• Journal of Biomedical Engineering Research
• /
• v.44 no.1
• /
• pp.25-32
• /
• 2023

Recognizing the size and location of prostate cancer is critical for prostate cancer diagnosis, treatment, and predicting prognosis. This paper proposes a model to classify the tumor region and normal tissue with cross-sectional visual images of prostatectomy tissue. We used specimen images of 44 prostate cancer patients who received prostatectomy at Gachon University Gil Hospital. A total of 289 prostate slice images consist of 200 slices including tumor region and 89 slices not including tumor region. Images were divided based on the presence or absence of tumor, and a total of 93 features from each slice image were extracted using Radiomics: 18 first order, 24 GLCM, 16 GLRLM, 16 GLSZM, 5 NGTDM, and 14 GLDM. We compared feature selection techniques such as LASSO, ANOVA, SFS, Ridge and RF, LR, SVM classifiers for the model's high performances. We evaluated the model's performance with AUC of the ROC curve. The results showed that the combination of feature selection techniques LASSO, Ridge, and classifier RF could be best with an AUC of 0.99±0.005.

• Journal of Broadcast Engineering
• /
• v.27 no.6
• /
• pp.872-884
• /
• 2022

With the development of the IoT industry, different types of time series data are being generated in various industries, and it is evolving into research that reproduces and utilizes it through re-integration. In addition, due to data processing speed and issues of the utilization system in the actual industry, there is a growing tendency to compress the size of data when using time series data and integrate it. However, since the guidelines for integrating time series data are not clear and each characteristic such as data description time interval and time section is different, it is difficult to use it after batch integration. In this paper, two integration methods are proposed based on the integration criteria setting method and the problems that arise during integration of time series data. Based on this, integration framework of a heterogeneous time series data was constructed that is considered the characteristics of time series data, and it was confirmed that different heterogeneous time series data compressed can be used for integration and various machine learning.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.27 no.1
• /
• pp.69-75
• /
• 2023

Small communication devices used in the Internet of Things are vulnerable to various hacking because they do not apply advanced encryption techniques due to their low memory capacity or slow computation speed. In order to increase the authentication reliability of small-sized transmitters operating in 433MHz band, we introduce an RF fingerprint and adopt a convolutional neural network (CNN) as a classification algorithm. The preamble signal transmitted by each transmitter are extracted and collected using software-defined-radio to constitute a training data set, which is used for training the CNN. We tested identification of 20 transmitters in four different scenarios and obtained high identification accuracy. In particular, the accuracy of 95.8% and 92.6% was obtained, respectively in the scenario where the test was performed at a location different from the transmitter's location at the time of collecting training data, and in the scenario where the transmitter moves at walking speed.

• The Journal of the Convergence on Culture Technology
• /
• v.8 no.6
• /
• pp.905-910
• /
• 2022

Drowsy driving requires a lot of social attention because it increases the incidence of traffic accidents and leads to fatal accidents. The number of accidents caused by drowsy driving is increasing every year. Therefore, in order to solve this problem all over the world, research for measuring various biosignals is being conducted. Among them, this paper focuses on non-contact biosignal analysis. Various noises such as engine, tire, and body vibrations are generated in a running vehicle. To measure the driver's heart rate and respiration rate in a driving vehicle with a piezoelectric sensor, a sensor plate that can cushion vehicle vibrations was designed and noise generated from the vehicle was reduced. In addition, we developed a system for classifying whether the driver is sleeping or not by extracting the model using the CNN-LSTM ensemble learning technique based on the signal of the piezoelectric sensor. In order to learn the sleep state, the subject's biosignals were acquired every 30 seconds, and 797 pieces of data were comparatively analyzed.

• Journal of the Korea Convergence Society
• /
• v.12 no.11
• /
• pp.91-97
• /
• 2021

This paper uses a data extension technique originally designed for the rule refinement problem to handling incomplete data. This technique is characterized in that each event can have a weight indicating importance, and each variable can be expressed as a probability value. Since the key problem in this paper is to find the probability that is closest to the missing value and replace the missing value with the probability, three different algorithms are used to find the probability for the missing value and then store it in this data structure format. And, after learning to classify each information area with the SVM classification algorithm for evaluation of each probability structure, it compares with the original information and measures how much they match each other. The three algorithms for the imputation probability of the missing value use the same data structure, but have different characteristics in the approach method, so it is expected that it can be used for various purposes depending on the application field.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.2
• /
• pp.520-541
• /
• 2023

In CRNs, SS is of utmost significance. Every CR user generates a sensing report during the training phase beneath various circumstances, and depending on a collective process, either communicates or remains silent. In the training stage, the fusion centre combines the local judgments made by CR users by a majority vote, and then returns a final conclusion to every CR user. Enough data regarding the environment, including the activity of PU and every CR's response to that activity, is acquired and sensing classes are created during the training stage. Every CR user compares their most recent sensing report to the previous sensing classes during the classification stage, and distance vectors are generated. The posterior probability of every sensing class is derived on the basis of quantitative data, and the sensing report is then classified as either signifying the presence or absence of PU. The ISVM technique is utilized to compute the quantitative variables necessary to compute the posterior probability. Here, the iterations of SVM are tuned by novel GO-PSA by combining GOA and PSO. Novel GO-PSA is developed since it overcomes the problem of computational complexity, returns minimum error, and also saves time when compared with various state-of-the-art algorithms. The dependability of every CR user is taken into consideration as these local choices are then integrated at the fusion centre utilizing an innovative decision combination technique. Depending on the collective choice, the CR users will then communicate or remain silent.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.21 no.6
• /
• pp.36-56
• /
• 2022

Traffic accidents occur in Korea are calculated with the 「Automobile Accident Negligence Ratio Certification Standard」 prepared by the 'General Insurance Association of Korea' and the insurance company's agreement or judgment is made. However, disputes are frequently occurring in calculating the negligence ratio. Therefore, it is thought that a more effective response would be possible if accident type according to the standard could be quickly identified using traffic accident information prepared by police. Therefore, this study aims to develop a model that learns the accident information prepared by the police and classifies it to match the accident type in the standard. In particular, through data mining, keywords necessary to classify the accident types of the standard were extracted from the accident data of the police. Then, models were developed to derive the types of accidents by learning the extracted keywords through decision trees and random forest models.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.18 no.3
• /
• pp.89-100
• /
• 2023

Recently, on-device artificial intelligence (AI) solutions using mobile devices and embedded edge devices have emerged in various fields, such as computer vision, to address network traffic burdens, low-energy operations, and security problems. Although vision transformer deep learning models have outperformed conventional convolutional neural network (CNN) models in computer vision, they require more computations and parameters than CNN models. Thus, they are not directly applicable to embedded edge devices with limited hardware resources. Many researchers have proposed various model compression methods or lightweight architectures for vision transformers; however, there are only a few studies evaluating the effects of model compression techniques of vision transformers on performance. Regarding this problem, this paper presents a performance evaluation of vision transformers on embedded platforms. We investigated the behaviors of three vision transformers: DeiT, LeViT, and MobileViT. Each model performance was evaluated by accuracy and inference time on edge devices using the ImageNet dataset. We assessed the effects of the quantization method applied to the models on latency enhancement and accuracy degradation by profiling the proportion of response time occupied by major operations. In addition, we evaluated the performance of each model on GPU and EdgeTPU-based edge devices. In our experimental results, LeViT showed the best performance in CPU-based edge devices, and DeiT-small showed the highest performance improvement in GPU-based edge devices. In addition, only MobileViT models showed performance improvement on EdgeTPU. Summarizing the analysis results through profiling, the degree of performance improvement of each vision transformer model was highly dependent on the proportion of parts that could be optimized in the target edge device. In summary, to apply vision transformers to on-device AI solutions, either proper operation composition and optimizations specific to target edge devices must be considered.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2023.05a
• /
• pp.239-239
• /
• 2023

최근 도시화 및 기후변화에 따른 재난의 피해가 증가하고 있다. 국내 기상청에서는 호우 및 태풍에 대한 예·경보(주의보, 경보)를 전국적으로 통일된 기준(3시간, 12시간 누적강우량)에 따라 발령하고 있다. 이에 따라 현재 예·경보 기준에는 피해가 발생한 사상에 대한 지역별 특성이 고려되지 않는 문제점이 있다. 본 연구에서는 이러한 문제점을 해결하기 위하여 서울특별시, 인천광역시, 경기도의 호우 및 태풍에 대한 재해사상별 발생한 피해액 및 누적강우량을 활용하여 재해강도의 단계별 기준을 수립하고, 입력자료로 관측된 강우값을 활용하여 발생할 수 있는 재해의 발생 강도를 분류하는 모형을 개발하고자 하였다. 본 연구에서는 호우 및 태풍에 의한 재해 피해액의 분위별로 재해강도 단계(관심, 주의, 경계, 심각)를 분류하였고, 재해강도 단계에 따른 누적강우량 기준을 지자체별로 제시하였으며, 분류한 재해의 강도 단계를 모형의 종속변수로 활용하였다. 재해피해가 발생하지 않은 무강우 지속시간을 산정하여 호우 사상을 분류하였다. 지자체별로 재해 발생강도 분류 모형 개발을 위하여 머신러닝 모형 4가지(의사결정나무, 서포트 벡터 머신, 랜덤 포레스트, XGBoost)를 활용하였다. 본 연구에서 분류한 피해가 발생하지 않은 호우사상 및 피해가 발생한 사상별로 강우량, 지속시간 최대 강우량(3시간, 12시간), 선행강우량, 누적강우량을 독립변수로 입력하여 종속변수인 재해 발생 강도를 분류하였다. 각 모형별로 F1 Score를 이용한 정확도 평가 결과, 의사결정나무의 F1 Score가 평균 0.56으로 가장 우수한 정확도를 가지는 것으로 평가되었다. 본 연구에서 제시하는 머신러닝 기반 재해 발생 강도 분류모형을 활용하면 호우 및 태풍에 의한 재해에 대하여 지자체별로 재해 발생 강도를 단계별로 파악할 수 있어, 재난 담당자들의 의사결정을 위한 참고 자료로 활용될 수 있을 것으로 판단된다.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2023.05a
• /
• pp.423-423
• /
• 2023

습지 생태계는 탄소저장고, 대기 온·습도 조절 등의 기능을 수행하는 만큼 면밀한 관리가 요구된다. 습지의 규모와 생태계는 밀접한 연관성을 가지므로 그 규모를 우선적으로 파악할 필요가 있으며, 이를 위해 지표면의 상태를 산지, 습지, 수역 등의 항목으로 구분한 토지피복지도가 고려될 수 있다. 현재, 환경부에서 운영 중인 환경공간정보서비스(https://egis.me.go.kr/)에서는 각각 30 m, 5 m, 1 m의 공간 해상도와 7, 22, 41가지 분류 항목을 갖는 대분류, 중분류, 세분류로 구분된 토지피복지도를 제공하며 이러한 자료들은 모두 1년 이상의 시간 해상도를 갖는다. 습지의 경우, 계절에 따른 환경 변화로 인한 규모의 변동성이 크게 나타날 수 있기 때문에 1년 이하의 시간 해상도를 갖는 고품질 토지피복 분류 정보가 요구된다. 따라서 본 연구에서는 기존 자료의 낮은 시간 해상도 보완을 목표로, 1개월과 30 m의 시·공간 해상도를 갖는 토지피복지도를 구축하기 위한 방법론을 제안하고자 한다. 이를 위해 Landsat-8 등과 같은 다양한 인공위성 자료를 수집하고, Support Vector Machine 등과 같은 머신러닝 기법을 적용하였다. 최종적으로 습지보전법에서 지정한 습지보호지역 중 내륙습지 26개소를 대상으로, 본 연구로부터 산출된 토지피복지도를 기존 환경공간정보서비스 내 대분류 토지피복지도와 비교·평가하였다.