• Journal of Korean Society for Atmospheric Environment
• /
• v.31 no.2
• /
• pp.157-163
• /
• 2015

To establish National Reference Method (NRM) for $PM_{2.5}$, operational performance of 5 different commercial gravimetric-based $PM_{2.5}$ measuring instruments was assessed at Bulkwang monitoring station from January 23, 2014 to February 28, 2014. First, physical properties, design, and functional performance of the instruments were assessed. Evaluation was carried out to determine whether operating method for the instruments and levels of QA/QC activities meet the data quality objectives (DQOs). To verify whether DQOs were satisfied, reproducibility of QA/QC procedures, accuracy, relative sensitivity, limit of detection, margin of error, and coefficient of determination of the instruments were also evaluated. Results of flow rate measurement of 15 candidate instruments indicated that all the instruments met performance criteria with accuracy deviation of 4.0% and reproducibility of 0.6%. Comparison of final $PM_{2.5}$ mass concentrations showed that the coefficient of determination ($R^2$) values were greater than or equal to 0.9995, and concentration gradient ranged from 0.97 to 1.03. All the instruments satisfied criteria for NRM with the estimated precision of 1.47~2.60%, accuracy of -1.90~3.00%, and absolute accuracy of 1.02~3.12%. This study found that one particular type of measuring instrument was proved to be excellent, with overall evaluation criteria satisfied.

• 전자공학회논문지 IE
• /
• v.47 no.4
• /
• pp.39-47
• /
• 2010

This paper proposes a numerical analysis to prove that the performance of the differential phase detections (DPDs) with the decision feedback, such as the decision feedback DPD (DF-DPD) and the DPD with recursively generated phase reference (DPD-RGPR), approach the performance of the coherent detection with differential decoding. The conventional differential phase detection for M-ary DPSK can make the receiver architecture simple, while it can make the bit-error rate (BER) performance poor because of the previous noisy phase as a reference phase. To improve the BER performance of the conventional differential detection, multiple symbol differential detection methods, including DF-DPD and DPD-RGPR, have been proposed. However, the studies on the analysis and on the comparison of these methods have been little performed. Then, this paper mathematically intends to analyze and compare the performance of the DPDs with the decision feedback. The analysis results show that the DPDs with the decision feedback can have the performance equal to that of the coherent detection with differential decoding and be available for the noncoherent detection in the improved performance. Considering the hardware complexity, the DPD RGPR with the simple detection process by using the recursively generated phase reference can be more simply implemented than the DF-DPD based on the architecture whose complexity increases according to the increasing detection length.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.40 no.6
• /
• pp.20-30
• /
• 2003

In a conventional method based on quadrature 2D Gator wavelets to extract iris features, the iris recognition is performed by a 256-byte iris code, which is computed by applying the Gabor wavelets to a given area of the iris. However, there is a code redundancy because the iris code is generated by basis functions without considering the characteristics of the iris texture. Therefore, the size of the iris code is increased unnecessarily. In this paper, we propose a new feature extraction algorithm based on the ICA (Independent Component Analysis) for a compact iris code. We implemented the ICA to generate optimal basis functions which could represent iris signals efficiently. In practice the coefficients of the ICA expansions are used as feature vectors. Then iris feature vectors are encoded into the iris code for storing and comparing an individual's iris patterns. Additionally, we introduce two methods to enhance the recognition performance of the ICA. The first is to reorganize the ICA bases and the second is to use a different ICA bases set. Experimental results show that our proposed method has a similar EER (Equal Error Rate) as a conventional method based on the Gator wavelets, and the iris code size of our proposed methods is four times smaller than that of the Gabor wavelets.

• Journal of the Korea Concrete Institute
• /
• v.21 no.5
• /
• pp.629-638
• /
• 2009

In this paper, the existing formulas of the step-by-step method were generalized for effective estimation of responses of complicated composite sections due to long-term deformation of concrete. The initial transformed section properties of the composite section were derived from material and section properties of concrete section and sections which confine the longterm deformation of concrete. The transformed section properties at each step were derived from the effective modulus of elasticity considered the creep coefficient variation. Improved formulas of the step-by-step method for generalized responses were derived by introducing 5 generalized parameters. The formulas can be more simplified by applying constant increment of creep coefficient at each step. The constant increment of creep coefficient at each step can also reduce computing time and make equal computing error of each step. The generalized responses for axial elastic strain of concrete section were most sensitive to the area rate of concrete section, and the ratio of the second moment of the confining section area was more sensitive than that of the concrete section. Those for elastic curvature of concrete section were most sensitive to the ratio of the second moment of concrete section area.

• The Journal of the Acoustical Society of Korea
• /
• v.39 no.5
• /
• pp.497-504
• /
• 2020

Constructing speaker embeddings in speaker verification is an important issue. In general, a self-attention mechanism has been applied for speaker embedding encoding. Previous studies focused on training the self-attention in a high-level layer, such as the last pooling layer. In this case, the effect of low-level layers is not well represented in the speaker embedding encoding. In this study, we propose Masked Cross Self-Attentive Encoding (MCSAE) using ResNet. It focuses on training the features of both high-level and low-level layers. Based on multi-layer aggregation, the output features of each residual layer are used for the MCSAE. In the MCSAE, the interdependence of each input features is trained by cross self-attention module. A random masking regularization module is also applied to prevent overfitting problem. The MCSAE enhances the weight of frames representing the speaker information. Then, the output features are concatenated and encoded in the speaker embedding. Therefore, a more informative speaker embedding is encoded by using the MCSAE. The experimental results showed an equal error rate of 2.63 % using the VoxCeleb1 evaluation dataset. It improved performance compared with the previous self-attentive encoding and state-of-the-art methods.

• The Journal of the Acoustical Society of Korea
• /
• v.39 no.5
• /
• pp.441-446
• /
• 2020

In this paper, we propose the α-Feature Map Scaling (α-FMS) method which extends the FMS method that was designed to enhance the discriminative power of feature maps of deep neural networks in Speaker Verification (SV) systems. The FMS derives a scale vector from a feature map and then adds or multiplies them to the features, or sequentially apply both operations. However, the FMS method not only uses an identical scale vector for both addition and multiplication, but also has a limitation that it can only add a value between zero and one in case of addition. In this study, to overcome these limitations, we propose α-FMS to add a trainable parameter α to the feature map element-wise, and then multiply a scale vector. We compare the performance of the two methods: the one where α is a scalar, and the other where it is a vector. Both α-FMS methods are applied after each residual block of the deep neural network. The proposed system using the α-FMS methods are trained using the RawNet2 and tested using the VoxCeleb1 evaluation set. The result demonstrates an equal error rate of 2.47 % and 2.31 % for the two α-FMS methods respectively.

• Journal of Korean Society for Geospatial Information Science
• /
• v.24 no.1
• /
• pp.89-98
• /
• 2016

In this paper, we proposed an automated areal feature matching method based on geometric similarity without user intervention and is applied into areal features of many-to-many relation, for confusion of spatial data-sets of different scale and updating cycle. Firstly, areal feature(node) that a value of inclusion function is more than 0.4 was connected as an edge in adjacency matrix and candidate corresponding areal features included many-to-many relation was identified by multiplication of adjacency matrix. For geometrical matching, these multiple candidates corresponding areal features were transformed into an aggregated polygon as a convex hull generated by a curve-fitting algorithm. Secondly, we defined matching criteria to measure geometrical quality, and these criteria were changed into normalized values, similarity, by similarity function. Next, shape similarity is defined as a weighted linear combination of these similarities and weights which are calculated by Criteria Importance Through Intercriteria Correlation(CRITIC) method. Finally, in training data, we identified Equal Error Rate(EER) which is trade-off value in a plot of precision versus recall for all threshold values(PR curve) as a threshold and decided if these candidate pairs are corresponding pairs or not. To the result of applying the proposed method in a digital topographic map and a base map of address system(KAIS), we confirmed that some many-to-many areal features were mis-detected in visual evaluation and precision, recall and F-Measure was highly 0.951, 0.906, 0.928, respectively in statistical evaluation. These means that accuracy of the automated matching between different spatial data-sets by the proposed method is highly. However, we should do a research on an inclusion function and a detail matching criterion to exactly quantify many-to-many areal features in future.

• Journal of KIISE
• /
• v.42 no.5
• /
• pp.642-651
• /
• 2015

A vanishing point is a point where parallel lines converge, and they become evident when a camera's lenses are used to project 3D space onto a 2D image plane. Vanishing point detection is the use of the information contained within an image to detect the vanishing point, and can be utilized to infer the relative distance between certain points in the image or for understanding the geometry of a 3D scene. Since parallel lines generally exist for the artificial structures within images, line-detection-based vanishing point-detection techniques aim to find the point where the parallel lines of artificial structures converge. To detect parallel lines in an image, we detect edge pixels through edge detection and then find the lines by using the Hough transform. However, the various textures and noise in an image can hamper the line-detection process so that not all of the lines converging toward the vanishing point are obvious. To overcome this difficulty, it is necessary to assign a different weight to each line according to the degree of possibility that the line passes through the vanishing point. While previous research studies assigned equal weight or adopted a simple weighting calculation, in this paper, we are proposing a new method of assigning weights to lines after noticing that the lines that pass through vanishing points typically belong to artificial structures. Experimental results show that our proposed method reduces the vanishing point-estimation error rate by 65% when compared to existing methods.

• The Journal of the Acoustical Society of Korea
• /
• v.38 no.5
• /
• pp.593-600
• /
• 2019

The similarity in tones between speakers can lower the performance of speaker verification. To improve the performance of speaker verification systems, we propose a multi-task learning technique using deep neural network to learn speaker information and age information. Multi-task learning can improve generalization performances, because it helps deep neural networks to prevent hidden layers from overfitting into one task. However, we found in experiments that learning of age information does not work well in the process of learning the deep neural network. In order to improve the learning, we propose a method to dynamically change the objective function weights of speaker identification and age estimation in the learning process. Results show the equal error rate based on RSR2015 evaluation data set, 6.91 % for the speaker verification system without using age information, 6.77 % using age information only, and 4.73 % using age information when weight change technique was applied.

• The Journal of the Acoustical Society of Korea
• /
• v.37 no.6
• /
• pp.483-488
• /
• 2018

Far field input utterance is one of the major causes of performance degradation of speaker verification systems. In this study, we used teacher student learning framework to compensate for the performance degradation caused by far field utterances. Teacher student learning refers to training the student deep neural network in possible performance degradation condition using the teacher deep neural network trained without such condition. In this study, we use the teacher network trained with near distance utterances to train the student network with far distance utterances. However, through experiments, it was found that performance of near distance utterances were deteriorated. To avoid such phenomenon, we proposed techniques that use trained teacher network as initialization of student network and training the student network using both near and far field utterances. Experiments were conducted using deep neural networks that input raw waveforms of 4-channel utterances recorded in both near and far distance. Results show the equal error rate of near and far-field utterances respectively, 2.55 % / 2.8 % without teacher student learning, 9.75 % / 1.8 % for conventional teacher student learning, and 2.5 % / 2.7 % with proposed techniques.