• Journal of Korea Multimedia Society
• /
• v.22 no.10
• /
• pp.1168-1177
• /
• 2019

In order to diagnose and prevent Alzheimer's Disease (AD), it is becoming increasingly important to develop a CAD(Computer-aided Diagnosis) system for AD diagnosis, which provides effective treatment for patients by analyzing 3D MRI images. It is essential to apply powerful deep learning algorithms in order to automatically classify stages of Alzheimer's Disease and to develop a Alzheimer's Disease support diagnosis system that has the function of detecting hippocampus and CSF(Cerebrospinal fluid) which are important biomarkers in diagnosis of Alzheimer's Disease. In this paper, for AD diagnosis, we classify a given MRI data into three categories of AD, mild cognitive impairment, and normal control according by applying 3D brain MRI image to the Faster R-CNN model and detect hippocampus and CSF in MRI image. To do this, we use the 2D MRI slice images extracted from the 3D MRI data of the Faster R-CNN, and perform the widely used majority voting algorithm on the resulting bounding box labels for classification. To verify the proposed method, we used the public ADNI data set, which is the standard brain MRI database. Experimental results show that the proposed method achieves impressive classification performance compared with other state-of-the-art methods.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.21 no.6
• /
• pp.117-122
• /
• 2021

Autonomous driving allows cars to drive without people and is being studied very actively thanks to the recent development of artificial intelligence technology. Among artificial intelligence technologies, deep reinforcement learning is used most effectively. Deep reinforcement learning requires us to build a neural network using an appropriate activation function. So far, many activation functions have been suggested, but different performances have been shown depending on the field of application. This paper compares and evaluates the performance of which activation function is effective when using deep reinforcement learning to learn autonomous driving on highways. To this end, the performance metrics to be used in the evaluation were defined and the values of the metrics according to each activation function were compared in graphs. As a result, when Mish was used, the reward was higher on average than other activation functions, and the difference from the activation function with the lowest reward was 9.8%.

• Journal of the Korea Convergence Society
• /
• v.12 no.12
• /
• pp.65-74
• /
• 2021

Recently, studies on Korean spelling correction have been actively conducted based on machine translation and automatic noise generation. These methods generate noise and use as train and data set. This has limitation in that it is difficult to accurately measure performance because it is unlikely that noise other than the noise used for learning is included in the test set In addition, there is no practical error type standard, so the type of error used in each study is different, making qualitative analysis difficult. This paper proposes new 'error type classification' for deep learning-based Korean spelling correction research, and error analysis perform on existing commercialized Korean spelling correctors (System A, B, C). As a result of analysis, it was found the three correction systems did not perform well in correcting other error types presented in this paper other than spacing, and hardly recognized errors in word order or tense.

• Imaging Science in Dentistry
• /
• v.51 no.3
• /
• pp.299-306
• /
• 2021

Purpose: This study aimed to propose a fully automatic landmark identification model based on a deep learning algorithm using real clinical data and to verify its accuracy considering inter-examiner variability. Materials and Methods: In total, 950 lateral cephalometric images from Yonsei Dental Hospital were used. Two calibrated examiners manually identified the 13 most important landmarks to set as references. The proposed deep learning model has a 2-step structure-a region of interest machine and a detection machine-each consisting of 8 convolution layers, 5 pooling layers, and 2 fully connected layers. The distance errors of detection between 2 examiners were used as a clinically acceptable range for performance evaluation. Results: The 13 landmarks were automatically detected using the proposed model. Inter-examiner agreement for all landmarks indicated excellent reliability based on the 95% confidence interval. The average clinically acceptable range for all 13 landmarks was 1.24 mm. The mean radial error between the reference values assigned by 1 expert and the proposed model was 1.84 mm, exhibiting a successful detection rate of 36.1%. The A-point, the incisal tip of the maxillary and mandibular incisors, and ANS showed lower mean radial error than the calibrated expert variability. Conclusion: This experiment demonstrated that the proposed deep learning model can perform fully automatic identification of cephalometric landmarks and achieve better results than examiners for some landmarks. It is meaningful to consider between-examiner variability for clinical applicability when evaluating the performance of deep learning methods in cephalometric landmark identification.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.21 no.5
• /
• pp.197-202
• /
• 2021

Biometric technology using finger veins is receiving a lot of attention due to its high security, convenience and accuracy. And the recent development of deep learning technology has improved the processing speed and accuracy for authentication. However, the training data is a subset of real data not in a certain order or method and the results are not constant. so the amount of data and the complexity of the artificial neural network must be considered. In this paper, the deep learning model of Inception-Resnet-v2 was used to improve the high accuracy of the finger vein recognizer and the performance of the authentication system, We compared and analyzed the performance of the deep learning model of DenseNet-201. The simulations used data from MMCBNU_6000 of Jeonbuk National University and finger vein images taken directly. There is no preprocessing for the image in the finger vein authentication system, and the results are checked through EER.

• Journal of IKEEE
• /
• v.25 no.3
• /
• pp.571-577
• /
• 2021

This paper describes ISO/TR 4804, an international standard to describe how to design and verify autonomous cars to ensure safety and cybersecurity. Goals of ISO/TR 4804 are (1) positive risk balance and (2) avoidance of unreasonable risk. It also 12 principles of safety and cybersecurity to achieve these goals. In the design procedures, it describes (1) 13 capabilities to achieve these safety and cybersecurity principles, (2) hardware and software elements to achieve these capabilities, and (3) a generic logical architecture to combine these elements. In the verification procedures, it describes (1) 5 challenges to ensure safety and cybersecurity, (2) test goals, platforms, and solutions to achieve these challenges, (3) simulation and field operation methods, and (4) verification methods for hardware and software elements. Especially, it regards deep neural network as a software component and it describe design and verification methods of autonomous cars.

• Geomechanics and Engineering
• /
• v.25 no.1
• /
• pp.1-16
• /
• 2021

Machine learning models have been widely used for landslide susceptibility assessment (LSA) in recent years. The large number of inputs or conditioning factors for these models, however, can reduce the computation efficiency and increase the difficulty in collecting data. Feature selection is a good tool to address this problem by selecting the most important features among all factors to reduce the size of the input variables. However, two important questions need to be solved: (1) how do feature selection methods affect the performance of machine learning models? and (2) which feature selection method is the most suitable for a given machine learning model? This paper aims to address these two questions by comparing the predictive performance of 13 feature selection-based machine learning (FS-ML) models and 5 ordinary machine learning models on LSA. First, five commonly used machine learning models (i.e., logistic regression, support vector machine, artificial neural network, Gaussian process and random forest) and six typical feature selection methods in the literature are adopted to constitute the proposed models. Then, fifteen conditioning factors are chosen as input variables and 1,017 landslides are used as recorded data. Next, feature selection methods are used to obtain the importance of the conditioning factors to create feature subsets, based on which 13 FS-ML models are constructed. For each of the machine learning models, a best optimized FS-ML model is selected according to the area under curve value. Finally, five optimal FS-ML models are obtained and applied to the LSA of the studied area. The predictive abilities of the FS-ML models on LSA are verified and compared through the receive operating characteristic curve and statistical indicators such as sensitivity, specificity and accuracy. The results showed that different feature selection methods have different effects on the performance of LSA machine learning models. FS-ML models generally outperform the ordinary machine learning models. The best FS-ML model is the recursive feature elimination (RFE) optimized RF, and RFE is an optimal method for feature selection.

• The Korean Journal of Applied Statistics
• /
• v.34 no.6
• /
• pp.979-993
• /
• 2021

Acoustic scene classification (ASC) categorizes an audio file based on the environment in which it has been recorded. This has long been studied in the detection and classification of acoustic scenes and events (DCASE). In this study, we considered the problem that ASC faces in real-world applications that the model used should have low-complexity. We compared several models that apply light-weight techniques. First, a base CNN model was proposed using log mel-spectrogram, deltas, and delta-deltas features. Second, depthwise separable convolution, linear bottleneck inverted residual block was applied to the convolutional layer, and Quantization was applied to the models to develop a low-complexity model. The model considering low-complexity was similar or slightly inferior to the performance of the base model, but the model size was significantly reduced from 503 KB to 42.76 KB.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.12
• /
• pp.45-52
• /
• 2021

In this paper, we propose a prediction model for osteopenia and osteoporosis based on a convolutional neural network(CNN) using computed tomography(CT) images. In a single CT image, CNN had a limitation in utilizing important local features for diagnosis. So we propose a compound model which has two identical structures. As an input, two different texture images are used, which are converted from a single normalized CT image. The two networks train different information by using dissimilarity loss function. As a result, our model trains various features in a single CT image which includes important local features, then we ensemble them to improve the accuracy of predicting osteopenia and osteoporosis. In experiment results, our method shows an accuracy of 77.11% and the feature visualize of this model is confirmed by using Grad-CAM.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.20 no.4
• /
• pp.43-49
• /
• 2019

The wired and wireless Internet is a useful window to easily acquire various types of media data. On the other hand, the public can easily get the media data including the object to which the personal information is exposed, which is a social problem. In this paper, we propose a method to robustly detect a target object that has exposed personal information using a learning algorithm and effectively block the detected target object area. In the proposed method, only the target object containing the personal information is detected using a neural network-based learning algorithm. Then, a grid-like mosaic is created and overlapped on the target object area detected in the previous step, thereby effectively blocking the object area containing the personal information. Experimental results show that the proposed algorithm robustly detects the object area in which personal information is exposed and effectively blocks the detected area through mosaic processing. The object blocking method presented in this paper is expected to be useful in many applications related to computer vision.