• Journal of KIISE:Databases
• /
• v.36 no.6
• /
• pp.434-445
• /
• 2009

Content-based image retrieval(CBIR) is the retrieval technique which uses the contents of images. However, in contrast to text data, multimedia data are ambiguous and there is a big difference between system's low-level representation and human's high-level concept. So it doesn't always mean that near points in the vector space are similar to user. We call this the semantic-gap problem. Due to this problem, performance of image retrieval is not good. To solve this problem, the relevance feedback(RF) which uses user's feedback information is used. But existing RF doesn't consider user's region-of-interest(ROI), and therefore, irrelevant regions are used in computing new query points. Because the system doesn't know user's ROI, RF is proceeded in the image-level. We propose a new ROI RF method which guides a user to select ROI from relevant images for the retrieval of complex satellite image, and this improves the accuracy of the image retrieval by computing more accurate query points in this paper. Also we propose a pruning technique which improves the accuracy of the image retrieval by using images not selected by the user in this paper. Experiments show the efficiency of the proposed ROI RF and the pruning technique.

• Korean Journal of Remote Sensing
• /
• v.38 no.3
• /
• pp.299-309
• /
• 2022

Recently, The sea water temperature around Korean Peninsula is steadily increasing. Water temperature changes not only affect the fishing ecosystem, but also are closely related to military operations in the sea. The purpose of this study is to suggest which model is more suitable for the field of water temperature prediction by attempting short-term water temperature prediction through various prediction models based on deep learning technology. The data used for prediction are water temperature data from the East Sea (Goseong, Yangyang, Gangneung, and Yeongdeok) from 2016 to 2020, which were observed through marine observation by the National Fisheries Research Institute. In addition, we use Long Short-Term Memory (LSTM), Bidirectional LSTM, and Gated Recurrent Unit (GRU) techniques that show excellent performance in predicting time series data as models for prediction. While the previous study used only LSTM, in this study, the prediction accuracy of each technique and the performance time were compared by applying various techniques in addition to LSTM. As a result of the study, it was confirmed that Bidirectional LSTM and GRU techniques had the least error between actual and predicted values at all observation points based on 1 hour prediction, and GRU was the fastest in learning time. Through this, it was confirmed that a method using Bidirectional LSTM was required for water temperature prediction to improve accuracy while reducing prediction errors. In areas that require real-time prediction in addition to accuracy, such as anti-submarine operations, it is judged that the method of using the GRU technique will be more appropriate.

• The Journal of the Acoustical Society of Korea
• /
• v.41 no.5
• /
• pp.492-499
• /
• 2022

This study experimentally investigates the effect of dimensionality reduction of vibration signal on fault diagnosis of a marine engine. By using the principal component analysis, a vibration signal having the dimension of 513 is converted into a low-dimensional signal having the dimension of 1 to 15, and the variation in fault diagnosis accuracy according to the dimensionality change is observed. The vibration signal measured from a full-scale marine generator diesel engine is used, and the contribution of the dimension-reduced signal is quantitatively evaluated using two kinds of variable importance analysis algorithms which are the integrated gradients and the feature permutation methods. As a result of experimental data analysis, the accuracy of the fault diagnosis is shown to improve as the number of dimensions used increases, and when the dimension approaches 10, near-perfect fault classification accuracy is achieved. This shows that the dimension of the vibration signal can be considerably reduced without degrading fault diagnosis accuracy. In the variable importance analysis, the dimension-reduced principal components show higher contribution than the conventional statistical features, which supports the effectiveness of the dimension-reduced signals on fault diagnosis.

• Asia-pacific Journal of Multimedia Services Convergent with Art, Humanities, and Sociology
• /
• v.7 no.4
• /
• pp.593-601
• /
• 2017

In this paper, we constructed the model of production/investment, distribution, and screening by using variables that can be considered at each stage according to the value chain stage of the movie industry. To increase the predictive power of the model, a regression analysis was used to derive meaningful variables. Based on the given variables, we compared the difference in predictive power between the artificial neural network, which is a machine learning analysis method, and the decision tree analysis method. As a result, the accuracy of artificial neural network was higher than that of decision trees when all variables were added in production/ investment model and distribution model. However, decision trees were more accurate when selected variables were applied according to regression analysis results. In the screening model, the accuracy of the artificial neural network was higher than the accuracy of the decision tree regardless of whether the regression analysis result was reflected or not. This paper has an implication which we tried to improve the performance of movie prediction model by using machine learning analysis. In addition, we tried to overcome a limitation of linear approach by reflecting the results of regression analysis to ANN and decision tree model.

• Journal of the Korea Society of Computer and Information
• /
• v.29 no.2
• /
• pp.31-41
• /
• 2024

Deep learning models show excellent performance in tasks such as image classification and object detection in the field of computer vision, and are used in various ways in actual industrial sites. Recently, research on improving robustness has been actively conducted, along with pointing out that this deep learning model is vulnerable to hostile examples. A hostile example is an image in which small noise is added to induce misclassification, and can pose a significant threat when applying a deep learning model to a real environment. In this paper, we tried to confirm the robustness of the edge-learning classification model and the performance of the adversarial example detection model using it for adversarial examples of various algorithms. As a result of robustness experiments, the basic classification model showed about 17% accuracy for the FGSM algorithm, while the edge-learning models maintained accuracy in the 60-70% range, and the basic classification model showed accuracy in the 0-1% range for the PGD/DeepFool/CW algorithm, while the edge-learning models maintained accuracy in 80-90%. As a result of the adversarial example detection experiment, a high detection rate of 91-95% was confirmed for all algorithms of FGSM/PGD/DeepFool/CW. By presenting the possibility of defending against various hostile algorithms through this study, it is expected to improve the safety and reliability of deep learning models in various industries using computer vision.

• Journal of the Korean Society of Marine Environment & Safety
• /
• v.30 no.5
• /
• pp.407-414
• /
• 2024

This study analyzed the spatiotemporal errors and limitations of LDAPS-based rainfall forecast data used in coastal ocean numerical models and verified their reliability by comparing them with the rainfall data for 2020 from three rain gauges located around Jinhae Bay, South Korea. The results indicated that although LDAPS-based rainfall data generally reproduced long-term trends, they exhibited significant discrepancies in short-term variations. The quantitative error in annual rainfall was 197.5 mm, which increased to 285.4 mm for summer, indicating that the difference in the cumulative rainfall prediction increases for seasons with high rainfall variability. Furthermore, the rainfall-time predictions exhibited a temporal delay of approximately 8 h, suggesting that temporal errors in the LDAPS-based rainfall data could significantly reduce the accuracy of coastal environment predictions. The indiscriminate use of these data, which fail to accurately reflect coastal rainfall, could lead to serious issues in predicting coastal environmental changes caused by pollutants or extreme rainfall events. Therefore, to appropriately use LDAPS-based rainfall data, it is necessary to improve their accuracy through comprehensive verifications and further refinements.

• Elastomers and Composites
• /
• v.35 no.1
• /
• pp.19-28
• /
• 2000

The fundamental study of additional silicone impression material has been performed by comparing the other import products. In order to estimate the possibility of usage of the impression material developed in this study, the several techniques such as IR, EDX, DSC, TGA, rubber rheometer, and contact angle measurement were used. According to the results, there were not any product satisfying all properties required in the impression material. The impression material developed in this study showed best mechanical properties among the all impression materials. However. the wetting property should be studied more by an introduction of a hydrophilic surfactant or modification of a base polymer.

• Journal of the Institute of Electronics Engineers of Korea TC
• /
• v.49 no.5
• /
• pp.7-14
• /
• 2012

The quantized beamforming systems always need the channel state information, which must be quantized into a finite set of vectors (named codebook), and feedback only sends the index representing the desired vector. Thereby it minimized the impact of feedback errors, caused by feedback overhead and delay. In this regard, index assignment (IA) methods, an exhaustive-search and group-based schemes, have been presented for minimizes the performance degradation without additional feedback bits. In this paper, we proposed enhanced group-based IA method, which used the optimal codebook design with chordal distance, having the adaptive properties in application of the existing IA methods. When the number of transmit antennas is 4 and LTE codebook is used, Monte-Carlo simulations verify that the proposed scheme has a power advantage of 0.5~1dB to obtain the same bit error rate than methods without IA, and it has 0.1~0.2 dB better performance compared with the existing IA methods over same environment.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.48 no.2
• /
• pp.47-54
• /
• 2011

This paper describes the method to reduce the time-cost for depth estimation of an image by learning, on the basis of the Genetic Algorithm, the image's features. The depth information is estimated from the relationship among features such as the energy value of an image and the gradient of the texture etc. The estimation-time increases due to the large dimension of an image's features used in the estimating process. And the use of the features without consideration of their importance can have an adverse effect on the performance. So, it is necessary to reduce the dimension of an image's features based on the significance of each feature. Evaluation of the method proposed in this paper using benchmark data provided by Stanford University found that the time-cost for feature extraction and depth estimation improved by 60% and the accuracy was increased by 0.4% on average and up to 2.5%.

• Journal of Digital Contents Society
• /
• v.18 no.6
• /
• pp.1049-1056
• /
• 2017

This study focused on the changes in the voices in determining the effects of speech training. This study aimed to make more visible and scientific evaluation of the changes in the voices among the substantial effects obtained from speech training. As a result, some objective differences from before the speech training could be found in the voice of every learner. Each learner showed gradual technical improvement in a variety of vocal elements, including resonance and timbre, accuracy of pronunciation, pause; that is, the voice became more powerful, more accurate pronounced, more pausing and more stable than before the speech training. This study determined if speech training could change a voice and the results are expected to help speech learners participate actively in speech training and see their speech ability improved.