• Proceedings of the Korean Information Science Society Conference
• /
• 2007.06b
• /
• pp.182-187
• /
• 2007

본 논문에서는 이미지 상에 나타난 색상 정보를 추출하기 위한 새로운 커널 메소드(Kernel method)인 Grid kernel을 제안한다. 제안한 Grid kernel은 Convolution kernel의 하나로 이미지 상에 나타나는 자질을 주변 픽셀에서 나타나는 자질로 정의 하고 이를 재귀적으로 적용함으로써 두 이미지를 비교한다. 본 논문에서는 제안한 커널을 차량 색상 인식 문제에 적용하여 차량 색상 인식 모델을 제안한다. 이미지 생성시 나타나는 주변 요인으로 인해 차량의 색상을 추출하는 것은 어려운 문제이다. 이미지가 야외에서 촬영되기 때문에 시간, 날씨 등의 주변 요인은 같은 차량이라 하더라도 다른 색상을 보이게 할 수 있다. 이를 해결하기 위해 Grid kernel이 적용된 차량 색상 인식 모델은 이미지를 HSV (Hue-Saturation-Value) 색상 공간으로 사상하여 명도를 배제하였다. 제안한 커널과 색상 인식 모델을 검증하기 위해 5가지 색상을 가진 차량 이미지를 이용하여 실험을 하였으며, 실험 결과 92.4%의 정확율과 92.0%의 재현율을 보였다.

• Journal of IKEEE
• /
• v.20 no.4
• /
• pp.351-360
• /
• 2016

Recently, machine learning algorithms, especially deep learning-based algorithms, have been receiving attention due to its high classification performance. Among the algorithms, Convolutional Neural Network(CNN) is known to be efficient for image processing tasks used for Advanced Driver Assistance Systems(ADAS). However, it is difficult to achieve real-time processing for CNN in vehicle embedded software environment due to the repeated operations contained in each layer of CNN. In this paper, we propose a hardware accelerator which enhances the execution time of CNN by parallelizing the repeated operations such as convolution. Xilinx ZC706 evaluation board is used to verify the performance of the proposed accelerator. For $36{\times}36$ input images, the hardware execution time of CNN is 2.812ms in 100MHz clock frequency and shows that our hardware can be executed in real-time.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.32 no.4
• /
• pp.629-635
• /
• 2022

Cyber threats are also increasing with recent social changes and the development of ICT technology. Malicious codes used in cyber threats are becoming more advanced and intelligent, such as analysis environment avoidance technology, concealment, and fileless distribution, to make analysis difficult. Machine learning technology is being used to effectively analyze these malicious codes, but a lot of effort is needed to increase the accuracy of classification. In this paper, we propose a malicious code detection technology based on API call interval characteristics to improve the classification performance of machine learning. The proposed technology uses API call characteristics for each section and entropy of binary to separate characteristic factors into sections based on the extraction malicious code and API call order of normal binary. It was verified that malicious code can be well analyzed using the support vector machine (SVM) algorithm for the extracted characteristic factors.

• Convergence Security Journal
• /
• v.23 no.1
• /
• pp.89-95
• /
• 2023

In order to help reduce the medical expenses of patients with auto insurance accidents, this study predicted the treatment period, which is the most important factor in the medical expenses of patients in their 40s and 50s, and analyzed the factors affecting the treatment period. To this end, a mechine learning model using five algorithms such as Decision Tree was created, and its performance was compared and analyzed between models. There were three algorithms that showed good performance including Decison Tree, Gradient Boost, and XGBoost. In addition, as a result of analyzing the factors affecting the prediction of the treatment period, the type of hospital, the treatment area, age, and gender were found. Through these studies, easy research methods such as the use of AutoML were presented, and we hope that the results of this study will help policies to reduce medical expenses for automobile insurance accidents.