• Phonetics and Speech Sciences
• /
• v.13 no.4
• /
• pp.47-53
• /
• 2021

Autoregressive Text-to-Speech (TTS) models suffer from inference instability and slow inference speed. Inference instability occurs when a poorly predicted sample at time step t affects all the subsequent predictions. Slow inference speed arises from a model structure that forces the predicted samples from time steps 1 to t-1 to predict the sample at time step t. In this study, an end-to-end non-autoregressive fast text-to-speech model is suggested as a solution to these problems. The results of this study show that this model's Mean Opinion Score (MOS) is close to that of Tacotron 2 - WaveNet, while this model's inference speed and stability are higher than those of Tacotron 2 - WaveNet. Further, this study aims to offer insight into the improvement of non-autoregressive models.

• Journal of Advanced Navigation Technology
• /
• v.26 no.2
• /
• pp.113-118
• /
• 2022

Motivated by people highlighting important phrases while reading or taking notes we propose a neural network training method by highlighting defective pixel areas to classify effectively defect types of images with complex background textures. To verify our proposed method we apply it to the problem of classifying the defect types of tire band fabric images that are too difficult to classify. In addition we propose a backlight highlighting technique which is tailored to the tire band fabric images. Backlight highlighting images can be generated by using both the GradCAM and simple image processing. In our experiment we demonstrated that the proposed highlighting method outperforms the traditional method in the view points of both classification accuracy and training speed. It achieved up to 13.4% accuracy improvement compared to the conventional method. We also showed that the backlight highlighting technique tailored for highlighting tire band fabric images is superior to a contour highlighting technique in terms of accuracy.

• Journal of The Korean Society of Agricultural Engineers
• /
• v.64 no.3
• /
• pp.63-73
• /
• 2022

A more accurate understanding of the irrigation water supply is necessary for efficient agricultural water management. Although we measure water levels in an irrigation canal using ultrasonic water level gauges, some errors occur due to malfunctions or the surrounding environment. This study aims to apply CNN (Convolutional Neural Network) Deep-learning-based image classification and segmentation models to the irrigation canal's CCTV (Closed-Circuit Television) images. The CCTV images were acquired from the irrigation canal of the agricultural reservoir in Cheorwon-gun, Gangwon-do. We used the ResNet-50 model for the image classification model and the U-Net model for the image segmentation model. Using the Natural Breaks algorithm, we divided water level data into 2, 4, and 8 groups for image classification models. The classification models of 2, 4, and 8 groups showed the accuracy of 1.000, 0.987, and 0.634, respectively. The image segmentation model showed a Dice score of 0.998 and predicted water levels showed R² of 0.97 and MAE (Mean Absolute Error) of 0.02 m. The image classification models can be applied to the automatic gate-controller at four divisions of water levels. Also, the image segmentation model results can be applied to the alternative measurement for ultrasonic water gauges. We expect that the results of this study can provide a more scientific and efficient approach for agricultural water management.

• Advances in nano research
• /
• v.12 no.6
• /
• pp.549-566
• /
• 2022

Mining of ore minerals (sfalerite, cinnabar, and chalcopyrite) from the old mine has led in significant environmental effects as contamination of soils and plants and acidification of water. Also, nanoparticles (NP) have obtained global importance because of their widespread usage in daily life, unique properties, and rapid development in the field of nanotechnology. Regarding their usage in various fields, it is suggested that soil is the final environmental sink for NPs. Nanoparticles with excessive reactivity and deliverability may be carried out as amendments to enhance soil quality, mitigate soil contaminations, make certain secure land-software of the traditional change substances and enhance soil erosion control. Meanwhile, there's no record on the usage of Nano superior substances for mine soil reclamation. In this study, five soil specimens have been tested at 4 sites inside the region of mine (<100 m) to study zeolites, and iron sulfide nanoparticles. Also, through using Artificial Neural Network (ANN) and Extreme Learning Machine (ELM), this study has tried to appropriately estimate the mechanical properties of soil under the effect of these Nano particles. Considering the RMSE and R2 values, Zeolite Nano materials could enhance the mine soil fine through increasing the clay-silt fractions, increasing the water holding capacity, removing toxins and improving nutrient levels. Also, adding iron sulfide minerals to the soils would possibly exacerbate the soil acidity problems at a mining site.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.32 no.5
• /
• pp.817-826
• /
• 2022

AI technology is being successfully introduced in many fields, and models deployed as a service are deployed with black box environment that does not expose the model's information to protect intellectual property rights and data. In a black box environment, attackers try to steal data or parameters used during training by using model output. This paper proposes a method of inferring the type of model to directly find out the composition of layer of the target model, based on the fact that there is no attack to infer the information about the type of model from the deep learning model. With ResNet, VGGNet, AlexNet, and simple convolutional neural network models trained with MNIST datasets, we show that the types of models can be inferred using the output values in the gray box and black box environments of the each model. In addition, we inferred the type of model with approximately 83% accuracy in the black box environment if we train the big and small relationship feature that proposed in this paper together, the results show that the model type can be infrerred even in situations where only partial information is given to attackers, not raw probability vectors.

• Journal of the Korean Geosynthetics Society
• /
• v.21 no.2
• /
• pp.11-19
• /
• 2022

In this paper, we propose a GoogleNet transfer learning and CNN-LSTM combination method to improve the time-series prediction performance for crack detection using crack data captured inside the sewer pipes. LSTM can solve the long-term dependency problem of CNN, so spatial and temporal characteristics can be considered at the same time. The predictive performance of the proposed method is excellent in all test variables as a result of comparing the RMSE(Root Mean Square Error) for time series sections using the crack data inside the sewer pipe. In addition, as a result of examining the prediction performance at the time of data generation, the proposed method was verified that it is effective in predicting crack detection by comparing with the existing CNN-only model. If the proposed method and experimental results obtained through this study are utilized, it can be applied in various fields such as the environment and humanities where time series data occurs frequently as well as crack data of concrete structures.

• Journal of Broadcast Engineering
• /
• v.28 no.1
• /
• pp.42-54
• /
• 2023

It is an important issue to compress huge holographic data in a digital format. In particular, research on the compression of phase-only holograms for commercialization is noteworthy. Conventional video coding standards optimized for natural images are not suitable for compressing phase signals, and neural network-based compression model that can be optimized for phase signals can achieve high performance, but has a memory issue in learning high-resolution holographic data. In this paper, we show that by applying a block-based learned image compression model that can solve memory problems to phase-only holograms, the proposed method can demonstrate significant performance improvement over standard codecs even under the same conditions as block-based. Block-based learned compression model can provide compatibility with conventional standard codecs, solve memory problems, and can perform significantly better against phase-only hologram compression.

• Korean Journal of Agricultural Science
• /
• v.49 no.2
• /
• pp.193-202
• /
• 2022

The recurrent neural network (RNN) algorithm has been widely used in water-related research areas, such as water level predictions and water quality predictions, due to its excellent time series learning capabilities. However, studies on water quality predictions using RNN algorithms are limited because of the scarcity of water quality data. Therefore, most previous studies related to water quality predictions were based on monthly predictions. In this study, the quality of the water in a reservoir in Nonsan, Chungcheongnam-do Republic of Korea was predicted using the RNN-LSTM algorithm. The study was conducted after constructing data that could then be, linearly interpolated as daily data. In this study, we attempt to predict the water quality on the 7th, 15th, 30th, 45th and 60th days instead of making daily predictions of water quality factors. For daily predictions, linear interpolated daily water quality data and daily weather data (rainfall, average temperature, and average wind speed) were used. The results of predicting water quality concentrations (chemical oxygen demand [COD], dissolved oxygen [DO], suspended solid [SS], total nitrogen [T-N], total phosphorus [TP]) through the LSTM algorithm indicated that the predictive value was high on the 7th and 15th days. In the 30th day predictions, the COD and DO items showed R² that exceeded 0.6 at all points, whereas the SS, T-N, and T-P items showed differences depending on the factor being assessed. In the 45th day predictions, it was found that the accuracy of all water quality predictions except for the DO item was sharply lowered.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2023.05a
• /
• pp.320-320
• /
• 2023

강우 데이터는 수문기상, 환경, 농업, 자연재해, 그리고 수자원 시스템 분야에서 가장 필수적인 기본 요소 중 하나이다. 또한 강우 데이터는 수문학적 분석에서 활용되는 필수 입력 자료 중 하나로 관측 데이터의 품질에 따라 수문 모형을 이용한 모의 결과물의 정확도가 결정된다고 할 수 있다. 따라서, 강우 관측소별로 강우 데이터의 품질을 어떻게 관리하느냐에 따라 수문 모형의 활용 범위 및 수자원 관리의 효율성이 결정될 수 있다. 강우의 시공간적 변동성은 수 많은 인자들과 직간접적으로 연계되어 있기 때문에 미계측 강우 자료에 대해 직접 관측이 아닌 수치 모형을 이용하여 강우의 발생과 강우량을 산정하는 것은 매우 복잡한 과제 중 하나이다. 현재 국내에서 운용되고 있는 강우 관측소의 경우에도 미계측 된 강우 데이터가 존재함으로써 강우 데이터의 활용에 제한이 생기는 경우가 있다. 따라서, 이러한 미계측 데이터의 추정 및 보완은 보다 효과적인 수재해 방지, 수자원 관리를 위한 필수 과제 중 하나이다. 일반적으로, 미계측 강우를 산정하기 위해서 Kriging, Thiessen, 등우선법, 그리고 역거리 관측법 등 다양한 수문학적 방법들이 적용되고 있다. 이러한 방법들은 산악효과나 강우 관측소의 분포 상태 등을 고려하지 못하기 때문에 측정하는 지역에 따라 강우 추정 오차가 커질 수 있다는 한계가 있다. 최근에는 데이터 관측 시스템과 빅데이터 기술의 발전과 활용 가능한 데이터의 양이 증가함에 따라 머신러닝을 활용한 사례가 증가하고 있다. 머신러닝은 데이터 사이의 관계를 기반으로 분류, 회귀, 그리고 예측 문제에 주로 사용되는 기법 중 하나이다. 따라서, 본 연구에서는 광주광역시 지역에 위치한 주요 강우 관측 지점들을 대상으로 미계측 된 시강우 데이터를 추정 및 복원하고자 한다. 여기서 데이터 추정 기술이란 미계측 강우의 발생 유무 및 강우량을 추정할 수 있는 기술을 의미한다. 이를 위해 대표적인 머신러닝 알고리즘인 인공신경망(Artificial Neural Network) 및 랜덤포레스트(Random Forest)를 적용하였다.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2023.05a
• /
• pp.34-34
• /
• 2023

The Chah-Nimeh reservoirs, which are a sort of natural lakes located in the border of Iran and Afghanistan, are the main drinking and agricultural water resources of Sistan arid region. Considering the occurrence of intense seasonal wind, locally known as levar wind, this study aims to explore the possibility to provide a TSM (Total Suspended Matter) monitoring model of Chah-Nimeh reservoirs using multi-temporal satellite images and in-situ wind speed data. The results show that a strong correlation between TSM concentration and wind speed are present. The developed empirical model indicated high performance in retrieving spatiotemporal distribution of the TSM concentration with R²=0.98 and RMSE=0.92g/m³. Following this observation, we also consider a machine learning-based model to predicts the average TSM using only wind speed. We connect our in-situ wind speed data to the TSM data generated from the inversion of multi-temporal satellite imagery to train a neural network based mode l(Wind2TSM-Net). Examining Wind2TSM-Net model indicates this model can retrieve the TSM accurately utilizing only wind speed (R²=0.88 and RMSE=1.97g/m³). Moreover, this results of this study show tha the TSM concentration can be estimated using only in situ wind speed data independent of the satellite images. Specifically, such model can supply a temporally persistent means of monitoring TSM that is not limited by the temporal resolution of imagery or the cloud cover problem in the optical remote sensing.