• Journal of Convergence for Information Technology
• /
• v.10 no.8
• /
• pp.23-34
• /
• 2020

Recently, the artificial intelligence deep learning field has been hard to commercialize due to the high computing power and the price problem of computing resources. In this paper, we apply a double pruning techniques to evaluate the performance of the in-depth neural network and various datasets. Double pruning combines basic Network-slimming and Parameter-prunning. Our proposed technique has the advantage of reducing the parameters that are not important to the existing learning and improving the speed without compromising the learning accuracy. After training various datasets, the pruning ratio was increased to reduce the size of the model.We confirmed that MobileNet-V3 showed the highest performance as a result of NetScore performance analysis. We confirmed that the performance after pruning was the highest in MobileNet-V3 consisting of depthwise seperable convolution neural networks in the Cifar 10 dataset, and VGGNet and ResNet in traditional convolutional neural networks also increased significantly.

• Journal of Korea Water Resources Association
• /
• v.53 no.12
• /
• pp.1159-1172
• /
• 2020

This study applied deep convolution neural network based on U-Net and SegNet using long period weather radar data to very short-term rainfall prediction. And the results were compared and evaluated with the translation model. For training and validation of deep neural network, Mt. Gwanak and Mt. Gwangdeoksan radar data were collected from 2010 to 2016 and converted to a gray-scale image file in an HDF5 format with a 1km spatial resolution. The deep neural network model was trained to predict precipitation after 10 minutes by using the four consecutive radar image data, and the recursive method of repeating forecasts was applied to carry out lead time 60 minutes with the pretrained deep neural network model. To evaluate the performance of deep neural network prediction model, 24 rain cases in 2017 were forecast for rainfall up to 60 minutes in advance. As a result of evaluating the predicted performance by calculating the mean absolute error (MAE) and critical success index (CSI) at the threshold of 0.1, 1, and 5 mm/hr, the deep neural network model showed better performance in the case of rainfall threshold of 0.1, 1 mm/hr in terms of MAE, and showed better performance than the translation model for lead time 50 minutes in terms of CSI. In particular, although the deep neural network prediction model performed generally better than the translation model for weak rainfall of 5 mm/hr or less, the deep neural network prediction model had limitations in predicting distinct precipitation characteristics of high intensity as a result of the evaluation of threshold of 5 mm/hr. The longer lead time, the spatial smoothness increase with lead time thereby reducing the accuracy of rainfall prediction The translation model turned out to be superior in predicting the exceedance of higher intensity thresholds (> 5 mm/hr) because it preserves distinct precipitation characteristics, but the rainfall position tends to shift incorrectly. This study are expected to be helpful for the improvement of radar rainfall prediction model using deep neural networks in the future. In addition, the massive weather radar data established in this study will be provided through open repositories for future use in subsequent studies.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2021.05a
• /
• pp.43-45
• /
• 2021

The short-term quantitative precipitation prediction (QPF) system is important socially and economically to prevent damage from severe weather. Recently, many studies for short-term QPF model applying the Deep Neural Network (DNN) has been conducted. These studies require the sophisticated pre-processing because the mistreatment of various and vast meteorological data sets leads to lower performance of QPF. Especially, for more accurate prediction of the non-linear trends in precipitation, the dataset needs to be carefully handled based on the physical and dynamical understands the data. Thereby, this paper proposes the following approaches: i) refining and combining major factors (weather radar, terrain, air temperature, and so on) related to precipitation development in order to construct training data for pattern analysis of precipitation; ii) producing predicted precipitation fields based on Convolutional with ConvLSTM. The proposed algorithm was evaluated by rainfall events in 2020. It is outperformed in the magnitude and strength of precipitation, and clearly predicted non-linear pattern of precipitation. The algorithm can be useful as a forecasting tool for preventing severe weather.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2018.06a
• /
• pp.213-216
• /
• 2018

본 논문에서는 사람의 인지 시각 특성 중 하나인 JND(Just Noticeable Difference)를 이용한 인지 비디오 부호화 기법을 제안한다. JND 기반 인지 부호화 방법은 사람의 인지 시각 특성을 이용해 시각적으로 인지가 잘 되지 않는 인지 신호를 제거함으로 부호화 효율을 높이는 방법이다. 제안된 방법은 기존 수학적 모델 기반의 JND 기법이 아닌 최근 각광 받고 있는 데이터 중심(data-driven) 모델링 방법인 심층 신경망 기반 JND 모델 생성 기법을 제안한다. 제안된 심층 신경망 기반 JND 모델은 비디오 부호화 과정에서 입력 영상에 대한 전처리를 통해 입력 영상의 인지 중복(perceptual redundancy)를 제거하는 역할을 수행한다. 부호화 실험에서 제안된 방법은 동일하거나 유사한 인지화질을 유지한 상태에서 평균 16.86 %의 부호화 비트를 감소 시켰다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2020.05a
• /
• pp.555-558
• /
• 2020

본 논문에서는 입력 영상에 담긴 다양한 물체들과 그들 간의 관계를 효과적으로 탐지하여, 하나의 장면 그래프로 표현해내는 새로운 심층 신경망 모델을 제안한다. 제안 모델에서는 물체와 관계의 효과적인 탐지를 위해, 합성 곱 신경망 기반의 시각 맥락 특징들뿐만 아니라 언어 맥락 특징들을 포함하는 다양한 멀티 모달 맥락 정보들을 활용한다. 또한, 제안 모델에서는 관계를 맺는 두 물체 간의 상호 의존성이 그래프 노드 특징값들에 충분히 반영되도록, 그래프 신경망을 이용해 맥락 정보를 임베딩한다. 본 논문에서는 Visual Genome 벤치마크 데이터 집합을 이용한 비교 실험들을 통해, 제안 모델의 효과와 성능을 입증한다.

• The Journal of the Convergence on Culture Technology
• /
• v.8 no.6
• /
• pp.867-871
• /
• 2022

Amid the recent rapid trend change, the change in design has a great impact on the sales of fashion companies, so it is inevitable to be careful in choosing new designs. With the recent development of the artificial intelligence field, various machine learning is being used a lot in the fashion market to increase consumers' preferences. To contribute to increasing reliability in the development of new products by quantifying abstract concepts such as preferences, we generate new images that do not exist through three adversarial generative neural networks (GANs) and numerically compare abstract concepts of preferences using pre-trained convolution neural networks (CNNs). Deep convolutional generative adversarial networks (DCGAN), Progressive growing adversarial networks (PGGAN), and Dual Discriminator generative adversarial networks (DANs), which were trained to produce comparative, high-level, and high-level images. The degree of similarity measured was considered as a preference, and the experimental results showed that D2GAN showed a relatively high similarity compared to DCGAN and PGGAN.

• Journal of the Institute of Convergence Signal Processing
• /
• v.23 no.4
• /
• pp.234-240
• /
• 2022

This paper proposes a customized AI exercise recommendation service for balancing the relative amount of exercise according to the working environment by each occupation. WISDM database is collected by using acceleration and gyro sensors, and is a dataset that classifies physical activities into 18 categories. Our system recommends a adaptive exercise using the analyzed activity type after classifying 18 physical activities into 3 physical activities types such as whole body, upper body and lower body. 1 Dimensional convolutional neural network is used for classifying a physical activity in this paper. Proposed model is composed of a convolution blocks in which 1D convolution layers with a various sized kernel are connected in parallel. Convolution blocks can extract a detailed local features of input pattern effectively that can be extracted from deep neural network models, as applying multi 1D convolution layers to input pattern. To evaluate performance of the proposed neural network model, as a result of comparing the previous recurrent neural network, our method showed a remarkable 98.4% accuracy.

• Journal of Broadcast Engineering
• /
• v.26 no.4
• /
• pp.419-428
• /
• 2021

Shadow removal plays a key role for the pre-processing of image processing techniques such as object tracking and detection. With the advances of image recognition based on deep convolution neural networks, researches for shadow removal have been actively conducted. In this paper, we propose a novel method for shadow removal, which utilizes self attention distillation to extract semantic features. The proposed method gradually refines results of shadow detection, which are extracted from each layer of the proposed network, via top-down distillation. Specifically, the training procedure can be efficiently performed by learning the contextual information for shadow removal without shadow masks. Experimental results on various datasets show the effectiveness of the proposed method for shadow removal under real world environments.

• KIPS Transactions on Software and Data Engineering
• /
• v.7 no.5
• /
• pp.195-204
• /
• 2018

In this paper, we propose both a new loss function and an online hard example mining scheme for improving the performance of single-stage object detectors which use deep convolutional neural networks. The proposed loss function and the online hard example mining scheme can not only overcome the problem of imbalance between the number of annotated objects and the number of background examples, but also improve the localization accuracy of each object. Therefore, the loss function and the mining scheme can provide intrinsically fast single-stage detectors with detection performance higher than or similar to that of two-stage detectors. In experiments conducted with the PASCAL VOC 2007 benchmark dataset, we show that the proposed loss function and the online hard example mining scheme can improve the performance of single-stage object detectors.

• Journal of the Korea Convergence Society
• /
• v.10 no.9
• /
• pp.23-29
• /
• 2019

In this paper, the mood of images was classified into eight categories through a deep convolutional neural network and video was automatically generated using proper background music. Based on the collected image data, the classification model is learned using a multilayer perceptron (MLP). Using the MLP, a video is generated by using multi-class classification to predict image mood to be used for video generation, and by matching pre-classified music. As a result of 10-fold cross-validation and result of experiments on actual images, each 72.4% of accuracy and 64% of confusion matrix accuracy was achieved. In the case of misclassification, by classifying video into a similar mood, it was confirmed that the music from the video had no great mismatch with images.