• The Journal of the Acoustical Society of Korea
• /
• v.39 no.5
• /
• pp.414-423
• /
• 2020

In this paper, we propose a Dilated Convolution Gate Linear Unit (DCGLU) to mitigate the lack of sparsity and small receptive field problems caused by the segmentation map extraction process in sound event detection with weak labels. In the advent of deep learning framework, segmentation map extraction approaches have shown improved performance in noisy environments. However, these methods are forced to maintain the size of the feature map to extract the segmentation map as the model would be constructed without a pooling operation. As a result, the performance of these methods is deteriorated with a lack of sparsity and a small receptive field. To mitigate these problems, we utilize GLU to control the flow of information and Dilated Convolutional Neural Networks (DCNNs) to increase the receptive field without additional learning parameters. For the performance evaluation, we employ a URBAN-SED and self-organized bird sound dataset. The relevant experiments show that our proposed DCGLU model outperforms over other baselines. In particular, our method is shown to exhibit robustness against nature sound noises with three Signal to Noise Ratio (SNR) levels (20 dB, 10 dB and 0 dB).

• Fire Science and Engineering
• /
• v.25 no.6
• /
• pp.168-177
• /
• 2011

The combustion test for real box of AC outdoor unit has been performed in this study in order to estimate the fire hazard in multi-system type of AC outdoor unit which is currently used for commercial use. The result showed that in test, there was explosion inside of AC outdoor unit, and flame erupted and fire spread through upper side grill. And then this fire burnt the combustibles such as wires, electronic control board, heat exchange copper plate and plastics etc inside the unit, refrigerant gas pipe was burst due to fire, and accelerated the explosion and flame eruption to outside while the refrigerant was erupting. It is found in this test that the maximum heat release rate of AC outdoor unit is 5,830 kW, the maximum internal temperature measured with infrared camera and thermocouple is $1,201^{\circ}C$, maximum ambient temperature is $881^{\circ}C$, and flame rose higher than about 5 m. It is concluded that the fire in AC outdoor unit cause fire to combustibles around the unit, and may give big damage by generating the secondary fire. It is expected that the result obtained from the test on the real object may be applied to fire realization of AC outdoor unit and estimation of fire spreading to the combustibles around in the future computer simulation.

• Journal of the Korea Society of Computer and Information
• /
• v.20 no.5
• /
• pp.53-63
• /
• 2015

Singing voice synthesis is the generation of a song using a computer given its lyrics and musical notes. Hidden Markov models (HMM) have been proved to be the models of choice for text to speech synthesis. HMMs have also been used for singing voice synthesis research, however, a huge database is needed for the training of HMMs for singing voice synthesis. And commercially available singing voice synthesis systems which use the piano roll music notation, needs to adopt the easy to read standard music notation which make it suitable for singing learning applications. To overcome this problem, we use a speech database for training context dependent HMMs, to be used for singing voice synthesis. Pitch and duration control methods have been devised to modify the parameters of the HMMs trained on speech, to be used as the synthesis units for the singing voice. This work describes a singing voice synthesis system which uses a MusicXML based music score editor as the front-end interface for entry of the notes and lyrics to be synthesized and a hidden Markov model based text to speech synthesis system as the back-end synthesizer. A perceptual test shows the feasibility of our proposed system.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.48 no.5
• /
• pp.88-98
• /
• 2011

Recently, according as the importance of GIS(geography information system) database security is embossed, much researches had been achieved about GIS network security. But most such researches are weak against sourceful illegal reproductions and distributions of GIS vector data map. In this paper, we proposed an efficient layer unit contents based partial encryption technique in the vector map compression domain to prevent illegal distributions and unauthorized accesses. This method achieves a partial encryption about each central coordinate and directional parameters of a MCA(minimum coding attribute) that is created at the vector map compression processing in the vector space. First, the position encryption is applied as permutating randomly the center coordinate of each record that is minimum unit of vector map shape. And second, the direction encryption that changing shapes of vector map topography is applied as encrypting the direction of vertices's coordinates of each record. In experimental results, we confirmed that our proposed method can encipher the large volumed vector map data effectively in low computational complexity. Also, we could minimize the decline of compression efficiency that occurred by conventional contents based encryption schemes using AES or DES algorithms.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.19 no.2
• /
• pp.89-103
• /
• 2020

As we face an aging society, the demand for personal mobility for disabled and aged people is increasing. In fact, as of 2017, the number of electric wheelchair in the country continues to increase to 90,000. However, people with disabilities and seniors are more likely to have accidents while driving, because their judgment and coordination are inferior to normal people. One of the causes of the accident is the interference of personal vehicle steering control due to unbalanced road surface conditions. In this paper, we introduce a encoder type semantic segmentation algorithm that can recognize road conditions at high speed to prevent such accidents. To this end, more than 1,500 training data and 150 test data including road surface damage were newly secured. With the data, we proposed a deep neural network composed of encoder stages, unlike the Auto-encoding type consisting of encoder and decoder stages. Compared to the conventional method, this deep neural network has a 4.45% increase in mean accuracy, a 59.2% decrease in parameters, and an 11.9% increase in computation speed. It is expected that safe personal transportation will be come soon by utilizing such high speed algorithm.

• Journal of the Korean Vacuum Society
• /
• v.19 no.2
• /
• pp.81-90
• /
• 2010

We introduce a novel and versatile approach for preparing self-assembled nanoporous multilayered films with antireflective properties. Protonated polystyrene-block-poly (4-vinylpyrine) (PS-b-P4VP) and anionic polystyrene-block-poly (acrylic acid) (PS-b-PAA) block copolymer micelles (BCM) were used as building blocks for the layer-by-layer assembly of BCM multilayer films. BCM film growth is governed by electrostatic and hydrogen-bonding interactions between the oppositely BCMs. Both film porosity and film thickness are dependent upon the charge density of the micelles, with the porosity of the film controlled by the solution pH and the molecular weight (Mw) of the constituents. PS7K-b-P4VP28K/PS2K-b-PAA8K films prepared at pH 4 (for PS7K-b-P4VP28K) and pH 6 (for PS2K-b-PAA8K) are highly nanoporous and antireflective. In contrast, PS7K-b-P4VP28K/PS2K-b-PAA8K films assembled at pH 4/4 show a relatively dense surface morphology due to the decreased charge density of PS2K-b-PAA8K. Films formed from BCMs with increased PS block and decreased hydrophilic block (P4VP or PAA) size (e.g., PS36K-b-P4VP12K/PS16K-b-PAA4K at pH 4/4) were also nanoporous. Furthermore, we demonstrate that the nanostructured electrochemical sensors based on patterning methods show the electrochemical activities. Anionic poly(styrene sulfonate) (PSS) layers were selectively and uniformly deposited onto the catalase (CAT)-coated surface using the micro-contact printing method. The pH-induced charge reversal of catalase can provide the selective deposition of consecutive PE multilayers onto patterned PSS layers by causing the electrostatic repulsion between next PE layer and catalase. Based on this patterning method, the hybrid patterned multilayers composed of platinum nanoparticles (PtNP) and catalase were prepared and then their electrochemical properties were investigated from sensing $H_2O_2$ and NO gas. This study was based on the papers reported by our group. (J. Am. Chem. Soc. 128, 9935 (2006); Adv. Mater. 19, 4364 (2007); Electro. Mater. Lett. 3, 163 (2007)).

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.18 no.11
• /
• pp.2709-2714
• /
• 2014

This paper has analyzed the change of threshold voltage and conduction path for the ratio of top and bottom gate oxide thickness of asymmetric double gate MOSFET. The asymmetric double gate MOSFET has the advantage that the factor to be able to control the current in the subthreshold region increases. The analytical potential distribution is derived from Poisson's equation to analyze the threshold voltage and conduction path for the ratio of top and bottom gate oxide thickness. The Gaussian distribution function is used as charge distribution. This analytical potential distribution is used to derive off-current and subthreshold swing. By observing the results of threshold voltage and conduction path with parameters of bottom gate voltage, channel length and thickness, projected range and standard projected deviation, the threshold voltage greatly changed for the ratio of top and bottom gate oxide thickness. The threshold voltage changed for the ratio of channel length and thickness, not the absolute values of those, and it increased when conduction path moved toward top gate. The threshold voltage and conduction path changed more greatly for projected range than standard projected deviation.

• Journal of Satellite, Information and Communications
• /
• v.9 no.2
• /
• pp.96-100
• /
• 2014

Recently, the management system of wastewater treatment facility has magnified due to the stringent regulations for the protection of the environment, and a sewage treatment plant efficiency and research of the car development are activated in large facilities or industrial park. however, the existing sewerage disposal system and specific water quality monitoring network reliability for real-time transmission of this building is insufficient. In this paper, we proposed a local wireless network design for sludge multi meter data collection and control for measuring the concentration of the sludge efficiently. Also, the collected data over the local wireless network to transmitted to the central monitoring system and accumulate the data in real time to calculate statistics is possible to monitor the status of the sewage treatment facilities. The proposed system uses a short-range wireless networks of IEEE 802.15.4 and configures an IEEE 802.11 network which can monitor real-time status in central system. Also, we install a sludge multimeter and communication network in sewage treatment facilities and confirm the usefulness of the proposed technique by demonstrating its effectiveness.

• Korean Chemical Engineering Research
• /
• v.57 no.2
• /
• pp.274-282
• /
• 2019

In recent years, big data analysis has been expanded to include automatic control through reinforcement learning as well as prediction through modeling. Research on the utilization of image data is actively carried out in various industrial fields such as chemical, manufacturing, agriculture, and bio-industry. In this paper, we applied NASNet, which is an AutoML reinforced learning algorithm, to DeepU-Net neural network that modified U-Net to improve image semantic segmentation performance. We used BRATS2015 MRI data for performance verification. Simulation results show that DeepU-Net has more performance than the U-Net neural network. In order to improve the image segmentation performance, remove dropouts that are typically applied to neural networks, when the number of kernels and filters obtained through reinforcement learning in DeepU-Net was selected as a hyperparameter of neural network. The results show that the training accuracy is 0.5% and the verification accuracy is 0.3% better than DeepU-Net. The results of this study can be applied to various fields such as MRI brain imaging diagnosis, thermal imaging camera abnormality diagnosis, Nondestructive inspection diagnosis, chemical leakage monitoring, and monitoring forest fire through CCTV.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.42 no.5
• /
• pp.689-699
• /
• 2022

Recently, high productivity capabilities have been improved due to the application of advanced technologies in various industries, but in the construction industry, productivity improvements have been relatively low. Research on advanced technology for the construction industry is being conducted quickly to overcome the current low productivity. Among advanced technologies, 3D scan technology is widely used for creating 3D digital terrain models at construction sites. In particular, the 3D digital terrain model provides basic data for construction automation processes, such as earthwork machine guidance and control. The quality of the 3D digital terrain model has a lot of influence not only on the performance and acquisition environment of the 3D scanner, but also on the denoising, registration and merging process, which is a preprocessing process for creating a 3D digital terrain model after acquiring terrain scan data. Therefore, it is necessary to improve the terrain scan data processing performance. This study seeks to solve the problem of density inhomogeneity in terrain scan data that arises during the pre-processing step. The study suggests a 'pixel-based point cloud comparison algorithm' and verifies the performance of the algorithm using terrain scan data obtained at an actual earthwork site.