• Journal of information and communication convergence engineering
• /
• v.18 no.2
• /
• pp.75-81
• /
• 2020

Approximately 1.3 million people die from traffic accidents each year, and smartphone usage while driving is one of the main causes of such accidents. Therefore, detection of smartphone usage by drivers has become an important part of distracted driving detection. Previous studies have used single camera-based methods to collect the driver images. However, smartphone usage detection by employing a single camera can be unsuccessful if the driver occludes the phone. In this paper, we present a driver smartphone usage detection system that uses multiple cameras to collect driver images from different perspectives, and then processes these images with ensemble convolutional neural networks. The ensemble method comprises three individual convolutional neural networks with a simple voting system. Each network provides a distinct image perspective and the voting mechanism selects the final classification. Experimental results verified that the proposed method avoided the limitations observed in single camera-based methods, and achieved 98.96% accuracy on our dataset.

• Journal of Semiconductor Engineering
• /
• v.1 no.1
• /
• pp.31-37
• /
• 2020

We present an approximate DRAM architecture for energy-efficient deep learning. Our key premise is that by bounding memory errors to non-critical information, we can significantly reduce DRAM refresh energy without compromising recognition accuracy of deep neural networks. To validate the key premise, we make extensive Monte-Carlo simulations for several well-known convolutional neural networks such as LeNet, ConvNet and AlexNet with the input of MINIST, CIFAR-10, and ImageNet, respectively. We assume that the highest-order 8-bits (in single precision) and 4-bits (in half precision) are protected from retention errors under the proposed architecture and then, randomly inject bit-errors to unprotected bits with various bit-error-rates. Here, recognition accuracies of the above convolutional neural networks are successfully maintained up to the 10^-5-order bit-error-rate. We simulate DRAM energy during inference of the above convolutional neural networks, where the proposed architecture shows the possibility of considerable energy saving up to 10 ~ 37.5% of total DRAM energy.

• Nuclear Engineering and Technology
• /
• v.55 no.6
• /
• pp.2276-2282
• /
• 2023

We deploy artificial neural networks to unfold neutron spectra from measured energy-integrated quantities. These neutron spectra represent an important parameter allowing to compute the absorbed dose and the kerma to serve radiation protection in addition to nuclear safety. The built architectures are inspired from convolutional neural networks. The first architecture is made up of residual transposed convolution's blocks while the second is a modified version of the U-net architecture. A large and balanced dataset is simulated following "realistic" physical constraints to train the architectures in an efficient way. Results show a high accuracy prediction of neutron spectra ranging from thermal up to fast spectrum. The dataset processing, the attention paid to performances' metrics and the hyper-optimization are behind the architectures' robustness.

• Journal of Korea Multimedia Society
• /
• v.24 no.8
• /
• pp.1020-1025
• /
• 2021

Facial alignment is very important task for human life. And facial landmark detection is one of the instrumental methods in face alignment. We introduce the stacked hourglass networks with transposed convolutional layers for facial landmark detection. our method substitutes nearest neighbor upsampling for transposed convolutional layer. Our method returns better accuracy in facial landmark detection compared to stacked hourglass networks with nearest neighbor upsampling.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.22 no.5
• /
• pp.728-733
• /
• 2018

Deep learning has been actively studied for predicting protein secondary structure based only on the sequence information of the amino acids constituting the protein. In this paper, we compared the performances of the convolutional neural networks of various structures to predict the protein secondary structure. To investigate the optimal depth of the layer of neural network for the prediction of protein secondary structure, the performance according to the number of layers was investigated. We also applied the structure of GoogLeNet and ResNet which constitute building blocks of many image classification methods. These methods extract various features from input data, and smooth the gradient transmission in the learning process even using the deep layer. These architectures of convolutional neural networks were modified to suit the characteristics of protein data to improve performance.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.17 no.5
• /
• pp.97-102
• /
• 2017

Recently deep learning techniques such as convolutional neural networks (CNN) have been introduced to classify high-resolution remote sensing data. In this paper, we investigated the possibility of applying CNN to crop classification of farmland images captured by drones. The farming area was divided into seven classes: rice field, sweet potato, red pepper, corn, sesame leaf, fruit tree, and vinyl greenhouse. We performed image pre-processing and normalization to apply CNN, and the accuracy of image classification was more than 98%. With the output of this study, it is expected that the transition from the existing image classification methods to the deep learning based image classification methods will be facilitated in a fast manner, and the possibility of success can be confirmed.

• Journal of Korea Multimedia Society
• /
• v.25 no.11
• /
• pp.1643-1652
• /
• 2022

Most of the existing machine reading research has used Recurrent Neural Network (RNN) and Convolutional Neural Network (CNN) algorithms as networks. Among them, RNN was slow in training, and Question Answering Network (QANet) was announced to improve training speed. QANet is a model composed of CNN and self-attention. CNN extracts semantic and syntactic information well from the local corpus, but there is a limit to extracting the corresponding information from the global corpus. Graph Convolutional Networks (GCN) extracts semantic and syntactic information relatively well from the global corpus. In this paper, to take advantage of this strength of GCN, we propose I-QANet, which changed the CNN of QANet to GCN. The proposed model performed 1.2 times faster than the baseline in the Stanford Question Answering Dataset (SQuAD) dataset and showed 0.2% higher performance in Exact Match (EM) and 0.7% higher in F1. Furthermore, in the Korean Question Answering Dataset (KorQuAD) dataset consisting only of Korean, the learning time was 1.1 times faster than the baseline, and the EM and F1 performance were also 0.9% and 0.7% higher, respectively.

• The Journal of Korean Institute of Information Technology
• /
• v.17 no.6
• /
• pp.85-94
• /
• 2019

Warhead and debris show the different micro-Doppler frequency shape in the spectrogram because of the different micro motion. So we can classify them using the micro-Doppler features. In this paper, we classified warhead and debris in the separation phase using CNN(Convolutional Neural Networks). For the input image of CNN, we used micro-Doppler spectrogram. In addition, to improve classification performance of warhead and debris, we applied the preprocessing using CA-CFAR to the micro-Doppler spectrogram. As a result, when the preprocessing of micro-Doppler spectrogram was used, classification performance is improved in all signal-to-noise ratio(SNR).

• The Journal of the Acoustical Society of Korea
• /
• v.37 no.2
• /
• pp.118-128
• /
• 2018

In this paper, we have studied how to search an underwater object by learning the image generated by the side scan sonar in the convolution neural network. In the method of human side analysis of the side scan image or the image, the convolution neural network algorithm can enhance the efficiency of the analysis. The image data of the side scan sonar used in the experiment is the public data of NSWC (Naval Surface Warfare Center) and consists of four kinds of synthetic underwater objects. The convolutional neural network algorithm is based on Faster R-CNN (Region based Convolutional Neural Networks) learning based on region of interest and the details of the neural network are self-organized to fit the data we have. The results of the study were compared with a precision-recall curve, and we investigated the applicability of underwater object detection in convolution neural networks by examining the effect of change of region of interest assigned to sonar image data on detection performance.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2014.10a
• /
• pp.316-319
• /
• 2014

People's interests in coffee are increasing with the expansion of coffee market. In this trend, people's taste becomes more luxurious and coffee bean's quality is considered to be very important. Currently, bean defects are mainly detected by experienced specialists. In this paper, a detection system of bean defects using machine learning is presented. This system concentrates on detecting two main defect types : bean's shape and insect damage. Convolutional Neural Networks are used for machine learning. The neural networks are comprised of two neural networks. The first neural network detects defects in the bean's shape, and the second one detects the bean's insect damage. The development of this system could be a starting point for automated coffee bean defects detection. Later, further research is needed to detect other bean defect types.