• Current Optics and Photonics
• /
• v.7 no.3
• /
• pp.237-247
• /
• 2023

In middle- and long-distance imaging systems, due to the atmospheric turbulence caused by temperature, wind speed, humidity, and so on, light waves propagating in the air are distorted, resulting in image-quality degradation such as geometric deformation and fuzziness. In remote sensing, astronomical observation, and traffic monitoring, image information loss due to degradation causes huge losses, so effective restoration of degraded images is very important. To restore images degraded by atmospheric turbulence, an image-restoration method based on improved compound multibranch feature fusion (CMFNetPro) was proposed. Based on the CMFNet network, an efficient channel-attention mechanism was used to replace the channel-attention mechanism to improve image quality and network efficiency. In the experiment, two-dimensional random distortion vector fields were used to construct two turbulent datasets with different degrees of distortion, based on the Google Landmarks Dataset v2 dataset. The experimental results showed that compared to the CMFNet, DeblurGAN-v2, and MIMO-UNet models, the proposed CMFNetPro network achieves better performance in both quality and training cost of turbulent-image restoration. In the mixed training, CMFNetPro was 1.2391 dB (weak turbulence), 0.8602 dB (strong turbulence) respectively higher in terms of peak signal-to-noise ratio and 0.0015 (weak turbulence), 0.0136 (strong turbulence) respectively higher in terms of structure similarity compared to CMFNet. CMFNetPro was 14.4 hours faster compared to the CMFNet. This provides a feasible scheme for turbulent-image restoration based on deep learning.

• Journal of Digital Policy
• /
• v.2 no.3
• /
• pp.1-7
• /
• 2023

In this paper, I proposes a method for performing single image super resolution by separating texture-spatial domains and then classifying features based on detailed information. In CNN (Convolutional Neural Network) based super resolution, the complex procedures and generation of redundant feature information in feature estimation process for enhancing details can lead to quality degradation in super resolution. The proposed method reduced procedural complexity and minimizes generation of redundant feature information by splitting input image into two channels: texture and spatial. In texture channel, a feature refinement process with step-wise skip connections is applied for detail restoration, while in spatial channel, a method is introduced to preserve the structural features of the image. Experimental results using proposed method demonstrate improved performance in terms of PSNR and SSIM evaluations compared to existing super resolution methods, confirmed the enhancement in quality.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.22 no.1
• /
• pp.103-114
• /
• 2023

This study collected various data of urban roadways to analyze the effect of travel speed change, and a GRU-based short-term travel speed prediction model was developed using such big data. The baseline model and the double exponential smoothing model were selected as comparison models, and prediction errors were evaluated using the RMSE index. The model evaluation results revealed that the average RMSE of the baseline model and the double exponential smoothing model were 7.46 and 5.94, respectively. The average RMSE predicted by the GRU model was 5.08. Although there are deviations for each of the 15 links, most cases showed minimal errors in the GRU model, and the additional scatter plot analysis presented the same result. These results indicate that the prediction error can be reduced, and the model application speed can be improved when applying the GRU-based model in the process of generating travel speed information on urban roadways.

• Journal of Korean Society on Water Environment
• /
• v.39 no.1
• /
• pp.46-60
• /
• 2023

Climate change causes fluctuations in water quality in the aquatic environment, which can cause changes in water circulation patterns and severe adverse effects on aquatic ecosystems in the future. Therefore, research is needed to predict and respond to water quality changes caused by climate change in advance. In this study, we tried to predict the dissolved oxygen (DO), chlorophyll-a, and turbidity of the Paldang reservoir for about two weeks using long short-term memory (LSTM) and gated recurrent units (GRU), which are deep learning algorithms based on recurrent neural networks. The model was built based on real-time water quality data and meteorological data. The observation period was set from July to September in the summer of 2021 (Period 1) and from March to May in the spring of 2022 (Period 2). We tried to select an algorithm with optimal predictive power for each water quality parameter. In addition, to improve the predictive power of the model, an important variable extraction technique using random forest was used to select only the important variables as input variables. In both Periods 1 and 2, the predictive power after extracting important variables was further improved. Except for DO in Period 2, GRU was selected as the best model in all water quality parameters. This methodology can be useful for preventive water quality management by identifying the variability of water quality in advance and predicting water quality in a short period.

• The Journal of the Acoustical Society of Korea
• /
• v.41 no.2
• /
• pp.235-241
• /
• 2022

Several studies using Convolutional Neural Network (CNN) have been conducted to detect and classify the sounds of marine mammals in underwater acoustic data collected through passive acoustic monitoring. In this study, the possibility of automatic classification of bearded seal sounds was confirmed using a CNN model based on the underwater acoustic spectrogram images collected from August 2017 to August 2018 in East Siberian Sea. When only the clear seal sound was used as training dataset, overfitting due to memorization was occurred. By evaluating the entire training data by replacing some training data with data containing noise, it was confirmed that overfitting was prevented as the model was generalized more than before with accuracy (0.9743), precision (0.9783), recall (0.9520). As a result, the performance of the classification model for bearded seals signal has improved when the noise was included in the training data.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.11a
• /
• pp.377-379
• /
• 2020

The hologram data is having a dependence on the pixel pitch of the SLM (spatial light modulator) and the wavelength of light, and the quality of the digital hologram is proportional to the unit pixel pitch and the total resolution. In addition, since each pixel has a complex value, the amount of data in the digital hologram also increases exponentially, and the size is bound to be very large. Therefore, in order to efficiently handle digital hologram files, it is essential to reduce the file size through a codec and store it. Recently, research on enhancing image quality damaged by the codec is actively underway. In this paper, the hologram image of JPEG Pleno, which is the standard hologram data, was used, and the image quality damage that occurs whenthe holographic image is encoded and decoded through the JPEG2000, AVC, and HEVC codec is enhanced with a deep learning network to find out whether the image quality can be improved. we also compare and quantitatively find out the degree of improvement in image quality.

• Phonetics and Speech Sciences
• /
• v.15 no.2
• /
• pp.43-51
• /
• 2023

In this paper, we propose an approach for dialect classification based on the speed and pause of speech utterances as well as the age and gender of the speakers. Dialect classification is one of the important techniques for speech analysis. For example, an accurate dialect classification model can potentially improve the performance of speaker or speech recognition. According to previous studies, research based on deep learning using Mel-Frequency Cepstral Coefficients (MFCC) features has been the dominant approach. We focus on the acoustic differences between regions and conduct dialect classification based on the extracted features derived from the differences. In this paper, we propose an approach of extracting underexplored additional features, namely the speed and the pauses of speech utterances along with the metadata including the age and the gender of the speakers. Experimental results show that our proposed approach results in higher accuracy, especially with the speech rate feature, compared to the method only using the MFCC features. The accuracy improved from 91.02% to 97.02% compared to the previous method that only used MFCC features, by incorporating all the proposed features in this paper.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2022.10a
• /
• pp.264-266
• /
• 2022

In order to apply the RNN model to the radio fingerprint-based indoor path generation technology, the data set must be continuous and sequential. However, Wi-Fi radio fingerprint data is not suitable as RNN data because continuity is not guaranteed as characteristic information about a specific location at the time of collection. Therefore, continuity information of sequential positions should be given. For this purpose, clustering is possible through classification of each region based on signal data. At this time, the continuity information between the clusters does not contain information on whether actual movement is possible due to the limitation of radio signals. Therefore, correlation information on whether movement between adjacent clusters is possible is required. In this paper, a deep learning network, a recurrent neural network (RNN) model, is used to predict the path of a moving object, and it reduces errors that may occur when predicting the path of an object by generating continuous location information for path generation in an indoor environment. We propose a method of giving correlation between clustering for generating an improved moving path that can avoid erroneous path prediction that cannot move on the predicted path.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.06a
• /
• pp.205-208
• /
• 2022

Multi-exposure high dynamic range (HDR) image reconstruction, the task of reconstructing an HDR image from multiple low dynamic range (LDR) images in a dynamic scene, often produces ghosting artifacts caused by camera motion and moving objects and also cannot deal with washed-out regions due to over or under-exposures. While there has been many deep-learning-based methods with motion estimation to alleviate these problems, they still have limitations for severely moving scenes. They also require large parameter counts, especially in the case of state-of-the-art methods that employ attention modules. To address these issues, we propose a frequency domain approach based on the idea that the transform domain coefficients inherently involve the global information from whole image pixels to cope with large motions. Specifically we adopt Residual Fast Fourier Transform (RFFT) blocks, which allows for global interactions of pixels. Moreover, we also employ Depthwise Overparametrized convolution (DO-conv) blocks, a convolution in which each input channel is convolved with its own 2D kernel, for faster convergence and performance gains. We call this LFFNet (Lightweight Frequency Fusion Network), and experiments on the benchmarks show reduced ghosting artifacts and improved performance up to 0.6dB tonemapped PSNR compared to recent state-of-the-art methods. Our architecture also requires fewer parameters and converges faster in training.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.22 no.5
• /
• pp.137-147
• /
• 2023

LiDAR is used in autonomous driving and various industrial fields to measure the size and distance of an object. In addition, the sensor also provides intensity images based on the amount of reflected light. This has a positive effect on sensor data processing by providing information on the shape of the object. LiDAR guarantees higher performance as the resolution increases but at an increased cost. These conditions also apply to LiDAR intensity images. Expensive equipment is essential to acquire high-resolution LiDAR intensity images. This study developed artificial intelligence to improve low-resolution LiDAR intensity images into high-resolution ones. Therefore, this study performed parameter analysis for the optimal super-resolution neural network model. The super-resolution algorithm was trained and verified using 2,500 LiDAR intensity images. As a result, the resolution of the intensity images were improved. These results can be applied to the autonomous driving field and help improve driving environment recognition and obstacle detection performance