• 에너지공학
• /
• 제13권2호
• /
• pp.152-160
• /
• 2004

The present study investigates on the experimental and numerical results of heat transfer in the acoustic fields induced by ultrasonic waves. The strong upwards flow which moves from the bottom surface in a cavity to the free surface called as "acoustic streaming" was visualized by a particle image velocimetry (PIV). In addition, the augmentation ratio of heat transfer was experimentally investigated in the presence of acoustic streaming and was compared with the profiles of acoustic pressure calculated by the numerical analysis. A coupled finite element-boundary element method (FE-BEM) was applied for a numerical analysis. The results of experimental and numerical studies clearly show that acoustic pressure variations caused by ultrasonic waves in a medium are closely related to the augmentation of heat transfer.

• 대한기계학회:학술대회논문집
• /
• 대한기계학회 2003년도 춘계학술대회
• /
• pp.1603-1608
• /
• 2003

The Present Study reported on the experimental and numerical results of heat transfer in the acoustic fields induced by ultrasonic waves. The strong upwards flow called as acoustic streaming was visualized by a particle image velocimetry (P.I.V). in addition, the augmentation of heat transfer was experimentally investigated in the presence of acoustic streaming and was compared with the profiles of acoustic pressure calculated by the numerical analysis. Experimental and numerical studies clearly show that acoustic pressure variations are closely related to the augmentation of heat transfer.

• 한국가시화정보학회지
• /
• 제19권1호
• /
• pp.68-73
• /
• 2021

In this study, the detection of rubber o-ring defects was carried out using k-fold cross validation and Support Vector Machine (SVM) algorithm. The data process was carried out in 3 steps. First, we proceeded with a frame alignment to eliminate unnecessary regions in the learning and secondly, we applied gray-scale changes for computational reduction. Finally, data processing was carried out using image augmentation to prevent data overfitting. After processing data, SVM algorithm was used to obtain normal and defect detection accuracy. In addition, we applied the SVM algorithm through the k-fold cross validation method to compare the classification accuracy. As a result, we obtain results that show better performance by applying the k-fold cross validation method.

• 한국컴퓨터정보학회:학술대회논문집
• /
• 한국컴퓨터정보학회 2022년도 제66차 하계학술대회논문집 30권2호
• /
• pp.539-540
• /
• 2022

본 논문에서는 Bounding Box가 포함된 증강 이미지 데이터셋을 손쉽게 생성할 수 있는 독립형 GUI 프로그램을 제안한다. 본 논문의 연구를 통하여 직관적인 마우스 클릭 동작만으로 적은 수의 이미지 파일과 annotation 파일로부터 필요한 만큼의 증강 이미지 데이터셋을 짧은 시간 내에 생성하고, 다양한 아키텍처의 학습용 이미지 데이터셋 증강에 적용할 수 있다.

• 방송공학회논문지
• /
• 제27권1호
• /
• pp.44-55
• /
• 2022

농업 관련 방송 콘텐츠에서 과일에 대한 자동적인 정보 제공을 위해서 대상 과일의 인스턴스 영상 분할이 요구된다. 또한, 해당 과일에 대한 3차원 자세에 대한 정보 제공도 의미있게 사용될 수 있다. 본 논문에서는 영상 콘텐츠에서 토마토에 대한 정보를 제공하는 연구를 다룬다. 인스턴스 영상 분할 기법을 학습하기 위해서는 다량의 데이터가 필요하지만 충분한 토마토 학습데이터를 얻기는 힘들다. 따라서 적은 양의 실사 영상을 바탕으로 데이터 증대기법을 통해 학습 데이터를 생성하였다. 실사 영상만을 통한 학습 결과 정확도에 비해서, 전경과 배경을 분리해서 만들어진 합성 영상을 통해 학습한 결과, 기존 대비 성능이 향상되는 것을 확인하였다. 영상 전처리 기법들을 활용해서 만들어진 영상을 사용한 데이터 증대 영상의 학습 결과, 전경과 배경을 분리한 합성 영상보다 높은 성능을 얻는 것을 확인하였다. 객체 검출 후 자세 추정을 하기 위해 RGB-D 카메라를 이용하여 포인트 클라우드를 획득하였고 최소제곱법을 이용한 실린더 피팅을 진행하였고, 실린더의 축 방향을 통해 토마토 자세를 추정하였다. 우리는 다양한 실험을 통해서 대상 객체에 대한 검출, 인스턴스 영상 분할, 실린더 피팅의 결과가 의미있게 나타난다는 것을 보였다.

• 한국컴퓨터정보학회논문지
• /
• 제28권4호
• /
• pp.65-73
• /
• 2023

본 논문에서는 콘크리트 균열 이미지 데이터셋을 효율적으로 얻기 위한 합성곱 신경망 네트워크 학습 기반의 데이터 증강기법을 제안한다. 실제 콘크리트 균열 이미지는 정형화된 형태가 없고 복잡한 패턴을 지니고 있어 얻기 어려울 뿐만 아니라, 데이터를 확보할 때 위험한 상황에 노출될 우려가 있다. 이러한 상황에 노출된 데이터셋 수집 문제를 본 논문에서는 벡터와 두께 기반의 데이터 증강 기법을 통해 비용과 시간적 측면에서 효율적으로 해결한다. 또한 제안한 방법을 효율성을 입증하고자 U-Net기반의 균열 검출을 통해 다양한 장면에서 실험을 진행했고, IoU 정확도로 측정했을 때 모든 장면에서 성능이 향상되었다. 콘크리트 균열 데이터를 증강하지 않았을 경우 잘못 예측된 경우의 비율이 약 25%였으나, 우리의 방법을 통해 데이터 증강을 했을 경우 잘못 예측된 비율이 3%까지 감소하였다.

• 대한임베디드공학회논문지
• /
• 제19권1호
• /
• pp.9-17
• /
• 2024

It is widely recognized that for 3D human pose estimation (HPE), dataset acquisition is expensive and the effectiveness of augmentation techniques of conventional visual recognition tasks is limited. We address these difficulties by presenting a simple but effective method that augments input images in terms of viewpoints when training a 3D human pose estimation (HPE) model. Our intuition is that meaningful variants of the input images for HPE could be obtained by viewing a human instance in the images from an arbitrary viewpoint different from that in the original images. The core idea is to synthesize new images that have self-occlusion and thus are difficult to predict at different viewpoints even with the same pose of the original example. We incorporate this idea into the training procedure of the 3D HPE model as an augmentation stage of the input samples. We show that a strategy for augmenting the synthesized example should be carefully designed in terms of the frequency of performing the augmentation and the selection of viewpoints for synthesizing the samples. To this end, we propose a new metric to measure the prediction difficulty of input images for 3D HPE in terms of the distance between corresponding keypoints on both sides of a human body. Extensive exploration of the space of augmentation probability choices and example selection according to the proposed distance metric leads to a performance gain of up to 6.2% on Human3.6M, the well-known pose estimation dataset.

• Journal of Multimedia Information System
• /
• 제5권1호
• /
• pp.35-42
• /
• 2018

Direct volume rendering (DVR) is an important 3D visualization method for medical images as it depicts the full volumetric data. However, because DVR renders the whole volume, regions of interests (ROIs) such as a tumor that are embedded within the volume maybe occluded from view. Thus, conventional 2D cross-sectional views are still widely used, while the advantages of the DVR are often neglected. In this study, we propose a new visualization algorithm where we augment the 2D slice of interest (SOI) from an image volume with volumetric information derived from the DVR of the same volume. Our occlusion-based DVR augmentation for SOI (ODAS) uses the occlusion information derived from the voxels in front of the SOI to calculate a depth parameter that controls the amount of DVR visibility which is used to provide 3D spatial cues while not impairing the visibility of the SOI. We outline the capabilities of our ODAS and through a variety of computer tomography (CT) medical image examples, compare it to a conventional fusion of the SOI and the clipped DVR.

• 한국방송∙미디어공학회:학술대회논문집
• /
• 한국방송∙미디어공학회 2019년도 하계학술대회
• /
• pp.163-165
• /
• 2019

The task of image denoising is gaining popularity in the computer vision research field. Its main objective of restoring the sharp image from given noisy input is demanded in all image processing procedure. In this work, we treat the process of residual hair removal on faces images similar to the task of image denoising. In particular, our method removes the residual hair that presents on the frontal or profile face images and in-paints it with the relevant skin color. To achieve this objective, we employ a deep neural network that able to perform both tasks in one time. Furthermore, simple technic of residual hair color augmentation is introduced to increase the number of training data. This approach is beneficial for improving the robustness of the network. Finally, we show that the experimental results demonstrate the superiority of our network in both quantitative and qualitative performances.

• Journal of Multimedia Information System
• /
• 제9권2호
• /
• pp.75-86
• /
• 2022

The existing image quality assessment (IQA) datasets have a small number of samples. Some methods based on transfer learning or data augmentation cannot make good use of image quality-related features. A No Reference (NR)-IQA method based on multi-task training and quality awareness is proposed. First, single or multiple distortion types and levels are imposed on the original image, and different strategies are used to augment different types of distortion datasets. With the idea of weak supervision, we use the Full Reference (FR)-IQA methods to obtain the pseudo-score label of the generated image. Then, we combine the classification information of the distortion type, level, and the information of the image quality score. The ResNet50 network is trained in the pre-train stage on the augmented dataset to obtain more quality-aware pre-training weights. Finally, the fine-tuning stage training is performed on the target IQA dataset using the quality-aware weights to predicate the final prediction score. Various experiments designed on the synthetic distortions and authentic distortions datasets (LIVE, CSIQ, TID2013, LIVEC, KonIQ-10K) prove that the proposed method can utilize the image quality-related features better than the method using only single-task training. The extracted quality-aware features improve the accuracy of the model.