Due to the immaturity of printing technology, printing defect is an inevitable problem in industrial production. It will greatly affect the appearance and performance of the product if defects could not be found and dealt with in time. Defect detection often requires high real-time performance and registration rate. Especially in the case of angular shift or image translation, the algorithm needs to have good registration effect. Feature matching is an important part of image registration. In the actual detection process, the speed and quality of matching directly affect the results of image registration algorithm. Based on deep understanding of SIFT, SURF, ORB and GMS, in the case of different lighting environments, different shooting angles, scale changes and fuzzy images, the superposition effect, registration time, number of feature point pairs and registration rate of four typical feature matching algorithms above are compared on the open data set respectively. The robustness, speed, advantages and disadvantages as well as applicable conditions of those four algorithms are analyzed and summarized. And an appropriate algorithm is selected based on the actual defect detection task. Experimental results show that ORB algorithm has the characteristics of fast speed, high registration rate and strong robustness in different environments. So, it is adopted in the actual defect detection task. Actual defect detection results show that ORB algorithm can not only accurately frame the defect area of printed matter, but also well meet the real-time production requirements of defect detection.
With the rise of the new generation of artificial intelligence technology, the object detection method based on deep learning has achieved remarkable results. In this paper, the detection accuracy of three popular object detection algorithms such as You Only Look Once (YOLO V3), Region-CNN (Faster R-CNN) and Single Shot MultiBox Detector (SSD) has been compared. Aiming at the actual detection problems of building block parts with irregular shape and different sizes, a method that combines deep convolutional generative adversarial networks (DCGAN) with deep learning based object detection algorithm is proposed to solve the problems of over fitting or weak generalization ability in the case of limited datasets, and to improve the detection accuracy of object detection algorithm. Experimental results show that: 1. Using public datasets, when the training data is reduced, the mean average precision (mAP) values of the above three algorithms are reduced respectively. Among those, SSD algorithm has the smallest change, which decreases 7.81%. 2. The control variable method is used to train the building block parts. In the case of insufficient training data, the detection accuracy of three object detection algorithms is low. 3. After combining SSD algorithm with DCGAN algorithm and applying it into the detection task of building block parts, the mAP value is improved from 79.63% to 83.32%, and the detection accuracy is obviously improved.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.