts the defectiveness information starting from the movement of the object without explicitly considering the physical characteristics of the light structure. Then, leveraging on RLEL, we present a hybrid learning (HL) technique capable of overcoming the data-driven approach used in classic deep learning (DL). By exploiting a multisteps training process, we combine the model-based descriptor RLEL and a classical data-driven convolutional neural network (CNN) to obtain an unconventional gray-box CNN, which exceeds the performance of popular DL solutions such as 3-D-inception and 3-D-residual DL networks. Remarkably, HL also shows its validity in comparing the performance of the same network structures trained not in a hybrid way, namely without the injection of the model-based information given by RLEL.
In this paper, an effective and novel automatic learning solution for the quality control of metallic objects surfaces is proposed, which can be seamlessly integrated into the industrial process. Such a system requires a coaxial illuminator to capture the object view with a single camera while lighting it with structured light: in this way, the object surface can be viewed in time as a dynamic scene under different illumination conditions. By relying on a linear model to describe the expected evolution of the light over the object of interest, the Residuals of Linear Evolution of Light (RLEL) algorithm is derived with the specific aim of identifying and characterizing anomalies and defects through the residuals of a least square approach. Then, a novel learning strategy is developed that exploits the model-based RLEL descriptor and thus promotes itself as an alternative strategy to the black box approach of Convolutional Neural Networks (CNNs). By combining both the data-driven and the model-based learning approaches to perform the inspection task, an Hybrid Learning (HL) procedure is defined: in a first phase, the HL exploits an Encoder-Decoder network to incorporate the model-based description while, in a second phase, it uses only the pre-trained encoder to drive the learning process of a 3D-CNN. In doing so, the proposed procedure reaches interesting results that exceed also the performance of state-of-the-art 3D-Inception and 3D-Residual networks.