The article addresses the application of Deep Learning (DL) methods and Computer Vision (CV) for the recognition and classification of surface defects in sheet metal. A specialized image preprocessing method has been developed, based on calculating statistical characteristics and subtracting them from averaged values. This method enhances defect detection, especially on low-contrast images, without requiring significant computational resources. The preprocessing algorithm includes considerations for image merging, the use of sliding windows with filters of various sizes, and image segmentation for input into the neural network. A defect detection model based on the Faster R-CNN Convolutional Neural Network (CNN) ensures high sensitivity in Automated Optical Inspection Systems (AOIS), detecting defects as small as 0.5×0.5 mm in real-time.

In recent years, many models based on deep learning have been proposed to accomplish the visual-inertial odometry (VIO) task, but these methods rely on combining visual and inertial data in each prediction. Processing visual data is considered computationally expensive, so this work will focus on relying on implementing VIO on inertial data in the first place, then using a reduced form of the visual data, and finally using the full form of the visual data, by training a policy network to control the selection of the visual data. Experimental results indicate a significant saving in the necessary computational power while maintaining close accuracy if the full form of visual data is used.

The paper focuses on using Deep Learning (DL) and Computer Vision (CV) techniques to detect surface defects in rolled metal products. By utilizing a Convolutional Neural Network (CNN), various surface defects can be detected and recognized, ultimately improving production standards and certification of the metal products. Two types of lighting, diffuse and side, are used to improve defect detection, and image preprocessing methods are employed to enhance the quality of the input data. The purpose of this work is to develop a method for recognizing and classifying defects in metal surfaces from their images in real time.

The design and development of surface defect detection and recognition systems for optical non-destructive testing (NDT) tasks is a complex, important and pressing problem today. Detection and classification of surface defects using Computer Vision (CV) and Machine Learning (ML) algorithms serves as an effective tool for production process control, quality management and increasing the profitability of enterprises. In this paper, Deep Learning (DL) and Computer Vision (CV) techniques are used to solve the problem of surface defect detection. Using Convolutional Neural Network (CNN), detection and recognition of various defects is carried out to improve production standards and process efficiency. The outcome of this paper is a comparative analysis of DL models and the selection of an algorithm designed to find and classify defects online. The application of such CNN models could allow the creation of a tool that considerably facilitates human work.

The paper describes an algorithm for localization and classification of quadcopter sensor (accelerometer and gyroscope) failures. Based on the obtained attributes; the conclusion is made about the serviceability of a quadcopter-type unmanned aerial vehicle (UAV). Two machine learning methods are used: logistic regression and random forest method. Their performance evaluated and compared using the data obtained by simulating the physical model of a quadcopter. A computer simulation of position sensor failure detection is presented. The proposed approach has shown its effectiveness and, therefore, it can be used to build a fault-tolerant adaptive motion control subsystem for UAVs.

This work is devoted to estimating the distance between moving cars to prevent dangerous traffic situations. The solution to this problem uses the approach of computer vision and calculating a depth map based on a stereoscopic pair. Using the Viola-Jones algorithm, the system detects vehicles while the vehicle is moving and calculates the distance to the objects moving in front, combining the received information with a depth map.