• Proceedings of the Korea Information Processing Society Conference
• /
• 2015.04a
• /
• pp.776-778
• /
• 2015

논문에서는 DNN(Deep Neural Network)의 dropout 기법을 이용하여 산란계가 고온 스트레스를 받고 있는지 여부를 닭의 울음소리 정보를 통해 탐지하는 방법을 제안한다. 실험에서는 $21^{\circ}C$ 정상 온도에서 100개의 소리 데이터, $35^{\circ}C$ 고온에서 200개의 소리 데이터를 사용한다. 먼저, DNN의 학습을 위해서 취득한 울음소리에서 54개의 소리 특징 정보를 추출한다. 둘째, CFS(Correlation Feature Selection)을 이용하여, 추출된 특징 중 온도 구분을 위한 중요한 특정 10개를 선택한다. 셋째, 선택된 소리특징을 DNN에 적용하여 온도 환경을 구분하는 시스템이다. DNN의 과적합(over-fitting) 영향을 감소시키고, 성능 향상을 위하여 dropout 비율을 조정하여 실험을 진행하였다. 본 연구에서는 실제 계사에서 수집된 소리 정보를 이용하여 모의실험을 수행한 결과 매우 우수한 성능을 보임을 확인하였다.

• Journal of Yeungnam Medical Science
• /
• v.36 no.2
• /
• pp.92-98
• /
• 2019

Forefoot disorders are often seen in clinical practice. Forefoot deformity and pain can deteriorate gait function and decrease quality of life. This review presents common forefoot disorders and conservative treatment using an insole or orthosis. Metatarsalgia is a painful foot condition affecting the metatarsal (MT) region of the foot. A MT pad, MT bar, or forefoot cushion can be used to alleviate MT pain. Hallux valgus is a deformity characterized by medial deviation of the first MT and lateral deviation of the hallux. A toe spreader, valgus splint, and bunion shield are commonly applied to patients with hallux valgus. Hallux limitus and hallux rigidus refer to painful limitations of dorsiflexion of the first metatarsophalangeal joint. A kinetic wedge foot orthosis or rocker sole can help relieve symptoms from hallux limitus or rigidus. Hammer, claw, and mallet toes are sagittal plane deformities of the lesser toes. Toe sleeve or padding can be applied over high-pressure areas in the proximal or distal interphalangeal joints or under the MT heads. An MT off-loading insole can also be used to alleviate symptoms following lesser toe deformities. Morton's neuroma is a benign neuroma of an intermetatarsal plantar nerve that leads to a painful condition affecting the MT area. The MT bar, the plantar pad, or a more cushioned insole would be useful. In addition, patients with any of the above various forefoot disorders should avoid tight-fitting or high-heeled shoes. Applying an insole or orthosis and wearing proper shoes can be beneficial for managing forefoot disorders.

• Geomechanics and Engineering
• /
• v.16 no.5
• /
• pp.539-545
• /
• 2018

Longitudinal Displacement Profile (LDP) is an appropriate tool for determination of the displacement magnitude of the tunnel walls as a function of the distance to the tunnel face. Some useful formulations for calculation of LDP have been developed based on the monitoring data on site or by 3D numerical simulations. However, the presented equations are only based on the tunnel dimensions and for different quality of rock masses proposed a unique LDP. In the present study, it is tried to present a new formulation, for calculation of LDP, on the basis of Rock mass quality. For this purpose, a comprehensive numerical simulation program was developed to investigate the effect of rock mass quality on the LDP. Results of the numerical modelling were analyzed and the least square technique was used for fitting an appropriate curve on the derived data from the numerical simulations. The proposed formulation in the present study, is a logistic function and the constants of the logistic function were predicted by rock mass quality index (GSI). Results of this study revealed that, the LDP curves of the tunnel surrounded by rock masses with high quality (GSI>60) match together; because the rock mass deformation varies over an elastic range.

• Environmental Engineering Research
• /
• v.23 no.4
• /
• pp.364-373
• /
• 2018

The present study assessed the efficient removal of nutrients and Chlorophyll-a (Chl-a) by using methyl esterified sericite (MES) and pine needle extracts (PNE), a low cost and abundant green hybrid material from nature. For this purpose, the optimal conditions were investigated, such as the pH, temperature, MES and PNE ratio, and MES-PNE dose. In addition, a Microcystis aeruginosa control using MES-PNE was also analyzed with various inhibition models. The removal of the nutrient and Chl-a onto MES-PNE was optimized for over 95% removal as follows: 2-2.5 for the MES-PNE ratio, 7-8 pH and a $22-25^{\circ}C$ temperature. In this respect, approximately 1.52-2.20 g/L of MES-PNE was required to remove each 1 g of dry weight/L of Chl-a. Total phosphorus (TP) has a greater influence on the increase in Chl-a than total nitrogen (TN) according to the correlation between TN, TP and Chl-a. Moreover, the Luong model was the best model for fitting the biodegradation kinetics data from Chl-a on MES-PNE from lake water. The novel hybrid material MES-PNE was very effective at removing TN, TP and Chl-a from the lake and can be applied in the field.

• International Journal of Naval Architecture and Ocean Engineering
• /
• v.12 no.1
• /
• pp.657-666
• /
• 2020

A reliable and cost-effective technique for the development of corrosion damage model is introduced to predict nonlinear time-dependent corrosion wastage of steel structures. A detailed explanation on how to propose a generalised mathematical formulation of the corrosion model is investigated in this paper (Part I), and verification and application of the developed method are covered in the following paper (Part II) by adopting corrosion data of a ship's ballast tank structure. In this study, probabilistic approaches including statistical analysis were applied to select the best fit probability density function (PDF) for the measured corrosion data. The sub-parameters of selected PDF, e.g., the largest extreme value distribution consisting of scale, and shape parameters, can be formulated as a function of time using curve fitting method. The proposed technique to formulate the refined time-dependent corrosion wastage model (TDCWM) will be useful for engineers as it provides an easy and accurate prediction of the 1) starting time of corrosion, 2) remaining life of the structure, and 3) nonlinear corrosion damage amount over time. In addition, the obtained outcome can be utilised for the development of simplified engineering software shown in Appendix B.

• Journal of Korea Multimedia Society
• /
• v.24 no.3
• /
• pp.382-395
• /
• 2021

In this study, character recognition using deep learning is performed among the various defects in the PCB, the purpose of which is to check whether the printed characters are printed correctly on top of components, or the incorrect parts are attached. Generally, character recognition may be perceived as not a difficult problem when considering MNIST, but the printed letters on the PCB component data are difficult to collect, and have very high redundancy. So if a deep learning model is trained with original data without any preprocessing, it can lead to over fitting problems. Therefore, this study aims to reduce the redundancy to the smallest dataset that can represent large amounts of data collected in limited production sites, and to create datasets through data enhancement to train a flexible deep learning model can be used in various production sites. Moreover, ResNet model verifies to determine which combination of datasets is the most effective. This study discusses how to reduce and augment data that is constantly occurring in real PCB production lines, and discusses how to select coresets to learn and apply deep learning models in real sites.

• Journal of information and communication convergence engineering
• /
• v.19 no.2
• /
• pp.79-83
• /
• 2021

This paper proposes a novel image classification method based on few-shot learning, which is mainly used to solve model overfitting and non-convergence in image classification tasks of small datasets and improve the accuracy of classification. This method uses model structure optimization to extend the basic convolutional neural network (CNN) model and extracts more image features by adding convolutional layers, thereby improving the classification accuracy. We incorporated certain measures to improve the performance of the model. First, we used general methods such as setting a lower learning rate and shuffling to promote the rapid convergence of the model. Second, we used the data expansion technology to preprocess small datasets to increase the number of training data sets and suppress over-fitting. We applied the model to 10 monkey species and achieved outstanding performances. Experiments indicated that our proposed method achieved an accuracy of 87.92%, which is 26.1% higher than that of the traditional CNN method and 1.1% higher than that of the deep convolutional neural network ResNet50.

• Journal of Korea Multimedia Society
• /
• v.24 no.12
• /
• pp.1624-1631
• /
• 2021

With the continued development of image processing technology, we live in a time when it is difficult to visually discriminate processed (or tampered) images from real images. However, as the risk of fake images being misused for crime increases, the importance of image forensic science for identifying fake images is emerging. Currently, various deep learning-based identifiers have been studied, but there are still many problems to be used in real situations. Due to the inherent characteristics of deep learning that strongly relies on given training data, it is very vulnerable to evaluating data that has never been viewed. Therefore, we try to find a way to improve generalization ability of deep learning-based fake image identifiers. First, images with various contents were added to the training dataset to resolve the over-fitting problem that the identifier can only classify real and fake images with specific contents but fails for those with other contents. Next, color spaces other than RGB were exploited. That is, fake image identification was attempted on color spaces not considered when creating fake images, such as HSV and YCbCr. Finally, dropout, which is commonly used for generalization of neural networks, was used. Through experimental results, it has been confirmed that the color space conversion to HSV is the best solution and its combination with the approach of increasing the training dataset significantly can greatly improve the accuracy and generalization ability of deep learning-based identifiers in identifying fake images that have never been seen before.

• Earthquakes and Structures
• /
• v.25 no.6
• /
• pp.443-454
• /
• 2023

Steel shear walls are used to strengthen steel and concrete structures. One such system is Partial Attached Steel Shear Walls (PASSW), which are only connected to frame beams. This system offers both structural and architectural advantages. This study first calibrated the numerical model of RC frames with and without PASSW using an experimental sample. The seismic performance of the RC frame was evaluated by 30 non-linear static analyses, which considered stiffness, ductility, lateral strength, and energy dissipation, to investigate the effect of PASSW width and column axial load. Based on numerical results and a curve fitting technique, a lateral stiffness equation was developed for frames equipped with PASSW. The effect of the shear wall location on the concrete frame was evaluated through eight analyses. Nonlinear dynamic analysis was performed to investigate the effect of the shear wall on maximum frame displacement using three earthquake records. The results revealed that if PASSW is designed with appropriate stiffness, it can increase the energy dissipation and ductility of the frame by 2 and 1.2 times, respectively. The stiffness and strength of the frame are greatly influenced by PASSW, while axial force has the most significant negative impact on energy dissipation. Furthermore, the location of PASSW does not affect the frame's behavior, and it is possible to have large openings in the frame bay.

• Nuclear Engineering and Technology
• /
• v.56 no.9
• /
• pp.3904-3913
• /
• 2024

Radiation detection at high count rate suffers from pulse pile-up, where the counting data and energy information of the system are affected by the overlapping of the system output pulses. There exist various pile-up correction strategies to recover the true information of the pulses, among which pulse-tail extrapolation is a well-known method focused on in this study. Present work aims to use a mono-exponential model for extrapolating the pileup-distorted trailing edge of a pulse, to provide a reference line for calculating the true amplitude of its subsequent overlapping pulse. To this goal, the auto-regression on linear operations (ARLO) method is examined and compared with two integration-based methods (the Foss and the Matheson methods), as well as the non-linear least squares (NLS) method. Despite a higher sensitivity to noise, the ARLO method was able to provide a simple, non-iterative solution with a performance over 400 times faster than the NLS algorithm, according to the analysis of a high count rate set of experimental pulses from a NaI(Tl) detection system. Foss and Matheson methods also provided solutions reasonably faster than NLS (but not surpassing ARLO), performing exactly the same as each other with results very close to NLS, benefiting from their non-iterative nature.