• KIPS Transactions on Software and Data Engineering
• /
• v.11 no.3
• /
• pp.125-132
• /
• 2022

Sentence compression is a natural language processing task that generates concise sentences that preserves the important meaning of the original sentence. For grammatically appropriate sentence compression, early studies utilized human-defined linguistic rules. Furthermore, while the sequence-to-sequence models perform well on various natural language processing tasks, such as machine translation, there have been studies that utilize it for sentence compression. However, for the linguistic rule-based studies, all rules have to be defined by human, and for the sequence-to-sequence model based studies require a large amount of parallel data for model training. In order to address these challenges, Deleter, a sentence compression model that leverages a pre-trained language model BERT, is proposed. Because the Deleter utilizes perplexity based score computed over BERT to compress sentences, any linguistic rules and parallel dataset is not required for sentence compression. However, because Deleter compresses sentences only considering perplexity, it does not compress sentences by reflecting the linguistic information of the words in the sentences. Furthermore, since the dataset used for pre-learning BERT are far from compressed sentences, there is a problem that this can lad to incorrect sentence compression. In order to address these problems, this paper proposes a method to quantify the importance of linguistic information and reflect it in perplexity-based sentence scoring. Furthermore, by fine-tuning BERT with a corpus of news articles that often contain proper nouns and often omit the unnecessary modifiers, we allow BERT to measure the perplexity appropriate for sentence compression. The evaluations on the English and Korean dataset confirm that the sentence compression performance of sentence-scoring based models can be improved by utilizing the proposed method.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.17 no.5
• /
• pp.897-904
• /
• 2022

Sports video is a very critical information resource. High-precision extraction of effective segments in sports video can better assist coaches in analyzing the player's actions in the video, and enable users to more intuitively appreciate the player's hitting action. Aiming at the shortcomings of the current sports video clip extraction results, such as strong subjectivity, large workload and low efficiency, a classification method of sports video clips based on MobileNetV3 is proposed to save user time. Experiments evaluate the effectiveness of effective segment extraction. Among the extracted segments, the effective proportion is 97.0%, indicating that the effective segment extraction results are good, and it can lay the foundation for the construction of the subsequent badminton action metadata video dataset.

• Smart Media Journal
• /
• v.12 no.1
• /
• pp.9-16
• /
• 2023

Recently, a convolutional neural network (CNN) based system is being developed to overcome the limitations of human resources in the apple quality classification of farmhouse. However, since convolutional neural networks receive only images of the same size, preprocessing such as sampling may be required, and in the case of oversampling, information loss of the original image such as image quality degradation and blurring occurs. In this paper, in order to minimize the above problem, to generate a image patch based graph of an original image and propose a random walk-based positional encoding method to apply the graph transformer model. The above method continuously learns the position embedding information of patches which don't have a positional information based on the random walk algorithm, and finds the optimal graph structure by aggregating useful node information through the self-attention technique of graph transformer model. Therefore, it is robust and shows good performance even in a new graph structure of random node order and an arbitrary graph structure according to the location of an object in an image. As a result, when experimented with 5 apple quality datasets, the learning accuracy was higher than other GNN models by a minimum of 1.3% to a maximum of 4.7%, and the number of parameters was 3.59M, which was about 15% less than the 23.52M of the ResNet18 model. Therefore, it shows fast reasoning speed according to the reduction of the amount of computation and proves the effect.

• Tunnel and Underground Space
• /
• v.29 no.3
• /
• pp.184-196
• /
• 2019

This study aims to extract a 3D image of micro-cracks generated by hydraulic fracturing tests, using the deep learning method and X-ray computed tomography images. The pixel-level cracks are difficult to be detected via conventional image processing methods, such as global thresholding, canny edge detection, and the region growing method. Thus, the convolutional neural network-based encoder-decoder network is adapted to extract and analyze the micro-crack quantitatively. The number of training data can be acquired by dividing, rotating, and flipping images and the optimum combination for the image augmentation method is verified. Application of the optimal image augmentation method shows enhanced performance for not only the validation dataset but also the test dataset. In addition, the influence of the original number of training data to the performance of the deep learning-based neural network is confirmed, and it leads to succeed the pixel-level crack detection.

• Smart Media Journal
• /
• v.13 no.3
• /
• pp.9-17
• /
• 2024

Due to the recent pandemic and the development of ICT technology, the use of non-face-to-face and unmanned systems is expanding, and it is very important to understand emotions in communication in non-face-to-face situations. As emotion recognition methods for various facial expressions are required to understand emotions, artificial intelligence-based research is being conducted to improve facial expression emotion recognition in image data. However, existing research on facial expression emotion recognition requires high computing power and a lot of learning time because it utilizes a large amount of data to improve accuracy. To improve these limitations, this paper proposes a method of recognizing facial expressions using age and gender, which are image meta information, as a method of recognizing facial expressions with even a small amount of data. For facial expression emotion recognition, a face was detected using the Yolo Face model from the original image data, and age and gender were classified through the VGG model based on image meta information, and then seven emotions were recognized using the EfficientNet model. The accuracy of the proposed data classification learning model was higher as a result of comparing the meta-information-based data classification model with the model trained with all data.

• Journal of Korea Society of Industrial Information Systems
• /
• v.21 no.6
• /
• pp.13-21
• /
• 2016

Feature selection is the one of important issue in the field of data mining and machine learning. It is the technique to find a subset of features which provides the best classification performance, from the source data. We propose a feature subset selection method using the neighborhood rough set model based on information granularity. To demonstrate the effectiveness of proposed method, it was applied to select the useful features associated with breast tumor diagnosis of 298 shape features extracted from 5,252 breast ultrasound images, which include 2,745 benign and 2,507 malignant cases. Experimental results showed that 19 diagnostic features were strong predictors of breast cancer diagnosis and then average classification accuracy was 97.6%.

• Journal of Internet Computing and Services
• /
• v.20 no.5
• /
• pp.1-8
• /
• 2019

This paper carried out a study to reduce damage from jellyfish whose population has increased due to global warming. The emergence of jellyfish on the beach could result in casualties from jellyfish stings and economic losses from closures. This paper confirmed from the preceding studies that the pattern of jellyfish's appearance is predictable through machine learning. This paper is an extension of The prediction model of emergence of Busan coastal jellyfish using SVM. In this paper, we used deep neural network to expand from the existing methods of predicting the existence of jellyfish to the classification by index. Due to the limitations of the small amount of data collected, the 84.57% prediction accuracy limit was sought to be resolved through data expansion using bootstraping. The expanded data showed about 7% higher performance than the original data, and about 6% better performance compared to the transfer learning. Finally, we used the test data to confirm the prediction performance of jellyfish appearance. As a result, although it has been confirmed that jellyfish emergence binary classification can be predicted with high accuracy, predictions through indexation have not produced meaningful results.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2022.05a
• /
• pp.46-46
• /
• 2022

유기물의 복잡한 혼합물인 CDOM(Colored or Chromophoric Dissolved Organic Matter)은 하천 내 BOD(Biological Oxygen Demand), COD(Chemical Oxygen Demand) 및 유기 오염물질과 상당한 관련이 있다. CDOM은 가시광선 영역에서 빛을 흡수하는 성질을 가지고 있으며, 최근 원격감지 기술로 CDOM을 모니터링하기 위한 연구가 진행되고 있다. 본 연구에서는 백제보 상류 23km 구간에서 3년(2016~2018) 중 13일의 초분광영상을 활용하여 머신러닝 기반 CDOM을 추정 알고리즘을 개발하고자 한다. 초분광영상은 400~970 nm의 범위의 4 nm 간격 127개 대역의 분광해상도와 2 m의 공간해상도를 가진 항공기 탑재 AsiaFENIX 초분광 센서를 통해 수집하였으며 CDOM은 Millipore polycarbonate filter (𝚽47, 0.2 ㎛)에서 여과된 CDOM 샘플 자료를 200~800 nm의 흡수계수 스펙트럼으로 추출하여 사용하였다. CDOM 값은 전체기간 동안 2.0~11.0 m^-1의 값 분포를 보였으며 5 m^-1이상의 고농도 구간 자료개수가 전체 153개 샘플자료 중 21개로 불균형하다. 따라서 ADASYN(Adaptive Synthesis Sampling Approach)의 oversampling 방법으로 생성된 합성 데이터를 사용하여 원본 데이터의 소수계층 데이터 불균형을 해결하고 모델 예측 성능을 개선하고자 하였다. 생성된 합성 데이터를 입력변수로 하여 ANN(Artificial Neural Netowk)을 활용한 CDOM 예측 알고리즘을 구축하였다. ADASYN 기법을 통한 합성 데이터는 관측된 데이터의 불균형을 해결하여 기계학습 모델의 CDOM 탐지 성능을 향상시킬 수 있으며, 저수지 내 유기 오염물질 관리를 위한 설계를 지원하는데 사용할 수 있을 것으로 판단된다.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.32 no.5
• /
• pp.987-995
• /
• 2022

Deep learning-based profiling side-channel analysis has been many proposed. Deep learning-based profiling analysis is a technique that trains the relationship between the side-channel information and the intermediate values to the neural network, then finds the secret key of the attack device using the trained neural network. Recently, cross-device profiling side channel analysis was proposed to consider the realistic deep learning-based profiling side channel analysis scenarios. However, it has a limitation in that attack performance is lowered if the profiling device and the attack device have not the same chips. In this paper, an environment in which the profiling device and the attack device have not the same chips is defined as the different-device, and a novel deep learning-based profiling side-channel analysis on different-device is proposed. Also, MCNN is used to well extract the characteristic of each data. We experimented with the six different boards to verify the attack performance of the proposed method; as a result, when the proposed method was used, the minimum number of attack traces was reduced by up to 25 times compared to without the proposed method.

• Smart Media Journal
• /
• v.10 no.4
• /
• pp.21-27
• /
• 2021

Recently, the importance of counseling is increasing due to the Corona Blue caused by COVID-19. Also, with the increase of non-face-to-face services, researches on chatbots that have changed the counseling media are being actively conducted. In non-face-to-face counseling through chatbot, it is most important to accurately understand the client's emotions. However, since there is a limit to recognizing emotions only in sentences written by the client, it is necessary to recognize the dimensional emotions embedded in the sentences for more accurate emotion recognition. Therefore, in this paper, the vector and sentence VAD (Valence, Arousal, Dominance) generated by learning the Word2Vec model after correcting the original data according to the characteristics of the data are learned using a deep learning algorithm to learn the multi-dimensional We propose an emotion recognition model. As a result of comparing three deep learning models as a method to verify the usefulness of the proposed model, R-squared showed the best performance with 0.8484 when the attention model is used.