學術出版;;Publications

Permanent URI for this communityhdl:11536/1

收集所有交通大學研究人員與學生個人或團體所發表之各項學術著作，涵蓋期刊論文、教師專書、學位論文等項。;;Collections of Publications of NCTU Researchers and Students Like Journal Papers and Thesis.

Browse

🏛️國立陽明交通大學機構典藏

Search Results

Now showing 1 - 10 of 49

Classification of Abnormality in Chest X-Ray Images by Transfer Learning of CheXNet
(2019-01-01)
Almuhayar, Mawanda
;
Lu, Henry Horng-Shing
;
Iriawan, Nur
;
統計學研究所
;
Institute of Statistics
Deep learning development nowadays has attracted a lot of attention because of its effectiveness and good performance. The performance of deep learning in medical images analysis already can compete with medical image experts. However, there are experts that still believe deep learning only efficient for the big datasets, because of deep learning performance in small datasets still not satisfying enough. In this study, it is aimed to build a deep learning model for image classification that can achieve high accuracy using chest X-ray images with a relatively small dataset. We classify chest X-ray into a binary classification which is a normal image and image with abnormalities. We built and experimented our model using the public dataset of Shenzen Hospital dataset. We also use a different type of input based on different images preprocessing so that the model can perform accurate classification. Based on the result, pre-trained CheXNet with a newly trained fully connected network on the cropped dataset can achieve the accuracy 0.8761, the sensitivity 0.8909, and the specificity 0.8621. The performance of the model also influenced by the certain region inside the images, such as other regions outside the lung region and black colored region outside the body region.
A Geometric Approach for Optimal Power Control and Relay Selection in NOMA Wireless Relay Networks
(2020-04-01)
Gau, Rung-Hung
;
Chiu, Hsiao-Ting
;
Liao, Chien-Hsun
;
電信工程研究所
;
Institute of Communications Engineering
In this paper, we propose a geometric approach for optimal power control and relay selection in NOMA wireless relay networks. First, for each pair of relays, to derive an optimal vector of transmission power that maximizes the network throughput, we formulate a non-convex optimization problem. To obtain a closed-form solution for the non-convex optimization problem, we adopt ellipses and hyperbolas for classification of transmission power vectors. In addition, the proposed geometric approach could be used to select an optimal pair of relays and an optimal vector of transmission power in polynomial time. Furthermore, we observe that allocating all transmission power of the BS to send data to the stronger relay is not necessarily optimal in a NOMA wireless relay network. Simulation results show that the proposed geometric approach could significantly increase the end-to-end sum rate and reduce the power consumption of wireless communications for NOMA wireless relay networks with low computational complexity.
Transfer function design for Fourier volume rendering and implementation using GPU
(2008-01-01)
Cheng, Chang-Chieh
;
Ching, Yu-Tai
;
資訊工程學系
;
Department of Computer Science
Volume rendering is a technique for volume visualization. Given a set of N x N x N volume data, the traditional volume rendering methods generally need O(N-3) rendering time. The FVR (Fourier Volume Rendering), that takes advantage of the Fourier slice theorem, takes O(N-2 log N) rendering time once the Fourier Transform of the volume data is available. Thus the FVR is favor to designing a real-time rendering algorithm with a preprocessing step. But the FVR has a disadvantage that resampling in the frequency domain causes artifacts in the spatial domain. Another problem is that the method for designing a transfer function is not obvious. In this paper, we report that by using the spatial domain zero-padding and tri-linear filtering can reduce the artifacts to an acceptable rendered image quality in spatial domain. To design the transfer function, we present a method that the user can define a transfer function by using a Bezier curve first. Based on the linear combination property of the Fourier transform and Bezier curve equation, the volume rendered result can be obtained by adding the weighted frequency domain signals. That mean, once a transfer function is given, we don't have to recompute the Fourier transform of the volume data after the transfer function applied. This technique makes real-time adjustment of transfer function possible.
Using a Novel LDA-Ensemble Framework to Classification of Motor Imagery Tasks for Brain-Computer Interface Applications
(2015-01-01)
Chiu, Ching-Yu
;
Chen, Chih-Yu
;
Lin, Yang-Yin
;
Chen, Shi-An
;
Lin, Chin-Teng
In this paper, we introduce a novel linear discriminate analysis (LDA) ensemble classifier utilizing the Mindo as a brain-computer interface (BCI) device to deal with the problem of motor imagery classification. With regard to the composition of the proposed system, we combine filter bank, sub-band common spatial pattern (SBCSP), LDA together for extracting features of EEG data and classifying the motor imagery with left or right states. In addition, we also employ a gradient descent (GD) algorithm to find the best weight associated with probability fusion function. This novel architecture not only boosts the accuracy of classification but maintains the computational efficiency of the system. Therefore, the proposed LDA-ensemble framework is able to be satisfied with each subject as demonstrated in Section III.
A machine learning approach for acquiring descriptive classification rules of shape contours
(1997-02-01)
Hsu, JC
;
Hwang, SY
;
資訊工程學系
;
Department of Computer Science
We devise a method to generate descriptive classification rules of shape contours by using inductive learning. The classification rules are represented in the form of logic programs. We first transform input objects from pixel representation into predicate representation. The transformation consists of preprocessing, feature extraction and symbolic transformation. We then use FOIL which is an indictive logic programming system to produce classification rules. Experiments on two sets of data were performed to justify our proposed method. Copyright (C) 1997 pattern Recognition Society.
FAST AND ACCURATE IMAGE RECOGNITION USING DEEPLY-FUSED BRANCHY NETWORKS
(2017-01-01)
Huang, Mou-Yue
;
Lai, Ching-Hao
;
Chen, Sin-Horng
;
電機工程學系
;
Department of Electrical and Computer Engineering
In order to achieve higher accuracy of image recognition, deeper and wider networks have been used. However, when the network size gets bigger, its forward inference time also takes longer. To address this problem, we propose Deeply Fused Branchy Network (DFB-Net) by adding small but complete side branches to the target baseline main branch. DFB-Net allows easy-to-discriminate samples to be classified faster. For hard-to-discriminate samples, DFB-Net makes probability fusion by averaging softmax probabilities to make collaborative predictions. Extensive experiments on the two CIFAR datasets show that DFB-Net achieves state-ofthe-art results to obtain an error rate of 3.07% on CIFAR-10 and 16.01% on CIFAR-100. Meanwhile, the forward inference time (with a batch size of 1 and averaged among all test samples) only takes 10.4 ms on CIFAR-10, 18.8 ms on CIFAR-100, using GTX 1080 GPU with cuDNN 5.1.
Agreement Rate Initialized Maximum Likelihood Estimator for Ensemble Classifier Aggregation and Its Application in Brain-Computer Interface
(2016-01-01)
Wu, Dongrui
;
Lawhern, Vernon J.
;
Gordon, Stephen
;
Lance, Brent J.
;
Lin, Chin-Teng
Ensemble learning is a powerful approach to construct a strong learner from multiple base learners. The most popular way to aggregate an ensemble of classifiers is majority voting, which assigns a sample to the class that most base classifiers vote for. However, improved performance can be obtained by assigning weights to the base classifiers according to their accuracy. This paper proposes an agreement rate initialized maximum likelihood estimator (ARIMLE) to optimally fuse the base classifiers. ARIMLE first uses a simplified agreement rate method to estimate the classification accuracy of each base classifier from the unlabeled samples, then employs the accuracies to initialize a maximum likelihood estimator (MLE), and finally uses the expectation-maximization algorithm to refine the MLE. Extensive experiments on visually evoked potential classification in a brain-computer interface application show that ARIMLE outperforms majority voting, and also achieves better or comparable performance with several other state-of-the-art classifier combination approaches.
Phase modulation-Based Response-Inhibition Outcome Prediction in Translational Scenario of Stop-Signal task
(2016-01-01)
Chikara, Rupesh Kumar
;
Ko, Li-Wei
;
生物科技學系
;
腦科學研究中心
;
Department of Biological Science and Technology
In this paper, a method is proposed to predict the resting-state outcomes of participants based on their electroencephalogram (EEG) signals recorded before the successful /unsuccessful response inhibition. The motivation of this study is to enhance the shooter performance for shooting the target, when their EEG patterns show that they are ready. This method can be used in brain-computer interface (BCI) system. In this study, multi-channel EEG from twenty participants are collected by the electrodes placed at different scalp locations in resting-state time. The EEG trials are used to predict two possible outcomes: successful or unsuccessful stop. Four classifiers (QDC, KNNC, PARZENDC, LDC) are used in this study to evaluation the accuracy of our system. Based on the collected time-domain EEG signals, the phase locking value (PLV) from 5-pair electrodes are calculated and then used as the feature input for the classifiers. Our experimental results show that the proposed method prediction accuracy (leave-one-out) was obtained 95% by QDC classifier.
Exploring resting-state EEG complexity before migraine attacks
(2018-06-01)
Cao, Zehong
;
Lai, Kuan-Lin
;
Lin, Chin-Teng
;
Chuang, Chun-Hsiang
;
Chou, Chien-Chen
Objective Entropy-based approaches to understanding the temporal dynamics of complexity have revealed novel insights into various brain activities. Herein, electroencephalogram complexity before migraine attacks was examined using an inherent fuzzy entropy approach, allowing the development of an electroencephalogram-based classification model to recognize the difference between interictal and preictal phases. Methods Forty patients with migraine without aura and 40 age-matched normal control subjects were recruited, and the resting-state electroencephalogram signals of their prefrontal and occipital areas were prospectively collected. The migraine phases were defined based on the headache diary, and the preictal phase was defined as within 72 hours before a migraine attack. Results The electroencephalogram complexity of patients in the preictal phase, which resembled that of normal control subjects, was significantly higher than that of patients in the interictal phase in the prefrontal area (FDR-adjusted p<0.05) but not in the occipital area. The measurement of test-retest reliability (n=8) using the intra-class correlation coefficient was good with r1=0.73 (p=0.01). Furthermore, the classification model, support vector machine, showed the highest accuracy (764%) for classifying interictal and preictal phases using the prefrontal electroencephalogram complexity. Conclusion Entropy-based analytical methods identified enhancement or normalization of frontal electroencephalogram complexity during the preictal phase compared with the interictal phase. This classification model, using this complexity feature, may have the potential to provide a preictal alert to migraine without aura patients.
Robust Feature-Based Automated Multi-View Human Action Recognition System
(2018-01-01)
Chou, Kuang-Pen
;
Prasad, Mukesh
;
Wu, Di
;
Sharma, Nabin
;
Li, Dong-Lin
Automated human action recognition has the potential to play an important role in public security, for example, in relation to the multiview surveillance videos taken in public places, such as train stations or airports. This paper compares three practical, reliable, and generic systems for multiview video-based human action recognition, namely, the nearest neighbor classifier, Gaussian mixture model classifier, and the nearest mean classifier. To describe the different actions performed in different views, view-invariant features are proposed to address multiview action recognition. These features are obtained by extracting the holistic features from different temporal scales which are modeled as points of interest which represent the global spatial-temporal distribution. Experiments and cross-data testing are conducted on the KTH, WEIZMANN, and MuHAVi datasets. The system does not need to be retrained when scenarios are changed which means the trained database can be applied in a wide variety of environments, such as view angle or background changes. The experiment results show that the proposed approach outperforms the existing methods on the KTH and WEIZMANN datasets.

學術出版;;Publications

Browse

Filters

Settings

Sort By

Results per page

🏛️國立陽明交通大學機構典藏

Search Results