• Journal of Information Processing Systems
• /
• v.17 no.2
• /
• pp.337-351
• /
• 2021

Automatically recognizing facial expressions in video sequences is a challenging task because there is little direct correlation between facial features and subjective emotions in video. To overcome the problem, a video facial expression recognition method using spatiotemporal recurrent neural network and feature fusion is proposed. Firstly, the video is preprocessed. Then, the double-layer cascade structure is used to detect a face in a video image. In addition, two deep convolutional neural networks are used to extract the time-domain and airspace facial features in the video. The spatial convolutional neural network is used to extract the spatial information features from each frame of the static expression images in the video. The temporal convolutional neural network is used to extract the dynamic information features from the optical flow information from multiple frames of expression images in the video. A multiplication fusion is performed with the spatiotemporal features learned by the two deep convolutional neural networks. Finally, the fused features are input to the support vector machine to realize the facial expression classification task. The experimental results on cNTERFACE, RML, and AFEW6.0 datasets show that the recognition rates obtained by the proposed method are as high as 88.67%, 70.32%, and 63.84%, respectively. Comparative experiments show that the proposed method obtains higher recognition accuracy than other recently reported methods.

• Phonetics and Speech Sciences
• /
• v.13 no.1
• /
• pp.45-51
• /
• 2021

End-to-end (E2E) automatic speech recognition (ASR) has achieved promising performance gains with the introduced self-attention network, Transformer. However, due to training time and the number of hyperparameters, finding the optimal hyperparameter set is computationally expensive. This paper investigates the impact of hyperparameters in the Transformer network to answer two questions: which hyperparameter plays a critical role in the task performance and training speed. The Transformer network for training has two encoder and decoder networks combined with Connectionist Temporal Classification (CTC). We have trained the model with Wall Street Journal (WSJ) SI-284 and tested on devl93 and eval92. Seventeen hyperparameters were selected from the ESPnet training configuration, and varying ranges of values were used for experiments. The result shows that "num blocks" and "linear units" hyperparameters in the encoder and decoder networks reduce Word Error Rate (WER) significantly. However, performance gain is more prominent when they are altered in the encoder network. Training duration also linearly increased as "num blocks" and "linear units" hyperparameters' values grow. Based on the experimental results, we collected the optimal values from each hyperparameter and reduced the WER up to 2.9/1.9 from dev93 and eval93 respectively.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.33 no.6
• /
• pp.1033-1042
• /
• 2023

This paper proposes a network intrusion detection system that identifies abnormal flows within the network. The majority of datasets commonly used in research lack time-series information, making it challenging to improve detection rates for attacks with fewer instances due to a scarcity of sample data. However, there is insufficient research regarding detection approaches. In this study, we build upon previous research by using the Artificial neural network(ANN) model and a stack ensemble technique in our approach. To address the aforementioned issues, we incorporate temporal information by leveraging adjacent flows and enhance the learning of samples from sparse attacks, thereby improving both the overall detection rate and the detection rate for sparse attacks.

• Journal of the Institute of Convergence Signal Processing
• /
• v.4 no.4
• /
• pp.30-39
• /
• 2003

In this paper, we carried out the performance evaluation of HM-Net(Hidden Markov Network) speech recognition system for Korean speech databases. We adopted to construct acoustic models using the HM-Nets modified by HMMs(Hidden Markov Models), which are widely used as the statistical modeling methods. HM-Nets are carried out the state splitting for contextual and temporal domain by PDT-SSS(Phonetic Decision Tree-based Successive State Splitting) algorithm, which is modified the original SSS algorithm. Especially it adopted the phonetic decision tree to effectively express the context information not appear in training speech data on contextual domain state splitting. In case of temporal domain state splitting, to effectively represent information of each phoneme maintenance in the state splitting is carried out, and then the optimal model network of triphone types are constructed by in the parameter. Speech recognition was performed using the one-pass Viterbi beam search algorithm with phone-pair/word-pair grammar for phoneme/word recognition, respectively and using the multi-pass search algorithm with n-gram language models for sentence recognition. The tree-structured lexicon was used in order to decrease the number of nodes by sharing the same prefixes among words. In this paper, the performance evaluation of HM-Net speech recognition system is carried out for various recognition conditions. Through the experiments, we verified that it has very superior recognition performance compared with the previous introduced recognition system.

• KIPS Transactions on Software and Data Engineering
• /
• v.8 no.3
• /
• pp.115-122
• /
• 2019

For Korean phoneme recognition, Hidden Markov-Gaussian Mixture model(HMM-GMM) or hybrid models which combine artificial neural network with HMM have been mainly used. However, current approach has limitations in that such models require force-aligned corpus training data that is manually annotated by experts. Recently, researchers used neural network based phoneme recognition model which combines recurrent neural network(RNN)-based structure with connectionist temporal classification(CTC) algorithm to overcome the problem of obtaining manually annotated training data. Yet, in terms of implementation, these RNN-based models have another difficulty in that the amount of data gets larger as the structure gets more sophisticated. This problem of large data size is particularly problematic in the Korean language, which lacks refined corpora. In this study, we introduce CTC algorithm that does not require force-alignment to create a Korean phoneme recognition model. Specifically, the phoneme recognition model is based on convolutional neural network(CNN) which requires relatively small amount of data and can be trained faster when compared to RNN based models. We present the results from two different experiments and a resulting best performing phoneme recognition model which distinguishes 49 Korean phonemes. The best performing phoneme recognition model combines CNN with 3hop Bidirectional LSTM with the final Phoneme Error Rate(PER) at 3.26. The PER is a considerable improvement compared to existing Korean phoneme recognition models that report PER ranging from 10 to 12.

• Korean Journal of Remote Sensing
• /
• v.31 no.3
• /
• pp.205-214
• /
• 2015

In this study, a novel land-cover classification framework for multi-temporal SAR data is presented that can combine multiple features extracted through data transforms and multiple classifiers. At first, data transforms using principle component analysis (PCA) and 3D wavelet transform are applied to multi-temporal SAR dataset for extracting new features which were different from original dataset. Then, three different classifiers including maximum likelihood classifier (MLC), neural network (NN) and support vector machine (SVM) are applied to three different dataset including data transform based features and original backscattering coefficients, and as a result, the diverse preliminary classification results are generated. These results are combined via a majority voting rule to generate a final classification result. From an experiment with a multi-temporal ENVISAT ASAR dataset, every preliminary classification result showed very different classification accuracy according to the used feature and classifier. The final classification result combining nine preliminary classification results showed the best classification accuracy because each preliminary classification result provided complementary information on land-covers. The improvement of classification accuracy in this study was mainly attributed to the diversity from combining not only different features based on data transforms, but also different classifiers. Therefore, the land-cover classification framework presented in this study would be effectively applied to the classification of multi-temporal SAR data and also be extended to multi-sensor remote sensing data fusion.

• The Transactions of the Korea Information Processing Society
• /
• v.7 no.1
• /
• pp.166-180
• /
• 2000

Recommendations of ITU-T and ISO stipulate the managerial abstraction of static and dynamic characteristics of network elements, management functions as well as management communication protocol. The current recommendations provide the formal mechanism for the structural parts of managed objects such as managed object class and attributes. But the current description method does not provide the formal mechanism for the behavioral characteristics of managed objects in clear manner but in natural language form, the complete specification of managed objects is not fully described. Also, the behaviour of managed objects is affected by their temporal and active properties. While the temporal properties representing periodic or repetitive internals are to describe managed objects behaviour in rather strict way, it will be more powerful if more dynamic temporal properties determined by external conditions are added to managed objects. In this paper, we added dynamic features to scheduling managed objects, and described, in GDMO, scheduling managed objects that support dynamic features. We also described behaviour of managed objects in newly defined BDL that has dynamic temporal properties. This paper showed that dynamic temporal managed objects provide a systematic and formal method in agent management function model.

• Journal of the Korea Convergence Society
• /
• v.10 no.12
• /
• pp.17-22
• /
• 2019

Network attack analysis and visualization methods using network traffic session data detect network anomalies by visualizing the sender's and receiver's IP addresses and the relationship between them. The traffic flow is a critical feature in detecting anomalies, but simply visualizing the source and destination IP addresses symmetrically from up-down or left-right would become a problematic factor for the analysis. Also, there is a risk of losing timely security situation when designing a visualization interface without considering the temporal characteristics of time-series traffic sessions. In this paper, we propose a visualization interface and analysis method that visualizes time-series traffic data by using the radial axis, divide IP addresses into network and host portions which then projects on the cylindrical coordinate system that could effectively monitor network attacks. The proposed method has the advantage of intuitively recognizing network attacks and identifying attack activity over time.

• Journal of Korea Water Resources Association
• /
• v.52 no.1
• /
• pp.51-60
• /
• 2019

The major purpose of this study is to construct an in-situ soil moisture verification network employing Frequency Domain Reflectometry (FDR) sensors for Cosmic-ray soil moisture observation system operation as well as long-term field-scale soil moisture monitoring. The test bed of Cosmic-ray and FDR verification network system was established at the Sulma Catchment, in connection with the existing instrumentations for integrated data provision of various hydrologic variables. This test bed includes one Cosmic-ray Neutron Probe (CRNP) and ten FDR stations with four different measurement depths (10 cm, 20 cm, 30 cm, and 40 cm) at each station, and has been operating since July 2018. Furthermore, to assess the reliability of the in-situ verification network, the volumetric water content data measured by FDR sensors were compared to those calculated through the core sampling method. The evaluation results of FDR sensors- measured soil moisture against sampling method during the study period indicated a reasonable agreement, with average values of $bias=-0.03m^3/m^3$ and RMSE $0.03m^3/m^3$, revealing that this FDR network is adequate to provide long-term reliable field-scale soil moisture monitoring at Sulmacheon basin. In addition, soil moisture time series observed at all FDR stations during the study period generally respond well to the rainfall events; and at some locations, the characteristics of rainfall water intercepted by canopy were also identified. The Temporal Stability Analysis (TSA) was performed for all FDR stations located within the CRNP footprint at each measurement depth to determine the representative locations for field-average soil moisture at different soil profiles of the verification network. The TSA results showed that superior performances were obtained at FDR 5 for 10 cm depth, FDR 8 for 20 cm depth, FDR2 for 30 cm depth, and FDR1 for 40 cm depth, respectively; demonstrating that those aforementioned stations can be regarded as temporal stable locations to represent field mean soil moisture measurements at their corresponding measurement depths. Although the limit on study duration has been presented, the analysis results of this study can provide useful knowledge on soil moisture variability and stability at the test bed, as well as supporting the utilization of the Cosmic-ray observation system for long-term field-scale soil moisture monitoring.

• Proceedings of the Korean Society for Bioinformatics Conference
• /
• 2005.09a
• /
• pp.373-378
• /
• 2005

The diagrammatic language for pathways is widely used for representing systems knowledge as a network of causal relations. Biologists infer and hypothesize with pathways to design experiments and verify models, and to identify potential drug targets. Although there have been many approaches to formalize pathways to simulate a system, reasoning with incomplete and high level knowledge has not been possible. We present a qualitative formalization of a pathway language with incomplete causal descriptions and its translation into propositional temporal logic to automate the reasoning process. Such automation accelerates the identification of drug targets in pathways.