• The Journal of Korea Robotics Society
• /
• v.14 no.3
• /
• pp.211-220
• /
• 2019

Surface electromyogram (sEMG), which is a bio-electrical signal originated from action potentials of nerves and muscle fibers activated by motor neurons, has been widely used for recognizing motion intention of robotic prosthesis for amputees because it enables a device to be operated intuitively by users without any artificial and additional work. In this paper, we propose a training-free unsupervised sEMG pattern recognition algorithm. It is useful for the gesture recognition for the amputees from whom we cannot achieve motion labels for the previous supervised pattern recognition algorithms. Using the proposed algorithm, we can classify the sEMG signals for gesture recognition and the calculated threshold probability value can be used as a sensitivity parameter for pattern registration. The proposed algorithm was verified by a case study of a patient with partial-hand amputation.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2019.05a
• /
• pp.344-346
• /
• 2019

In this paper, we propose a distance estimation method using the convolutional neural network in Ultra-Wideband (UWB) systems. The training data set used to learn the deep learning model using the convolutional neural network is generated by the MATLAB program and utilizes the IEEE 802.15.4a standard. The performance of the proposed distance estimation method is verified by comparing the threshold based distance estimation technique and the performance comparison used in the conventional distance estimation.

• Journal of the Korean Society for Precision Engineering
• /
• v.10 no.1
• /
• pp.77-88
• /
• 1993

In this paper, we propose a stereo vision system to solve correspondence problem with large disparity and sudden change in environment which result from small distance between camera and working objects. First of all, a specific feature is divided by predfined elementary feature. And then these are combined to obtain coded data for solving correspondence problem. We use Neural Network to extract elementary features from specific feature and to have adaptability to noise and some change of the shape. Fourier transformation and Log-polar mapping are used for obtaining appropriate Neural Network input data which has a shift, scale, and rotation invariability. Finally, we use associative memory to obtain coded data of the specific feature from the combination of elementary features. In spite of specific feature with some variation in shapes, we could obtain satisfactory 3-dimensional data from corresponded codes.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.26 no.2
• /
• pp.93-98
• /
• 2016

For accurate precipitation forecasts the choice of weather factors and prediction method is very important. Recently, machine learning has been widely used for forecasting precipitation, and artificial neural network, one of machine learning techniques, showed good performance. In this paper, we suggest a new method for forecasting precipitation using DBN, one of deep learning techniques. DBN has an advantage that initial weights are set by unsupervised learning, so this compensates for the defects of artificial neural networks. We used past precipitation, temperature, and the parameters of the sun and moon's motion as features for forecasting precipitation. The dataset consists of observation data which had been measured for 40 years from AWS in Seoul. Experiments were based on 8-fold cross validation. As a result of estimation, we got probabilities of test dataset, so threshold was used for the decision of precipitation. CSI and Bias were used for indicating the precision of precipitation. Our experimental results showed that DBN performed better than MLP.

• Journal of Internet Computing and Services
• /
• v.20 no.4
• /
• pp.91-102
• /
• 2019

In autonomous driving systems, the ability to classify pedestrians in images captured by cameras is very important for pedestrian safety. In the past, after extracting features of pedestrians with HOG(Histogram of Oriented Gradients) or SIFT(Scale-Invariant Feature Transform), people classified them using SVM(Support Vector Machine). However, extracting pedestrian characteristics in such a handcrafted manner has many limitations. Therefore, this paper proposes a method to classify pedestrians reliably and effectively using CNN's(Convolutional Neural Network) deep features and transfer learning. We have experimented with both the fixed feature extractor and the fine-tuning methods, which are two representative transfer learning techniques. Particularly, in the fine-tuning method, we have added a new scheme, called M-Fine(Modified Fine-tuning), which divideslayers into transferred parts and non-transferred parts in three different sizes, and adjusts weights only for layers belonging to non-transferred parts. Experiments on INRIA Person data set with five CNN models(VGGNet, DenseNet, Inception V3, Xception, and MobileNet) showed that CNN's deep features perform better than handcrafted features such as HOG and SIFT, and that the accuracy of Xception (threshold = 0.5) isthe highest at 99.61%. MobileNet, which achieved similar performance to Xception and learned 80% fewer parameters, was the best in terms of efficiency. Among the three transfer learning schemes tested above, the performance of the fine-tuning method was the best. The performance of the M-Fine method was comparable to or slightly lower than that of the fine-tuningmethod, but higher than that of the fixed feature extractor method.

• Journal of the Korean Data and Information Science Society
• /
• v.27 no.2
• /
• pp.353-362
• /
• 2016

By Wikipedia, data mining is the process of discovering patterns in a big data set involving methods at the intersection of association rule, decision tree, clustering, artificial intelligence, machine learning. and database systems. Association rule is a method for discovering interesting relations between items in large transactions by interestingness measures. Association rule interestingness measures play a major role within a knowledge discovery process in databases, and have been developed by many researchers. Among them, the Hellinger measure is a good association threshold considering the information content and the generality of a rule. But it has the drawback that it can not determine the direction of the association. In this paper we proposed a signed Hellinger measure to be able to interpret operationally, and we checked three conditions of association threshold. Furthermore, we investigated some aspects through a few examples. The results showed that the signed Hellinger measure was better than the Hellinger measure because the signed one was able to estimate the right direction of association.

• Journal of Biomedical Engineering Research
• /
• v.20 no.4
• /
• pp.435-442
• /
• 1999

A fuzzy-ART(adaptive resonance theory) network for the PVC(premature ventricular contraction) classification using wavelet coefficient is designed. This network consists of the feature extraction and learning of the fuzzy-ART network. In the first step, we have detected the QRS from the ECG signal in order to set the threshold range for feature extraction and the detected QRS was divided into several frequency bands by wavelet transformation using Haar wavelet. Among the low-frequency bands, only the 6th coefficient(D6) are selected as the input feature. After that, the fuzzy-ART network for classification of the PVC is learned by using input feature which comprises of binary data converted by applying threshold to D6. The MIT/BIH database including the PVC is used for the evaluation. The designed fuzzy-ART network showed the PVC classification ratio of 96.52%.

• The KIPS Transactions:PartB
• /
• v.10B no.7
• /
• pp.853-862
• /
• 2003

Many more use e-mail purely on a personal basis and the pool of e-mail users is growing daily. Also, the amount of mails, which are transmitted in electronic commerce, is getting more and more. Because of its convenience, a mass of spam mails is flooding everyday. And yet automated techniques for learning to filter e-mail have yet to significantly affect the e-mail market. This paper suggests Web Mail Filtering Agent for Personalized Classification, which automatically manages mails adjusting to the user. It is based on web mail, which can be logged in any time, any place and has no limitation in any system. In case new mails are received, it first makes some personal rules in use of the result of observation ; and based on the personal rules, it automatically classifies the mails into categories according to the contents of mails and saves the classified mails in the relevant folders or deletes the unnecessary mails and spam mails. And, we applied Bayesian Algorithm using Dynamic Threshold for our system's accuracy.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.48 no.5
• /
• pp.55-63
• /
• 2011

This paper describes new moving target detection technique combining two algorithms to detect targets and reject clutters in video frame images for surveillance system: One obtains the region of moving target using phase correlation method using $N{\times}M$ sub-block images in frequency domain. The other uses adaptive threshold using learning weight for extracting target candidates in subtracted image. The block region with moving target can be obtained using the characteristics that the highest value of phase correlation depends on the movement of largest image in block. This technique can be used in camera motion environment calculating and compensating camera movement using FFT phase correlation between input video frame images. The experimental results show that the proposed algorithm accurately detects target(s) with a low false alarm rate in variety environment using the receiver operating characteristics (ROC) curve.

• Journal of Intelligence and Information Systems
• /
• v.19 no.2
• /
• pp.125-140
• /
• 2013

We call a data set in which the number of records belonging to a certain class far outnumbers the number of records belonging to the other class, 'imbalanced data set'. Most of the classification techniques perform poorly on imbalanced data sets. When we evaluate the performance of a certain classification technique, we need to measure not only 'accuracy' but also 'sensitivity' and 'specificity'. In a customer churn prediction problem, 'retention' records account for the majority class, and 'churn' records account for the minority class. Sensitivity measures the proportion of actual retentions which are correctly identified as such. Specificity measures the proportion of churns which are correctly identified as such. The poor performance of the classification techniques on imbalanced data sets is due to the low value of specificity. Many previous researches on imbalanced data sets employed 'oversampling' technique where members of the minority class are sampled more than those of the majority class in order to make a relatively balanced data set. When a classification model is constructed using this oversampled balanced data set, specificity can be improved but sensitivity will be decreased. In this research, we developed a hybrid model of support vector machine (SVM), artificial neural network (ANN) and decision tree, that improves specificity while maintaining sensitivity. We named this hybrid model 'hybrid SVM model.' The process of construction and prediction of our hybrid SVM model is as follows. By oversampling from the original imbalanced data set, a balanced data set is prepared. SVM_I model and ANN_I model are constructed using the imbalanced data set, and SVM_B model is constructed using the balanced data set. SVM_I model is superior in sensitivity and SVM_B model is superior in specificity. For a record on which both SVM_I model and SVM_B model make the same prediction, that prediction becomes the final solution. If they make different prediction, the final solution is determined by the discrimination rules obtained by ANN and decision tree. For a record on which SVM_I model and SVM_B model make different predictions, a decision tree model is constructed using ANN_I output value as input and actual retention or churn as target. We obtained the following two discrimination rules: 'IF ANN_I output value <0.285, THEN Final Solution = Retention' and 'IF ANN_I output value ${\geq}0.285$, THEN Final Solution = Churn.' The threshold 0.285 is the value optimized for the data used in this research. The result we present in this research is the structure or framework of our hybrid SVM model, not a specific threshold value such as 0.285. Therefore, the threshold value in the above discrimination rules can be changed to any value depending on the data. In order to evaluate the performance of our hybrid SVM model, we used the 'churn data set' in UCI Machine Learning Repository, that consists of 85% retention customers and 15% churn customers. Accuracy of the hybrid SVM model is 91.08% that is better than that of SVM_I model or SVM_B model. The points worth noticing here are its sensitivity, 95.02%, and specificity, 69.24%. The sensitivity of SVM_I model is 94.65%, and the specificity of SVM_B model is 67.00%. Therefore the hybrid SVM model developed in this research improves the specificity of SVM_B model while maintaining the sensitivity of SVM_I model.