The detection of product defects is essential in quality control in manufacturing. This study surveys stateoftheart deep-learning methods in defect detection. First, we classify the defects of products, such as electronic components, pipes, welded parts, and textile materials, into categories. Second, recent mainstream techniques and deep-learning methods for defects are reviewed with their characteristics, strengths, and shortcomings described. Third, we summarize and analyze the application of ultrasonic testing, filtering, deep learning, machine vision, and other technologies used for defect detection, by focusing on three aspects, namely method and experimental results. To further understand the difficulties in the field of defect detection, we investigate the functions and characteristics of existing equipment used for defect detection. The core ideas and codes of studies related to high precision, high positioning, rapid detection, small object, complex background, occluded object detection and object association, are summarized. Lastly, we outline the current achievements and limitations of the existing methods, along with the current research challenges, to assist the research community on defect detection in setting a further agenda for future studies.

Image classification has always been a hot research direction in the world, and the emergence of deep learning has promoted the development of this field. Convolutional neural networks (CNNs) have gradually become the mainstream algorithm for image classification since 2012, and the CNN architecture applied to other visual recognition tasks (such as object detection, object localization, and semantic segmentation) is generally derived from the network architecture in image classification. In the wake of these successes, CNN-based methods have emerged in remote sensing image scene classification and achieved advanced classification accuracy. In this review, which focuses on the application of CNNs to image classification tasks, we cover their development, from their predecessors up to recent state-of-the-art (SOAT) network architectures. Along the way, we analyze (1) the basic structure of artificial neural networks (ANNs) and the basic network layers of CNNs, (2) the classic predecessor network models, (3) the recent SOAT network algorithms, (4) comprehensive comparison of various image classification methods mentioned in this article. Finally, we have also summarized the main analysis and discussion in this article, as well as introduce some of the current trends.

We adopted actual intelligent production requirements and proposed a tiny part defect detection method to obtain a stable and accurate real-time tiny part defect detection system and solve the problems of manually setting conveyor speed and industrial camera parameters in defect detection for factory products. First, we considered the important influences of the properties of tiny parts and the environmental parameters of a defect detection system on its stability. Second, we established a correlation model between the detection capability coefficient of the part system and the moving speed of the conveyor. Third, we proposed a defect detection algorithm for tiny parts that are based on a single short detector network (SSD) and deep learning. Finally, we combined an industrial real-time detection platform with the missed detection algorithm for mechanical parts based on intermediate variables to address the problem of missed detections. We used a 0.8 cm darning needle as the experimental object. The system defect detection accuracy was the highest when the speed of the conveyor belt was 7.67 m/min.

This study proposes a modified convolutional neural network (CNN) algorithm that is based on dropout and the stochastic gradient descent (SGD) optimizer (MCNN-DS), after analyzing the problems of CNNs in extracting the convolution features, to improve the feature recognition rate and reduce the time-cost of CNNs. The MCNN-DS has a quadratic CNN structure and adopts the rectified linear unit as the activation function to avoid the gradient problem and accelerate convergence. To address the overfitting problem, the algorithm uses an SGD optimizer, which is implemented by inserting a dropout layer into the all-connected and output layers, to minimize cross entropy. This study used the datasets MNIST, HCL2000, and EnglishHand as the benchmark data, analyzed the performance of the SGD optimizer under different learning parameters, and found that the proposed algorithm exhibited good recognition performance when the learning rate was set to [0.05, 0.07]. The performances of WCNN, MLP-CNN, SVM-ELM, and MCNN-DS were compared. Statistical results showed the following: (1) For the benchmark MNIST, the MCNN-DS exhibited a high recognition rate of 99.97%, and the time-cost of the proposed algorithm was merely 21.95% of MLP-CNN, and 10.02% of SVM-ELM; (2) Compared with SVM-ELM, the average improvement in the recognition rate of MCNN-DS was 2.35% for the benchmark HCL2000, and the time-cost of MCNN-DS was only 15.41%; (3) For the EnglishHand test set, the lowest recognition rate of the algorithm was 84.93%, the highest recognition rate was 95.29%, and the average recognition rate was 89.77%.

With the rapid development of machine learning, its powerful function in the machine vision field is increasingly reflected. The combination of machine vision and robotics to achieve the same precise and fast grasping as that of humans requires high-precision target detection and recognition, location and reasonable grasp strategy generation, which is the ultimate goal of global researchers and one of the prerequisites for the large-scale application of robots. Traditional machine learning has a long history and good achievements in the field of image processing and robot control. The CNN (convolutional neural network) algorithm realizes training of large-scale image datasets, solves the disadvantages of traditional machine learning in large datasets, and greatly improves accuracy, thereby positioning CNNs as a global research hotspot. However, the increasing difficulty of labeled data acquisition limits their development. Therefore, unsupervised learning, self-supervised learning and reinforcement learning, which are less dependent on labeled data, have also undergone rapid development and achieved good performance in the fields of image processing and robot capture. According to the inherent defects of vision, this paper summarizes the research achievements of tactile feedback in the fields of target recognition and robot grasping and finds that the combination of vision and tactile feedback can improve the success rate and robustness of robot grasping. This paper provides a systematic summary and analysis of the research status of machine vision and tactile feedback in the field of robot grasping and establishes a reasonable reference for future research.

The complexity of the background and the similarities between different types of precision parts, especially in the high-speed movement of conveyor belts in complex industrial scenes, pose immense challenges to the object recognition of precision parts due to diversity in illumination. This study presents a real-time object recognition method for 0.8 cm darning needles and KR22 bearing machine parts under a complex industrial background. First, we propose an image data increase algorithm based on directional flip, and we establish two types of dataset, namely, real data and increased data. We focus on increasing recognition accuracy and reducing computation time, and we design a multilayer feature fusion network to obtain feature information. Subsequently, we propose an accurate method for classifying precision parts on the basis of non-maximal suppression, and then form an improved You Only Look Once (YOLO) V3 network. We implement this method and compare it with models in our real-time industrial object detection experimental platform. Finally, experiments on real and increased datasets show that the proposed method outperforms the YOLO V3 algorithm in terms of recognition accuracy and robustness.

Recent research has shown that the ubiquitous use of cameras and voice monitoring equipment in a home environment can raise privacy concerns and affect human mental health. This can be a major obstacle to the deployment of smart home systems for elderly or disabled care. This study uses a social robot to detect embarrassing situations. Firstly, we designed an improved neural network structure based on the You Only Look Once (YOLO) model to obtain feature information. By focusing on reducing area redundancy and computation time, we proposed a bounding-box merging algorithm based on region proposal networks (B-RPN), to merge the areas that have similar features and determine the borders of the bounding box. Thereafter, we designed a feature extraction algorithm based on our improved YOLO and B-RPN, called F-YOLO, for our training datasets, and then proposed a real-time object detection algorithm based on F-YOLO (RODA-FY). We implemented RODA-FY and compared models on our MAT social robot. Secondly, we considered six types of situations in smart homes, and developed training and validation datasets, containing 2580 and 360 images, respectively. Meanwhile, we designed three types of experiments with four types of test datasets composed of 960 sample images. Thirdly, we analyzed how a different number of training iterations affects our prediction estimation, and then we explored the relationship between recognition accuracy and learning rates. Our results show that our proposed privacy detection system can recognize designed situations in the smart home with an acceptable recognition accuracy of 94.48%. Finally, we compared the results among RODA-FY, Inception V3, and YOLO, which indicate that our proposed RODA-FY outperforms the other comparison models in recognition accuracy.

No abstract