• 한국정보통신학회논문지
• /
• 제16권9호
• /
• pp.1870-1877
• /
• 2012

천이신호는 지속시간이 짧으면서 길이의 변화가 크고, 시변성 및 비정재성 특성을 갖는다. 이러한 천이신호의 식별에는 분석 프레임 단위로 참조신호에 대한 기준패턴을 만들어 입력신호와의 유사도를 비교하는 방법이 효과적일 수 있다. 본 연구에서는 참조신호의 기준패턴으로 프레임 기반의 특징벡터들에 대해 확률통계 모형인 정규혼합모델을 적용하는 방법을 제안하고, 다양한 수중 천이신호에 대한 식별 실험을 통해 제안한 방법의 타당성을 검증하였다.

• 한국음향학회지
• /
• 제30권7호
• /
• pp.383-389
• /
• 2011

본 논문에서는 음성향상을 위하여 환경잡음분류를 적용한 향상된 음성부재확률 추정방법을 제안한다. 기존의 음성부재확률 추정방법에서는 마이크로폰 입력신호와 추정된 잡음신호 기반의 a posteriori SNR값에 문턱값을 적용하여 음성부재확률을 구하는데 필요한 음성부재의 a priori 확률을 도출하였다. 본 논문에서 제안된 알고리즘은 보다 효과적인 음성부재확률 추정을 위하여 고정된 문턱값과 스무딩 (smoothing)파라미터를 사용하는 기존의 방법과는 달리 잡음분류 알고리즘인 가우시안 혼합 모델 (Gaussian mixture model)을 사용하여 잡음마다 최적화된 파라미터를 적용한다. 제안된 음성 향상 기법은 ITU-T P.862 PESQ (perceptual evaluation of speech quality)와 composite measure를 이용하여 다양한 환경에서 평가하였으며, 제안된 알고리즘이 기존의 음성부재확률 추정방법보다 향상된 결과를 보였다.

• ETRI Journal
• /
• 제43권6호
• /
• pp.991-1003
• /
• 2021

Automatic modulation classification is essential in radar emitter identification. We propose a cascade classifier by combining a support vector machine (SVM) and convolutional neural network (CNN), considering that noise might be taken as radar signals. First, the SVM distinguishes noise signals by the main ridge slice feature of signals. Second, the complex envelope features of the predicted radar signals are extracted and placed into a designed CNN, where a modulation classification task is performed. Simulation results show that the SVM-CNN can effectively distinguish radar signals from noise. The overall probability of successful recognition (PSR) of modulation is 98.52% at 20 dB and 82.27% at -2 dB with low computation costs. Furthermore, we found that the accuracy of intermediate frequency estimation significantly affects the PSR. This study shows the possibility of training a classifier using complex envelope features. What the proposed CNN has learned can be interpreted as an equivalent matched filter consisting of a series of small filters that can provide different responses determined by envelope features.

• 한국소음진동공학회논문집
• /
• 제16권10호
• /
• pp.1005-1013
• /
• 2006

A survey and auditory experiment on multiple residential noises such as floor impact, airborne, bathroom, drainage and traffic noises were conducted to develop a combined rating system and to establish criteria for multiple residential noises. Subjective reactions such as annoyance, activity disturbance, sleep disturbance, and satisfaction to overall noise environment and each residential noise were recorded. The effect of individual noise perception on the evaluation of the overall noise environment was also investigated. The survey results showed that satisfaction for floor impact noise most greatly affects the overall satisfaction for overall noise environment and annoyance most greatly affects the satisfaction for individual noise sources. Auditory experiments were undertaken to determine the percent satisfaction for individual noise levels. Result of auditory experiment showed that the noise level corresponding to 40 % satisfaction is 49 dB $(L_{i,Fmax,AW})$ for floor impact and is about 40 dB(A) for airborne, drainage and traffic noise. From the results of the survey and the auditory experiments, an equation for predicting the overall satisfaction for multiple noises was developed and a classification of multiple residential noises was proposed.

• 한국인터넷방송통신학회논문지
• /
• 제10권5호
• /
• pp.133-138
• /
• 2010

본 논문에서는 이미지에 대한 공간 특성(Spatial properties) 및 통계적 특성(Statistical properties)을 포함한 특징이미지를 구성하고, 지역 분산 크기를 이용한 공분산 행렬을 생성하여 텍스쳐 분류에 이용함으로서 조도(illumination) 및 노이즈(Noise) 그리고 회전(Rotation)에 강인한 텍스쳐 분류 방법을 제안한다. 또한 영역 합계의 빠른 연산을 위해 사용된 중간 이미지 표현인 적분 이미지(Integral Image)를 이용함으로서 텍스쳐 검출 프로세스의 수행 시간을 최소화 하는 방법을 제공한다. 제안한 방법의 성능 평가를 위해 브로다츠(Brodatz) 질감 이미지를 이용하여 잡음 추가 및 히스토그램 명세화 그리고 회전 이미지를 생성하여 실험하였으며, 96% 이상의 성능을 얻을 수 있었다.

• Journal of information and communication convergence engineering
• /
• 제21권3호
• /
• pp.208-215
• /
• 2023

Deep learning techniques provide powerful solutions to several pattern-recognition problems, including Raman spectral classification. However, these networks require large amounts of labeled data to perform well. Labeled data, which are typically obtained in a laboratory, can potentially be alleviated by data augmentation. This study investigated various data augmentation techniques and applied multiple deep learning methods to Raman spectral classification. Raman spectra yield fingerprint-like information about chemical compositions, but are prone to noise when the particles of the material are small. Five augmentation models were investigated to build robust deep learning classifiers: weighted sums of spectral signals, imitated chemical backgrounds, extended multiplicative signal augmentation, and generated Gaussian and Poisson-distributed noise. We compared the performance of nine state-of-the-art convolutional neural networks with all the augmentation techniques. The LeNet5 models with background noise augmentation yielded the highest accuracy when tested on real-world Raman spectral classification at 88.33% accuracy. A class activation map of the model was generated to provide a qualitative observation of the results.

• 한국소음진동공학회논문집
• /
• 제24권9호
• /
• pp.682-686
• /
• 2014

Nowadays, the needs to revise the classification criteria for noise emission facilities have been suggested by the related industries. Because there existed many reasonable factors in the criteria regarding the noise emission facilities. And the noise emission facility classification criterion of the print machine changed from 50 HP to 100 HP in 2013. But the increasement of the noise emission facility classification criterion of the print machine can cause adverse effects like the bigger noise. So, in this paper, we measured the print machine's sound power level according to the changes of the print machine's power to assess the adverse effects. The measurement method applied with KS I ISO 9614-2(1996). The corelation between the sound power level and the power of print machines was analyzed by regression analysis. In this paper, we found that the sound power level of the print machines can increase about 1.3 dB in the condition of that the power of print machine increases from 50 HP to 100 HP. And we found that the sound power level of the print machines can increase about 1.0 dB for a increasement of 1,000 SPH(sheet per hour) of printing speed. The noise emission characteristics of print machine stuied in this paper will be useful to design the noise reduction plan in the future.

• 한국음향학회지
• /
• 제37권4호
• /
• pp.256-261
• /
• 2018

차량 내부에는 BSR(Buzz, Squeak, Rattle) 세 가지 유형의 소음이 발생한다. 본 논문에서는 심층 컨볼루션 신경망으로 추출한 소음 특징에 기반하여 자동으로 차량 내부의 BSR 소음을 분류하는 분류기를 제안한다. 차량 내부의 소음은 전처리 단계에서 STFT(Short-time Fourier Transform) 알고리즘을 사용하여 소음 맵으로 표현된다. 생성된 소음 맵 내부에서 실제 소음의 위치를 정확하게 파악하기 어려운 문제에 대처하기 위해서 슬라이딩 윈도우 방법으로 분할하였다. 본 논문에서는 t-SNE(t-Stochastic Neighbor Embedding) 알고리즘을 사용하여 심층 컨볼루션 신경망 내부 파라미터를 시각화하고 정성적인 방식으로 오분류데이터를 분석하였다. 분류된 데이터의 정량적인 분석을 위해 소음의 종류별 유사도를 SSIM(Structural Similarity Index) 수치에 기반하여 정량화하여 리트랙터의 떨림음이 정상주행음과 가장 유사하다는 것을 밝혔다. 제안하는 방법의 분류기는 기타 기계학습 알고리즘 대비 최고 분류 정확도를 달성하였다(99.15%).

• ETRI Journal
• /
• 제39권1호
• /
• pp.87-96
• /
• 2017

Ultrasonic surgical devices are routinely used for surgical procedures. The incision and coagulation of tissue generate a temperature of $40^{\circ}C-150^{\circ}C$ and depend on the controllable output power level of the surgical device. Recently, research on the classification of grasped tissues to automatically control the power level was published. However, this research did not consider the specific characteristics of the surgical device, tissue denaturalization, and so on. Therefore, this research proposes a robust algorithm that simulates noise to resemble real situations and classifies tissue using conventional classifier algorithms. In this research, the bioimpedance spectrum for six tissues (liver, large intestine, kidney, lung, muscle, and fat) is measured, and five classifier algorithms are used. A signal-to-noise ratio of additive white Gaussian noise diversifies the testing sets, and as a result, each classifier's performance exhibits a difference. The k-nearest neighbors algorithm shows the highest classification rate of 92.09% (p < 0.01) and a standard deviation of 1.92%, which confirms high reproducibility.

• 한국군사과학기술학회지
• /
• 제24권3호
• /
• pp.264-271
• /
• 2021

Raman spectroscopy is an equipment that is widely used for classifying chemicals in chemical defense operations. However, the classification performance of Raman spectrum may deteriorate due to dark current noise, background noise, spectral shift by vibration of equipment, spectral shift by pressure change, etc. In this paper, we compare the classification accuracy of various machine learning algorithms including k-nearest neighbor, decision tree, linear discriminant analysis, linear support vector machine, nonlinear support vector machine, and convolutional neural network under noisy and spectral shifted conditions. Experimental results show that convolutional neural network maintains a high classification accuracy of over 95 % despite noise and spectral shift. This implies that convolutional neural network can be an ideal classification algorithm in a real combat situation where there is a lot of noise and spectral shift.