• Proceedings of the IEEK Conference
• /
• 1998.10a
• /
• pp.1273-1276
• /
• 1998

The HMM-Net is an architecture for a neural network that implements a hidden Markov model (HMM). The architecture is developed for the purpose of combining the discriminant power of neural networks with the time-domain modeling capability of HMMs. Criteria used for learning HMM-Net classifiers are maximum likelihood (ML), maximum mutual information (MMI), and minimization of mean squared error(MMSE). In this paper we propose an efficient learning method of HMM-Net classifiers using hybrid criteria, ML/MMSE and MMI/MMSE, and report the results of an experimental study comparing the performance of HMM-Net classifiers trained by the gradient descent algorithm with the above criteria. Experimental results for the isolated numeric digits from /0/ to /9/ show that the performance of the proposed method is better than the others in the respects of learning and recognition rates.

• Proceedings of the IEEK Conference
• /
• 1998.06a
• /
• pp.933-935
• /
• 1998

The HMM-Net is an architecture for a neural network that implements a hidden markov model (HMM). The architecture is developed for the purpose of combining the discriminant power of neural networks with the time-domain modeling capability of HMMs. Criteria used for learning HMM-Net classifiers are maximum likelihood(ML) and minimization of mean squared error(MMSE). In this paper we propose an efficient learning method of HMM_Net classifiers using a ML-MMSE hybrid criterion and report the results of an experimental study comparing the performance of HMM_Net classifiers trained by the gradient descent algorithm with the above criteria. Experimental results for the isolated numeric digits from /0/ to /9/ show that the performance of the proposed method is better than the others in the repects of learning and recognition rates.

• Journal of the Korean Society of Fisheries and Ocean Technology
• /
• v.51 no.1
• /
• pp.146-153
• /
• 2015

A model net experiment of the gape net for anchovy in Jindo, Jeollanam-do was carried out to investigate the net shape and hydrodynamic resistance using circulating water channel. The model net was made 1/33 down scale by Tauti's similarity method and the range of experimental current speed was from 0.5 knot to 3.5 knot (increasing 0.5 knot interval). The net mouth height in 0.5 knot of the minimum experiment current speed was shown 26.0 cm (full-scale conversion value 8.58 m). The net mouth height and mouth area in 1.5 knot of the same current speed with a gape net fishing ground were shown 20.0 cm (full-scale conversion value : 6.60 m) and about $507.9cm^2$ (full-scale conversion value : $55.31m^2$). The net mouth height and area were decreased with increase the experimental current speed. The hydrodynamic resistance of the model net in 1.5 knot current speed was shown 1.11 kgf and the value of full-scale conversion by Tauti's method was shown 3.996 ton.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.26 no.7
• /
• pp.1426-1437
• /
• 2002

Using the Lagrangian explicit time-integration finite element code NET3D which can treat three-dimensional high-velocity impact problems, oblique penetration processes of long rod projectile with yaw against thin plate are simulated. Through the comparison of simulation result with experimental result and other code's computational result, the adaptability and accuracy of NET3D is evaluated under the complex situation in which yaw angle and oblique angle exist simultaneously. Main research contents to be handled in this paper include the followings. First, the accuracy and efficiency estimation of NET3D code result obtained from the oblique penetration simulations of long rod projectile with yaw against thin plate. Second, the effect of increasing impact velocity. Third, the effect of initial yaw for the spaced-plate target. Residual velocities, residual lengths, angular velocities, and final deformed configurations obtained from the NET3D computations are compared with the experimental results and other code's computational results such as Eulerian code MESA and Lagrangian code EPIC. As a result of comparisons, it has been found that NET3D code is superior to EPIC code and MESA code in the prediction capability of residual velocity and residual length of penetrator. The key features obtained from the experiment can be successfully reproduced through NET3D simulations. Throughout the study, the applicability and accuracy of NET3D as a metallic armor system design tool is verified.

• Journal of the Korean Society of Fisheries and Ocean Technology
• /
• v.59 no.1
• /
• pp.74-84
• /
• 2023

This paper is the result of a comparative analysis of crew members' workloads using the EMG and OWAS methods according to the fishing process before and after the improvement of the operating system of the experimental vessel, with the aim of improving operational efficiency and safety work of coastal improved stow net fishing vessels. The target crew members were three people with at least five years of experience on board the same fishing vessel as the experimental vessel. After improving the operating system, such as installing a power block crane and two capstans, change the location of the ball-roller, the time required for setting and hauling work decreased (p < .01), and the evaluation results by OWAS showed that the overall workload for setting and hauling net work decreased. The results of muscle activity analysis showed a decrease in lashing anchor work (p < .01) in the case of hauling net and in anchor dropping work in the case of setting net (p < .001). It is judged that the use of ball rollers in net handling has been reduced; consequently, work safety has been improved.

• Journal of Korea Multimedia Society
• /
• v.22 no.6
• /
• pp.665-675
• /
• 2019

In this paper, we compare and analyze the classification performance of deep learning algorithm Convolutional Neural Network(CNN) ac cording to ensemble generation and combining techniques. We used several CNN models(VGG16, VGG19, DenseNet121, DenseNet169, DenseNet201, ResNet18, ResNet34, ResNet50, ResNet101, ResNet152, GoogLeNet) to create 10 ensemble generation combinations and applied 6 combine techniques(average, weighted average, maximum, minimum, median, product) to the optimal combination. Experimental results, DenseNet169-VGG16-GoogLeNet combination in ensemble generation, and the product rule in ensemble combination showed the best performance. Based on this, it was concluded that ensemble in different models of high benchmarking scores is another way to get good results.

• Journal of the Korean Society of Fisheries and Ocean Technology
• /
• v.49 no.3
• /
• pp.218-226
• /
• 2013

The objective of this study is to estimate the optimal net twine thickness of drift net for yellow croaker. The filed experiments of the fishing performance were conducted with three kinds of drift net with different monofilament diameters (No.3=0.284mm, No.4=0.330mm, No.5=0.370mm) the total eight times in the southwest coastal sea of Korea. And the physical properties tests on the monofilaments of experimental net were carried out to estimated breaking load and softness in dry and wet conditions, respectively. From the results, the No.3 mono. was the strongest break load per unit area in dry and wet conditions. And the softness showed that the No. 3 mono. was the most soft than another experimental monofilaments. The fishing performance was; the No. 3 drift net showed the most catches. Conversely, the catches of No. 4 and No. 5 drift net showed the half on the catches of No. 3 drift net. Consequently, the diameter of monofilament in the drift net for yellow croaker should carefully choose to consider the economic sides such as the amount of catches and the cost of catches.

• Journal of the Korean Society of Fisheries and Ocean Technology
• /
• v.44 no.3
• /
• pp.165-173
• /
• 2008

Sea experiment for this study was carried out to develop a Trawl Escapement Net(TEN) using net material to improve the survival rate of undersized fishes that escaped from trawls in the Korean southern sea in May and November, 2005. Three kinds of Trawl Escapement Net(TEN) were examined to increase the escapement rate for fishes escaped from the TEN, and total catches in the codend and cover-net were investigated. The main catch species in the experimental fishing trials using a V shape TEN(A type) in May were Korean pomfret(Pampus echinogaster), slimy(Leiognathus nuchalis), gizzard shad(Konosirus punctatus), white croaker(Pennahia argentata) and smelt(Sillago sihama). In the experimental fishing trials using a V shape TEN with a protector for undersized fishes(B type) and a panel shape TEN with a protector(C type) in November, the main catch species were horse mackerel(Trachurus japonicus), sea pike (Sphyraena japonica), large-head hairtail(Trichiurus lepturus), kammal thryssa(Thryssa kammalensis), white croaker(Pennahia argentata), butterfish(Psenopsis anomala) and red gurnard(Lepidotrigla microptera). Most of large-sized fish species like Korean pomfret and Spanish mackerel could not escape through the TEN in the B type and C type. Catches of horse mackerel were highest in the experimental fishing of November and escapement rates for most fish species including horse mackerel were higher in the C type compared with B type.

• Phonetics and Speech Sciences
• /
• v.10 no.1
• /
• pp.33-38
• /
• 2018

Although deep neural network-based acoustic models have greatly improved the performance of automatic speech recognition (ASR), reverberation still degrades the performance of distant speech recognition in indoor environments. In this paper, we adopt the DenseNet, which has shown great performance results in image classification tasks, to improve the performance of reverberant speech recognition. The DenseNet enables the deep convolutional neural network (CNN) to be effectively trained by concatenating feature maps in each convolutional layer. In addition, we extend the concept of multi-resolution CNN to multi-resolution DenseNet for robust speech recognition in reverberant environments. We evaluate the performance of reverberant speech recognition on the single-channel ASR task in reverberant voice enhancement and recognition benchmark (REVERB) challenge 2014. According to the experimental results, the DenseNet-based acoustic models show better performance than do the conventional CNN-based ones, and the multi-resolution DenseNet provides additional performance improvement.

• Journal of the Korean Institute of Telematics and Electronics A
• /
• v.30A no.5
• /
• pp.52-62
• /
• 1993

This paper describes the implementation of a layout generation system for the gate matrix style to implement multi-level logic. To achieve improved layouts from general net lists, the proposed system performs flexible net binding for series nets. Also the system reassings gates by the heuristic information that shorter net lengths are better for the track minimization. By track minimizing with subdividing layout column information, the system decreases the number of necessary layout tracks. Experimental results show that the system generates more area-reduced (approximately 7.46%) layouts than those by previous gate matrix generation systems using net list inputs.