http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Convolutional neural network based surface inspection system for non-patterned welding defects
박제강,Woo-Hyun An,강동중 한국정밀공학회 2019 International Journal of Precision Engineering and Vol.20 No.3
In this paper, we propose a convolutional neural network (CNN) based method that inspects non-patterned welding defects (craters, pores, foreign substances and fissures) on the surface of the engine transmission using a single RGB camera. The proposed method consists of two steps: first, extracting the welding area to be inspected from the captured image, and then determining whether the extracted area includes defects. In the first step, to extract the welding area from the captured image, a CNN based approach is proposed to detect a center of the engine transmission in the image. In the second stage, the extracted area is identified by another CNN as defective or non-defective. To train the second stage CNN stably, we propose a class-specific batch sampling method. With our sampling method, biased learning caused by data imbalance (number of collected defective images is much less than that of non-defective images) is effectively prevented. We evaluated our system with a large amount of samples (about 32,000 images) collected manually from the production line, and our system shows a remarkable performance in all experiments.
Machine Learning-Based Imaging System for Surface Defect Inspection
박제강,강동중,권배근,박준협 한국정밀공학회 2016 International Journal of Precision Engineering and Vol.3 No.3
Modern inspection systems based on smart sensor technology like image processing and machine vision have been widely spread into several fields of industry such as process control, manufacturing, and robotics applications in factories. Machine learning for smart sensors is a key element for the visual inspection of parts on a product line that has been manually inspected by people. This paper proposes a method for automatic visual inspection of dirties, scratches, burrs, and wears on surface parts. Imaging analysis with CNN (Convolution Neural Network) of training samples is applied to confirm the defect's existence in the target region of an image. In this paper, we have built and tested several types of deep networks of different depths and layer nodes to select adequate structure for surface defect inspection. A single CNN based network is enough to test several types of defects on textured and non-textured surfaces while conventional machine learning methods are separately applied according to type of each surface. Experiments for surface defects in real images prove the possibility for use of imaging sensors for detection of different types of defects. In terms of energy saving, the experiment result shows that proposed method has several advantages in time and cost saving and shows higher performance than traditional manpower inspection system.
Multi-task Convolutional Neural Network System for License Plate Recognition
김홍현,박제강,오주희,강동중 제어·로봇·시스템학회 2017 International Journal of Control, Automation, and Vol.15 No.6
License plate recognition is an active research field as demands sharply increase with the developmentof Intelligent Transportation System (ITS). However, since the license plate recognition(LPR) is sensitive to theconditions of the surrounding environment such as a complicated background in the image, viewing angle andillumination change, it is still difficult to correctly recognize letters and digits on LPR. This study applies DeepConvolutional Neural Network (DCNN) to the license plate recognition. The DCNN is a method of which theperformance has recently been proven to have an excellent generalization error rate in the field of image recognition. The proposed layer structure of the DCNN used in this study consists of a combination of a layer for judging theexistence of a license plate and a layer for recognizing digits and characters. This learning method is based on Multi-Task Learning (MTL). Through experiments using real images, this study shows that this layer structure classifiesdigits and characters more accurately than the DCNN using a conventional layer does. We also use artificial imagesgenerated directly for training model.
민정원(Jeong-Won Min),박제강(Je-Kang Park),강동중(Dong-Joong Kang) 제어로봇시스템학회 2017 제어·로봇·시스템학회 논문지 Vol.23 No.6
In this paper, we propose a method to detect the region of the traffic light using the Convolution Neural Network. Since the existing methods detect the traffic light based on the color component from pixel by pixel operation, the false detection rate and the white noise occurrence frequency are high for pixels having similar color components. On the other hand, the proposed method learns a convolution filter that performs region to pixel operation. Since the shape of the traffic light and the surrounding color components are also reflected in the learning, it is possible to reduce the white noise and detect a region relatively close to the object blob. In order to verify the proposed method, we performed a comparative experiment with the existing methods under the same conditions and the performance of the proposed method is much higher than that of the conventional methods.
박용규(Young-Kyu Park),박제강(Je-Kang Park),온한익(Han-Ik On),강동중(Dong-Joong Kang) 제어로봇시스템학회 2015 제어·로봇·시스템학회 논문지 Vol.21 No.11
This paper proposes a method for detecting the front side of vehicles. The method can find the car side with a license plate even with complicated and cluttered backgrounds. A convolutional neural network (CNN) is used to solve the detection problem as a unified framework combining feature detection, classification, searching, and localization estimation and improve the reliability of the system with simplicity of usage. The proposed CNN structure avoids sliding window search to find the locations of vehicles and reduces the computing time to achieve real-time processing. Multiple responses of the network for vehicle position are further processed by a weighted clustering and probabilistic threshold decision method. Experiments using real images in parking lots show the reliability of the method.