• 한국정보통신학회:학술대회논문집
• /
• 한국정보통신학회 2014년도 추계학술대회
• /
• pp.342-345
• /
• 2014

기계학습을 이용한 네트워크 침입탐지시스템은 선택된 특징 조합에 따라 정확성 및 효율성 측면에서 크게 영향을 받는다. 하지만 일반적으로 사용되는 침입탐지용 특징들로부터 최적의 조합을 찾아내는 일은 많은 계산량을 요구한다. 예를 들어 n개로 구성된 특징들로부터 가능한 특징조합은 $2^n-1$ 개이다. 본 논문에서는 이러한 문제를 해결하기 위한 최적 특징 선택 알고리즘을 제시한다. 제안한 알고리즘은 최적화 문제 해결을 위한 대표적인 메타 휴리스틱 알고리즘인 지역탐색 알고리즘에 기반 한다. 또한 특징 조합을 평가를 위해 선택된 특징 요소와 k-means 군집화 알고리즘을 이용해 구해진 군집화의 정확성을 비용함수로 사용한다. 제안한 특징 선택 알고리즘의 평가를 위해 NSL-KDD 데이터와 인공 신경망을 사용해 특징 모두를 사용한 경우와 비교한다.

• 융합정보논문지
• /
• 제12권3호
• /
• pp.17-24
• /
• 2022

모바일 악성 앱이 급증하고 있으며, 전 세계 모바일 OS 시장의 대부분을 차지하고 있는 안드로이드가 모바일 사이버 보안 위협의 주요 대상이 되고 있다. 따라서 빠르게 진화하는 악성 앱에 대응하기 위해 인공지능 구현기술 중 하나인 기계학습을 활용한 악성 앱 탐지 기법의 필요성이 대두되고 있다. 본 논문은 악성 앱의 탐지성능을 향상할 수 있는 특성 선택 및 특성 추출을 이용한 특성 선별 방법을 제안하였다. 특성 선별 과정에서 특성 개수에 따라 탐지 성능이 향상되었으며, 권한보다 API가 상대적으로 좋은 탐지 성능을 보였고, 두 특성을 조합하면 평균 93% 이상의 높은 탐지 정밀도를 보여 적절한 특성의 조합이 탐지 성능을 높일 수 있음을 확인하였다.

• Current Optics and Photonics
• /
• 제8권3호
• /
• pp.215-224
• /
• 2024

In medium- and long-range optical imaging systems, atmospheric turbulence causes blurring and distortion of images, resulting in loss of image information. An image-restoration method based on an adaptive feature-fusion multi-input-multi-output (MIMO) dense U-shaped network (Unet) is proposed, to restore a single image degraded by atmospheric turbulence. The network's model is based on the MIMO-Unet framework and incorporates patch-embedding shallow-convolution modules. These modules help in extracting shallow features of images and facilitate the processing of the multi-input dense encoding modules that follow. The combination of these modules improves the model's ability to analyze and extract features effectively. An asymmetric feature-fusion module is utilized to combine encoded features at varying scales, facilitating the feature reconstruction of the subsequent multi-output decoding modules for restoration of turbulence-degraded images. Based on experimental results, the adaptive feature-fusion MIMO dense U-shaped network outperforms traditional restoration methods, CMFNet network models, and standard MIMO-Unet network models, in terms of image-quality restoration. It effectively minimizes geometric deformation and blurring of images.

• 한국실내디자인학회논문집
• /
• 제23권5호
• /
• pp.76-85
• /
• 2014

When conscious and unconscious exploring information of space-visitors which is contained in the information acquired in the process of seeing any space is analyzed, it can be found what those visitors pick up as factors in the space for its selection as visual information in order to put it into action. This study, with the object of the space reproduced in three dimensions from the cafe which was visited for conversation, has analyzed the process of acquiring space-information by sex to find out the features of scanning path, findings of which are the followings. First, the rate of scanning type of males was "Combination (50.5%)- Circulation (31.0%) and that of females "Horizontal (32.5%) - Combination (32.1%)", which shows that there was a big difference by sex in the scanning path which took place in the process of observing any space. Second, when the features of continuous observation frequency by sex is looked into, the trends of increased "horizontal" scanning and decreased "Combination" scanning of both showed the same as the frequency of continuous observations increased, while in case of "Circulation" scanning, that of females was found to decrease but that of males showed the aspect of confusion. Third, the 'Combination' scanning of males was found strong at the short observation time with three times of continuous observation frequency defined as "Attention Concentration" while the distinct feature was seen that the scanning type was dispersed to "combination-circulation" as the frequency of continuous observation increased. Females start the information acquirement with "combination-circulation" but in the process of visual appreciation they showed a strong "Horizontal" These scanning features can be defined as those by sex for acquiring space information and therefore are very significant because they are fundamental studies which will enable any customized space-design by sex.

• 대한임베디드공학회논문지
• /
• 제9권5호
• /
• pp.261-268
• /
• 2014

Face detection is essential to the full automation of face image processing application system such as face recognition, facial expression recognition, age estimation and gender identification. It is found that local image features which includes Haar-like, LBP, and MCT and the Adaboost algorithm for classifier combination are very effective for real time face detection. In this paper, we present a face detection method using local pixel direction code(PDC) feature and lookup table classifiers. The proposed PDC feature is much more effective to dectect the faces than the existing local binary structural features such as MCT and LBP. We found that our method's classification rate as well as detection rate under equal false positive rate are higher than conventional one.

• 대한음성학회지:말소리
• /
• 제52호
• /
• pp.145-160
• /
• 2004

In this paper, a new noise compensation method based on the eigenvoice framework in feature space is proposed to reduce the mismatch between training and testing environments. The difference between clean and noisy environments is represented by the linear combination of K eigenvectors that represent the variation among environments. In the proposed method, the performance improvement of speech recognition systems is largely affected by how to construct the noisy models and the bias vector set. In this paper, two methods, the one based on MAP adaptation method and the other using stereo DB, are proposed to construct the noisy models. In experiments using Aurora 2 DB, we obtained 44.86% relative improvement with eigen-environment method in comparison with baseline system. Especially, in clean condition training mode, our proposed method yielded 66.74% relative improvement, which is better performance than several methods previously proposed in Aurora project.

• Journal of Computational Design and Engineering
• /
• 제3권4호
• /
• pp.370-384
• /
• 2016

Computer-aided materials design requires new modeling approaches to characterize and represent fine-grained geometric structures and material compositions at multiple scales. Recently, a dual-Rep approach was developed to model materials microstructures based on a new basis function, called surfacelet. As a combination of implicit surface and wavelets, surfacelets can efficiently identify and represent planar, cylindrical, and ellipsoidal geometries in material microstructures and describe the distribution of compositions and properties. In this paper, these primitive surfacelets are extended and composite surfacelets are proposed to model more complex geometries. Composite surfacelets are constructed by Boolean operations on the primitives. The surfacelet transform is applied to match geometric features in three-dimensional images. The composition of the material near the identified features can then be modeled. A cubic surfacelet and a v-joint surfacelet are developed to demonstrate the reverse engineering process of retrieving material compositions from material images.

• 한국생산제조학회지
• /
• 제9권2호
• /
• pp.144-150
• /
• 2000

Surface roughness is one of the most important parameters to estimate quality of products. As this reason so many studies were car-ried out through various attempts that were contact or non-contact using computer vision. Even through these efforts there were few good results in this research., however texture analysis making a important role to solve these problems in various fields including universe aviation living thing and fibers. In this study feature value of co-occurrence matrix was calculated by statistic method and roughness value of worked surface was classified, of it. Experiment was carried out using input vector of neural network with characteristic value of texture calculated from worked surface image. It's found that recognition rate of 74% was obtained when adapting texture features. In order to enhance recogni-tion rate combination type in characteristics value of texture was changed into input vector. As a result high recognition rate of 92.6% was obtained through these processes.

• The Journal of the Acoustical Society of Korea
• /
• 제21권2E호
• /
• pp.98-104
• /
• 2002

The present study describes a combination method to recognize the human affective states such as anger, happiness, sadness, or surprise. For this, we extracted emotional features from voice signals and facial expressions, and then trained them to recognize emotional states using hidden Markov model (HMM) and neural network (NN). For voices, we used prosodic parameters such as pitch signals, energy, and their derivatives, which were then trained by HMM for recognition. For facial expressions, on the other hands, we used feature parameters extracted from thermal and visible images, and these feature parameters were then trained by NN for recognition. The recognition rates for the combined parameters obtained from voice and facial expressions showed better performance than any of two isolated sets of parameters. The simulation results were also compared with human questionnaire results.

• Journal of Information Technology Applications and Management
• /
• 제15권1호
• /
• pp.243-260
• /
• 2008

To prevent low yields in the semiconductor industry is crucial to the success of that industry. However, to prevent low yields is difficult because of too many factors to affect yield variation and their complex relation in the semiconductor manufacturing process. This study presents a new efficient detection methodology for detecting abnormal yields including high and low yields, which can forecast the yield level of a production unit (namely a lot) based on yield-related feature variables' behaviors. In the methodology, we use C5.0 to identify the yield-related feature variables that are the combination of correlated process variables associated with yield, use SOM (Self-Organizing Map) neural networks to extract and classify significant patterns of past abnormal yield lots and finally use C5.0 to generate classification rules for detecting abnormal yield lot. We illustrate the effectiveness of our methodology using a semiconductor manufacturing company's field data.