• Korean Journal of Radiology
• /
• v.23 no.1
• /
• pp.139-149
• /
• 2022

Objective: To compare the effects of bone suppression imaging using deep learning (BSp-DL) based on a generative adversarial network (GAN) and bone subtraction imaging using a dual energy technique (BSt-DE) on radiologists' performance for pulmonary nodule detection on chest radiographs (CXRs). Materials and Methods: A total of 111 adults, including 49 patients with 83 pulmonary nodules, who underwent both CXR using the dual energy technique and chest CT, were enrolled. Using CT as a reference, two independent radiologists evaluated CXR images for the presence or absence of pulmonary nodules in three reading sessions (standard CXR, BSt-DE CXR, and BSp-DL CXR). Person-wise and nodule-wise performances were assessed using receiver-operating characteristic (ROC) and alternative free-response ROC (AFROC) curve analyses, respectively. Subgroup analyses based on nodule size, location, and the presence of overlapping bones were performed. Results: BSt-DE with an area under the AFROC curve (AUAFROC) of 0.996 and 0.976 for readers 1 and 2, respectively, and BSp-DL with AUAFROC of 0.981 and 0.958, respectively, showed better nodule-wise performance than standard CXR (AUAFROC of 0.907 and 0.808, respectively; p ≤ 0.005). In the person-wise analysis, BSp-DL with an area under the ROC curve (AUROC) of 0.984 and 0.931 for readers 1 and 2, respectively, showed better performance than standard CXR (AUROC of 0.915 and 0.798, respectively; p ≤ 0.011) and comparable performance to BSt-DE (AUROC of 0.988 and 0.974; p ≥ 0.064). BSt-DE and BSp-DL were superior to standard CXR for detecting nodules overlapping with bones (p < 0.017) or in the upper/middle lung zone (p < 0.017). BSt-DE was superior (p < 0.017) to BSp-DL in detecting peripheral and sub-centimeter nodules. Conclusion: BSp-DL (GAN-based bone suppression) showed comparable performance to BSt-DE and can improve radiologists' performance in detecting pulmonary nodules on CXRs. Nevertheless, for better delineation of small and peripheral nodules, further technical improvements are required.

• The Journal of Korean Medicine
• /
• v.44 no.4
• /
• pp.136-149
• /
• 2023

Objectives: Facial diagnosis is an important part of clinical diagnosis in traditional East Asian Medicine. In this paper, we proposed a model to quantitatively classify cold sensitivity using a fully automated facial image analysis system. Methods: We investigated cold sensitivity in 452 subjects. Cold sensitivity was determined using a questionnaire and the Cold Pattern Score (CPS) was used for analysis. Subjects with a CPS score below the first quartile (low CPS group) belonged to the cold non-sensitivity group, and subjects with a CPS score above the third quartile (high CPS group) belonged to the cold sensitivity group. After splitting the facial images into train/validation/test sets, the train and validation set were input into a convolutional neural network to learn the model, and then the classification accuracy was calculated for the test set. Results: The classification accuracy of the low CPS group and high CPS group using facial images in all subjects was 76.17%. The classification accuracy by sex was 69.91% for female and 62.86% for male. It is presumed that the deep learning model used facial color or facial shape to classify the low CPS group and the high CPS group, but it is difficult to specifically determine which feature was more important. Conclusions: The experimental results of this study showed that the low CPS group and the high CPS group can be classified with a modest level of accuracy using only facial images. There was a need to develop more advanced models to increase classification accuracy.

• Annual Conference on Human and Language Technology
• /
• 2017.10a
• /
• pp.184-189
• /
• 2017

With the advent of robust deep learning method, Neural machine translation has recently become a dominant paradigm and achieved adequate results in translation between popular languages such as English, German, and Spanish. However, its results in under-resourced languages Korean and Vietnamese are still limited. This paper reports an attempt at constructing a bidirectional Korean-Vietnamese Neural machine translation system with the supporting of Korean analysis tool - UTagger, which includes morphological analyzing, POS tagging, and WSD. Experiment results demonstrate that UTagger can significantly improve translation quality of Korean-Vietnamese NMT system in both translation direction. Particularly, it improves approximately 15 BLEU scores for the translation from Korean to Vietnamese direction and 3.12 BLEU scores for the reverse direction.

• Journal of Korea Multimedia Society
• /
• v.23 no.3
• /
• pp.412-420
• /
• 2020

This paper proposes a performance-improving face recognition system based on a super resolution method for low-resolution images. The conventional face recognition algorithm has a rapidly decreased accuracy rate due to small image resolution by a distance. To solve the previously mentioned problem, this paper generates a super resolution images based o deep learning method. The proposed method improved feature information from low-resolution images using a super resolution method and also applied face recognition using a feature extraction and an classifier. In experiments, the proposed method improves the face recognition rate when compared to conventional methods.

• 한국어정보학회:학술대회논문집
• /
• 2017.10a
• /
• pp.184-189
• /
• 2017

With the advent of robust deep learning method, Neural machine translation has recently become a dominant paradigm and achieved adequate results in translation between popular languages such as English, German, and Spanish. However, its results in under-resourced languages Korean and Vietnamese are still limited. This paper reports an attempt at constructing a bidirectional Korean-Vietnamese Neural machine translation system with the supporting of Korean analysis tool - UTagger, which includes morphological analyzing, POS tagging, and WSD. Experiment results demonstrate that UTagger can significantly improve translation quality of Korean-Vietnamese NMT system in both translation direction. Particularly, it improves approximately 15 BLEU scores for the translation from Korean to Vietnamese direction and 3.12 BLEU scores for the reverse direction.

• Annual Conference of KIPS
• /
• 2016.10a
• /
• pp.62-63
• /
• 2016

딥러닝의 대표적 개발 환경 중 하나인 Caffe를 임베디드 시스템의 메모리 구조를 고려하여 최적화하고 실제 측정 실험으로 기존의 방식보다 처리시간과 소비 전력량의 이득이 있다는 것을 확인하였다. 구체적으로 통합 메모리를 사용하는 임베디드 시스템 환경의 특성에 적합한 zero-copy기법을 적용하여 CPU와 GPU 모두 접근이 가능하도록 메모리 영역을 맵핑하는 방식으로 메모리 복제에 따른 오버헤드를 줄였으며, GoogLeNet 네트워크 모델에 대하여 10%의 처리 속도 향상과, 36% 소비 전력 감소를 확인하였다.

• Annual Conference of KIPS
• /
• 2018.10a
• /
• pp.808-810
• /
• 2018

Pedestrian tracking is a particular problem of object tracking, and an important component in various vision-based applications, such as autonomous cars or surveillance systems. After several years of development, pedestrian tracking in videos is still a challenging problem because of various visual properties of objects and surrounding environment. In this research, we propose a tracking-by-detection system for pedestrian tracking, which incorporates Convolutional Neural Network (CNN) and color information. Pedestrians in video frames are localized by a CNN, then detected pedestrians are assigned to their corresponding tracklets based on similarities in color distributions. The experimental results show that our system was able to overcome various difficulties to produce highly accurate tracking results.

• Journal of the Semiconductor & Display Technology
• /
• v.20 no.3
• /
• pp.117-119
• /
• 2021

With the development of the industry, the use of drones in specific mission flight is being actively studied. These drones fly a specified path and perform repetitive tasks. if the drone system will detect objects in real time, the performance of these mission flight will increase. In this paper, we implement object detection system and mount GPU acceleration to maximize the efficiency of limited device resources with drone video using Tensorflow Lite which enables in-device inference from a mobile device and Mobile SDK of DJI, a drone manufacture. For performance comparison, the average processing time per frame was measured when object detection was performed using only the CPU and when object detection was performed using the CPU and GPU at the same time.

• Journal of the Semiconductor & Display Technology
• /
• v.19 no.4
• /
• pp.71-76
• /
• 2020

In this paper, we propose a super-resolution method that reconstructs compressed low-resolution images into high-resolution images. We propose a CNN model with a small number of parameters, and even if quantization is applied to the proposed model, super-resolution can be implemented without deteriorating the image quality. To further improve the quality of the compressed low-resolution image, a new degradation model was proposed instead of the existing bicubic degradation model. The proposed degradation model is used only in the training process and can be applied by changing only the parameter values to the original CNN model. In the super-resolution image applying the proposed degradation model, visual artifacts caused by image compression were effectively removed. As a result, our proposed method generates higher PSNR values at compressed images and shows better visual quality, compared to conventional CNN-based SR methods.

• Journal of Korea Multimedia Society
• /
• v.24 no.2
• /
• pp.215-221
• /
• 2021

The agricultural is plagued by agricultural damage from wild boars every year. As a result, research on systems to repelling wild boars continues, and most of the systems are to detect objects with body temperature through sensors and then repelling them with actions such as light and sound. The problems of these system are operating regardless of wild boars and people, which can cause significant accident when using electric fence. In addition, If the same repelling action is repeated, wild boars can be adapted to that repelling action. As a solution to the two problems, Adaptation problem can be solved by random sounds and distinction problem can be solved by YOLO V4.