• 한국컴퓨터정보학회논문지
• /
• 제26권6호
• /
• pp.1-8
• /
• 2021

본 논문은 저해상도 적외선영상을 사양이 낮은 임베디드 시스템에서 추론 가능하도록 강화된 VGG 스타일과 Global Average Pooling 조합으로 정확도를 증가시키면서 연산량을 최소화하는 딥러닝 컨볼루션 신경망을 이용한 저해상도 적외선 표적 분류 방법을 제안한다. 제안한 알고리즘은 OKTAL-SE로 생성한 합성영상 클래스 9개 3,723,328개를 분류하였다. 최초 임베디드 추론 가능하도록 파라메터 수가 최소화된 최대풀링 레이어 기준 입력단 8개와 출력단 8개 조합에 비해 강화된 VGG 스타일을 적용한 입력단 4개와 출력단 16개 필터수 조합을 이용하여 연산량은 약 34% 감소시켰으며, 정확도는 약 2.4% 증가시켜 최종 정확도 96.1%을 획득하였다. 추가로 C 코드로 포팅하여 수행시간을 확인하였으며, 줄어든 연산량 만큼 수행 시간이 약 32% 줄어든 것을 확인할 수 있었다.

• 항공우주시스템공학회지
• /
• 제10권3호
• /
• pp.63-69
• /
• 2016

The properties of composite materials, when compared to those of metallic materials, are highly variable due to many factors including the batch-to-batch variability of raw materials, the prepreg manufacturing process, material handling, part-fabrication techniques, ply-stacking sequences, environmental conditions, and test procedures. It is therefore necessary to apply reliable statistical-analysis techniques to obtain the design allowables of composite materials. A new composite-material qualification process has been developed by the Advanced General Aviation Transport Experiments (AGATE) consortium to yield the lamina-design allowables of composite materials according to standardized coupon-level tests and statistical techniques; moreover, the generated allowables database can be shared among multiple users without a repeating of the full qualification procedure by each user. In 2005, NASA established the National Center for Advanced Materials Performance (NCAMP) with the purpose of refining and enhancing the AGATE process to a self-sustaining level to serve the entire aerospace industry. In this paper, the statistical techniques and procedures for the generation of the allowables of aerospace composite materials will be discussed with a focus on the pooling-across-environments method.

• 전기전자학회논문지
• /
• 제21권3호
• /
• pp.244-247
• /
• 2017

본 논문에서는 GP-GPU를 활용한 보행자 추론 컨볼루션 뉴럴 네트워크를 구현했다. CNN은 구조를 정한 후, 학습에서 얻은 가중치를 이용해 기존 연구인 256개의 스레드를 가지는 GP-GPU를 활용해 추론을 수행했다. 학습에는 Inter i7-4470 CPU와 Matlab을 사용했다. Dataset은 Daimler Pedestrian Dataset을 사용했다. GP-GPU는 PCIe를 이용해 PC로부터 제어를 받으며, FPGA로 동작한다. 각 레이어의 depth와 size에 따라 스레드를 할당했다. 풀링 레이어의 경우는 over warpping pooling을 사용했기 때문에 횡영역과 종영역에 추가적인 연산을 수행했다. 한 번의 추론에는 약 12ms가 걸린다.

• 한국산학기술학회논문지
• /
• 제11권7호
• /
• pp.2603-2609
• /
• 2010

최근 항만에서는 이송장비의 운영 효율성 제고를 위해 운영방식의 개선에 관한 다양한 시도가 이루어지고 있다. 기존 조별운영 방식에서는 특정 안벽크레인에 지연이 발생하면 배정된 YT들도 함께 작업을 중단해야만 하는 문제가 생겼다. 이 문제를 해결하기 위해 본 연구에서는 운영 방식을 풀링운영 방식으로 전환하고 운영 효과를 높이기 위해 장치장 내에 무선 측위 장비를 설치하고 YT의 위치와 혼잡도를 측정하였다. 본 연구를 통해 풀링운영을 위한 YT의 이동을 실시간으로 추적할 수 있게 되었고 장치장의 혼잡도를 객관적으로 분석하고 평가하는 작업통제 방법이 마련되었다.

• 대한핵의학회지
• /
• 제39권3호
• /
• pp.151-162
• /
• 2005

Hemangioma is the most common benign tumor of the liver, with a prevalence estimated as high as 7%. Tc-99m red blood cell (RBC) hepatic blood pool scan with single photon omission computed tomography (SPECT) imaging is extremely useful for the confirmation or exclusion of hepatic hemangiomas. The classic finding of absent or decreased perfusion and increased blood pooling ("perfusion/blood pool mismatch") is the key diagnostic element in the diagnosis of hemangiomas. The combination of early arterial flow and delayed blood pooling ("perfusion/blood pool match") is shown uncommonly. In giant hemangioma, filling with radioactivity appears first in the periphery, with progressive central fill-in on sequential RBC blood pool scan. However, the reverse filling pattern, which begins first in the center with progressive peripheral filling, is also rarely seen. Studies with false-positive blood pooling have been reported infrequently in nonhemangiomas, including hemangiosarcoma, hepatocellular carcinoma, hepatic adenoma, and metastatic carcinomas (adenocarcinoma of the colon, small cell carcinoma of the lung, neruroendocrine carcinoma). False-negative results have been also reported rarely except for small hemagniomas that are below the limits of spatial resolution of gamma camera.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제15권5호
• /
• pp.1778-1797
• /
• 2021

Document image binarization is an important pre-processing step in document analysis and archiving. The state-of-the-art models for document image binarization are variants of encoder-decoder architectures, such as FCN (fully convolutional network) and U-Net. Despite their success, they still suffer from three limitations: (1) reduced feature map resolution due to consecutive strided pooling or convolutions, (2) multiple scales of target objects, and (3) reduced localization accuracy due to the built-in invariance of deep convolutional neural networks (DCNNs). To overcome these three challenges, we propose an improved semantic segmentation model, referred to as DP-LinkNet, which adopts the D-LinkNet architecture as its backbone, with the proposed hybrid dilated convolution (HDC) and spatial pyramid pooling (SPP) modules between the encoder and the decoder. Extensive experiments are conducted on recent document image binarization competition (DIBCO) and handwritten document image binarization competition (H-DIBCO) benchmark datasets. Results show that our proposed DP-LinkNet outperforms other state-of-the-art techniques by a large margin. Our implementation and the pre-trained models are available at https://github.com/beargolden/DP-LinkNet.

• Communications for Statistical Applications and Methods
• /
• 제29권5호
• /
• pp.547-559
• /
• 2022

For a chi-squared test, which is a statistical method used to test the independence of a contingency table of two factors, the expected frequency of each cell must be greater than 5. The percentage of cells with an expected frequency below 5 must be less than 20% of all cells. However, there are many cases in which the regional expected frequency is below 5 in general small area studies. Even in large-scale surveys, it is difficult to forecast the expected frequency to be greater than 5 when there is small area estimation with subgroup analysis. Another statistical method to test independence is to use the Bayes factor, but since there is a high ratio of data dependency due to the nature of the Bayesian approach, the low expected frequency tends to decrease the precision of the test results. To overcome these limitations, we will borrow information from areas with similar characteristics and pool the data statistically to propose a pooled Bayes test of independence in target areas. Jo et al. (2021) suggested hierarchical Bayesian pooling models for small area estimation of categorical data, and we will introduce the pooled Bayes factors calculated by expanding their restricted pooling model. We applied the pooled Bayes factors using bone mineral density and body mass index data from the Third National Health and Nutrition Examination Survey conducted in the United States and compared them with chi-squared tests often used in tests of independence.

• Asian-Australasian Journal of Animal Sciences
• /
• 제20권3호
• /
• pp.316-327
• /
• 2007

In this study, a Korean native cattle strain (Hanwoo) evidencing high performance in terms of both meat quality and quantity was employed in the generation of 150,000 BAC clones with an average insert size of 140 kb, and corresponding to about a 6X coverage of bovine chromosomal DNA. The BAC clones were pooled in a mini-scale via three rounds of a pooling protocol, and the efficiency of this pooling protocol was evaluated by testing the accuracy of accessibility to the positive clones, via a PCR-based screening method. Two sets of primers designed from each of two known genes were tested, and each yielded 2 or 3 positive clones for each gene, thereby indicating that the BAC library pooling system was appropriate with regard to the accession of the target BAC clones. Analyses of $3.3{\times}10^6$ base pairs obtained from the 7,090 BAC end sequence (BES) showed that 34.88% of the DNA sequence harbored the repetition sequence. Analysis of the 7,090 BES to the $1^{st}$ and $2^{nd}$ generation radiation hybrid map of the cattle genome, using the COMPASS program designed for the construction of a cattle-human comparative mapping, resulted in the localization of a total of 1,374 clones proximal to 339 $1^{st}$ generation markers, and 1,721 clones proximal to 664 $2^{nd}$ generation markers. Collectively, the BAC library and pooling system of the BAC clones from the Korean cattle, coupled with the chromosome-localized BAC clones, will provide us with novel tools for the excavation of desired clones for genome mapping and sequencing, and will also furnish us with additional information regarding breed differences in cattle.

• 한국생물정보학회:학술대회논문집
• /
• 한국생물정보시스템생물학회 2005년도 BIOINFO 2005
• /
• pp.4-4
• /
• 2005

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제9권2호
• /
• pp.763-774
• /
• 2015

Hand posture recognition has been a wide region of applications in Human Computer Interaction and Computer Vision for many years. The problem arises mainly due to the high dexterity of hand and self-occlusions created in the limited view of the camera or illumination variations. To remedy these problems, a hand posture recognition method using 3-D point cloud is proposed to explicitly utilize 3-D information from depth maps in this paper. Firstly, hand region is segmented by a set of depth threshold. Next, hand image normalization will be performed to ensure that the extracted feature descriptors are scale and rotation invariant. By robustly coding and pooling 3-D facets, the proposed descriptor can effectively represent the various hand postures. After that, SVM with Gaussian kernel function is used to address the issue of posture recognition. Experimental results based on posture dataset captured by Kinect sensor (from 1 to 10) demonstrate the effectiveness of the proposed approach and the average recognition rate of our method is over 96%.