• Journal of IKEEE
• /
• v.26 no.4
• /
• pp.639-646
• /
• 2022

In this paper, to solve the large dataset problem, we collect images through an image collection method called web crawling and build datasets for use in image classification models through a data preprocessing process. We also propose a lightweight model that can automatically classify images by adding category values by incorporating transfer learning into the image classification model and an image classification model that reduces training time and achieves high accuracy.

• Journal of IKEEE
• /
• v.23 no.1
• /
• pp.112-119
• /
• 2019

In this paper, we propose a deep learning structure suitable for embedded system. The flame detection process of the proposed deep learning structure consists of four steps : flame area detection using flame color model, flame image classification using deep learning structure for flame color specialization, $N{\times}N$ cell separation in detected flame area, flame image classification using deep learning structure for flame shape specialization. First, only the color of the flame is extracted from the input image and then labeled to detect the flame area. Second, area of flame detected is the input of a deep learning structure specialized in flame color and is classified as flame image only if the probability of flame class at the output is greater than 75%. Third, divide the detected flame region of the images classified as flame images less than 75% in the preceding section into $N{\times}N$ units. Fourthly, small cells divided into $N{\times}N$ units are inserted into the input of a deep learning structure specialized to the shape of the flame and each cell is judged to be flame proof and classified as flame images if more than 50% of cells are classified as flame images. To verify the effectiveness of the proposed deep learning structure, we experimented with a flame database of ImageNet. Experimental results show that the proposed deep learning structure has an average resource occupancy rate of 29.86% and an 8 second fast flame detection time. The flame detection rate averaged 0.95% lower compared to the existing deep learning structure, but this was the result of light construction of the deep learning structure for application to embedded systems. Therefore, the deep learning structure for flame detection proposed in this paper has been proved suitable for the application of embedded system.

• The Journal of the Convergence on Culture Technology
• /
• v.10 no.3
• /
• pp.909-914
• /
• 2024

This study suggests a quantitative evaluation of transfer learning, which is widely used in various AI fields, including image recognition for robot vision. Quantitative and qualitative analyses of results applying transfer learning are presented, but transfer learning itself is not discussed. Therefore, this study proposes a quantitative evaluation of transfer learning itself based on MNIST, a handwritten digit database. For the reference network, the change in recognition accuracy according to the depth of the transfer learning frozen layer and the ratio of transfer learning data and pre-training data is tracked. It is observed that when freezing up to the first layer and the ratio of transfer learning data is more than 3%, the recognition accuracy of more than 90% can be stably maintained. The transfer learning quantitative evaluation method of this study can be used to implement transfer learning optimized according to the network structure and type of data in the future, and will expand the scope of the use of robot vision and image analysis AI in various environments.

• Database Research
• /
• v.34 no.3
• /
• pp.45-57
• /
• 2018

The drowsiness that occurs in the driving is a very dangerous driver condition that can be directly linked to a major accident. In order to prevent drowsiness, there are traditional drowsiness detection methods to grasp the driver's condition, but there is a limit to the generalized driver's condition recognition that reflects the individual characteristics of drivers. In recent years, deep learning based state recognition studies have been proposed to recognize drivers' condition. Deep learning has the advantage of extracting features from a non-human machine and deriving a more generalized recognition model. In this study, we propose a more accurate state recognition model than the existing deep learning method by learning image and PPG at the same time to grasp driver's condition. This paper confirms the effect of driver's image and PPG data on drowsiness detection and experiment to see if it improves the performance of learning model when used together. We confirmed the accuracy improvement of around 3% when using image and PPG together than using image alone. In addition, the multimodal deep learning based model that classifies the driver's condition into three categories showed a classification accuracy of 96%.

• Journal of the Korean Society of Industry Convergence
• /
• v.23 no.3
• /
• pp.515-521
• /
• 2020

Deep learning methods have been effectively used to provide great improvement in various research fields such as machine learning, image processing and computer vision. One of the most frequently used deep learning methods in image processing is the convolutional neural networks. Compared to the traditional artificial neural networks, convolutional neural networks do not use the predefined kernels, but instead they learn data specific kernels. This property makes them to be used as feature extractors as well. In this study, we compared the quality of CNN features for traditional texture feature extraction methods. Experimental results demonstrate the superiority of the CNN features. Additionally, the recognition process and result of a pioneering CNN on MNIST database are presented.

• International Journal of Advanced Culture Technology
• /
• v.4 no.4
• /
• pp.23-29
• /
• 2016

In this paper, the learning technique for CNN processor on vehicle is proposed. In the case of conventional CNN processors, weighted values learned through training are stored for use, but when there is distortion in the image due to the weather conditions, the accuracy is decreased. Therefore, the method of enhancing the input image for classification is general, but it has the weakness of increasing the processor size. To solve this problem, the CNN performance was improved in this paper through the learning method of the distorted image. As a result, the proposed method showed improvement of approximately 38% better accuracy than the conventional method.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.06a
• /
• pp.98-101
• /
• 2019

At present, deep convolutional neural networks have made a very important contribution in single-image super-resolution. Through the learning of the neural networks, the features of input images are transformed and combined to establish a nonlinear mapping of low-resolution images to high-resolution images. Some previous methods are difficult to train and take up a lot of memory. In this paper, we proposed a simple and compact deep recursive residual network learning the features for single image super resolution. Global residual learning and local residual learning are used to reduce the problems of training deep neural networks. And the recursive structure controls the number of parameters to save memory. Experimental results show that the proposed method improved image qualities that occur in previous methods.

• Journal of Information Processing Systems
• /
• v.18 no.2
• /
• pp.197-207
• /
• 2022

The applications and user interfaces (UIs) of smart mobile devices are constantly diversifying. For example, deep learning can be an innovative solution to classify widgets in screen images for increasing convenience. To this end, the present research leverages captured images and the ReDraw dataset to write deep learning datasets for image classification purposes. First, as the validation for datasets using ResNet50 and EfficientNet, the experiments show that the dataset composed in this study is helpful for classification according to a widget's functionality. An implementation for widget detection and classification on RetinaNet and EfficientNet is then executed. Finally, the research suggests the Widg-C and Widg-D datasets-a deep learning dataset for identifying the widgets of smart devices-and implementing them for use with representative convolutional neural network models.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.17 no.1
• /
• pp.41-49
• /
• 2022

This paper proposes a novel lane detection algorithm based on inverse perspective transformation and machine learning in lightweight embedded system. The inverse perspective transformation method is presented for obtaining a bird's-eye view of the scene from a perspective image to remove perspective effects. This method requires only the internal and external parameters of the camera without a homography matrix with 8 degrees of freedom (DoF) that maps the points in one image to the corresponding points in the other image. To improve the accuracy and speed of lane detection in complex road environments, machine learning algorithm that has passed the first classifier is used. Before using machine learning, we apply a meaningful first classifier to the lane detection to improve the detection speed. The first classifier is applied in the bird's-eye view image to determine lane regions. A lane region passed the first classifier is detected more accurately through machine learning. The system has been tested through the driving video of the vehicle in embedded system. The experimental results show that the proposed method works well in various road environments and meet the real-time requirements. As a result, its lane detection speed is about 3.85 times faster than edge-based lane detection, and its detection accuracy is better than edge-based lane detection.

• Journal of the Korean Society of Manufacturing Process Engineers
• /
• v.20 no.12
• /
• pp.1-7
• /
• 2021

The deep learning image processing technique was used to prevent accidents in lathe work caused by worker negligence. During lathe operation, when the chuck is rotated, it is very dangerous if the operator's hand is near the chuck. However, if the chuck is stopped during operation, it is not dangerous for the operator's hand to be in close proximity to the chuck for workpiece measurement, chip removal or tool change. We used YOLO (You Only Look Once), a deep learning image processing program for object detection and classification. Lathe work images such as hand, chuck rotation and chuck stop are used for learning, object detection and classification. As a result of the experiment, object detection and class classification were performed with a success probability of over 80% at a confidence score 0.5. Thus, we conclude that the artificial intelligence deep learning image processing technique can be effective in preventing incidents resulting from worker negligence in future manufacturing systems.