• Journal of Broadcast Engineering
• /
• v.24 no.2
• /
• pp.234-242
• /
• 2019

In this paper, we propose a deep neural network which extracts SIFT feature points by determining whether the center pixel of a cropped image is a SIFT feature point. The data set of this network consists of a DIV2K dataset cut into $33{\times}33$ size and uses RGB image unlike SIFT which uses black and white image. The ground truth consists of the RobHess SIFT features extracted by setting the octave (scale) to 0, the sigma to 1.6, and the intervals to 3. Based on the VGG-16, we construct an increasingly deep network of 13 to 23 and 33 convolution layers, and experiment with changing the method of increasing the image scale. The result of using the sigmoid function as the activation function of the output layer is compared with the result using the softmax function. Experimental results show that the proposed network not only has more than 99% extraction accuracy but also has high extraction repeatability for distorted images.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.21 no.2
• /
• pp.189-194
• /
• 2020

This paper analyzes the gradient descent method, which is the one most used for learning neural networks. Learning means updating a parameter so the loss function is at its minimum. The loss function quantifies the difference between actual and predicted values. The gradient descent method uses the slope of the loss function to update the parameter to minimize error, and is currently used in libraries that provide the best deep learning algorithms. However, these algorithms are provided in the form of a black box, making it difficult to identify the advantages and disadvantages of various gradient descent methods. This paper analyzes the characteristics of the stochastic gradient descent method, the momentum method, the AdaGrad method, and the Adadelta method, which are currently used gradient descent methods. The experimental data used a modified National Institute of Standards and Technology (MNIST) data set that is widely used to verify neural networks. The hidden layer consists of two layers: the first with 500 neurons, and the second with 300. The activation function of the output layer is the softmax function, and the rectified linear unit function is used for the remaining input and hidden layers. The loss function uses cross-entropy error.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.7
• /
• pp.842-848
• /
• 2020

Music markets have entered the era of streaming. In order to select and propose music that suits the taste of music consumers, there is an active demand and research on an automatic music genre classification system. We propose a method to improve the accuracy of genre unclassified songs, which was a lack of the previous system, by using a generative adversarial network (GAN) to further develop the automatic voting system for deep learning music genre using Softmax proposed in the previous paper. In the previous study, if the spectrogram of the song was ambiguous to grasp the genre of the song, it was forced to leave it as an unclassified song. In this paper, we proposed a system that increases the accuracy of genre classification of unclassified songs by converting the spectrogram of unclassified songs into an easy-to-read spectrogram using GAN. And the result of the experiment was able to derive an excellent result compared to the existing method.

• Journal of Aerospace System Engineering
• /
• v.17 no.1
• /
• pp.33-41
• /
• 2023

As seen in the case of the Boeing 737 Max accident, the proportion of aircraft software is rapidly increasing. However, it is vulnerable to safety issues. In case of domestic aircraft software, to operate a Light Unmanned Aerial Vehicle (LUAV) less than an empty weight of 150 kg, safety certification is required for an Ultra-Light Vehicle (ULV). However, software certification procedure is not included. Since the use of LUAVs has increased recently, software verification is required. This paper proposed a checklist of LUAV software that could be applied to LUAV referring DO-178C, an aviation software certification standard. A case study of applying the proposed checklist to the Model-based Development-based Helicopter Flight Control Computer (FCC) project currently used by domestic and foreign advanced companies and institutions was conducted.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.67 no.1
• /
• pp.114-123
• /
• 2018

This paper concerns a design and learning method of softmax function neural networks based on K-means clustering. The partial discharge data Information is preliminarily processed through simulation using an Epoxy Mica Coupling sensor and an internal Phase Resolved Partial Discharge Analysis algorithm. The obtained information is processed according to the characteristics of the pattern using a Motor Insulation Monitoring System program. At this time, the processed data are total 4 types that void discharge, corona discharge, surface discharge and slot discharge. The partial discharge data with high dimensional input variables are secondarily processed by principal component analysis method and reduced with keeping the characteristics of pattern as low dimensional input variables. And therefore, the pattern classifier processing speed exhibits improved effects. In addition, in the process of extracting the partial discharge data through the MIMS program, the magnitude of amplitude is divided into the maximum value and the average value, and two pattern characteristics are set and compared and analyzed. In the first half of the proposed partial discharge pattern classifier, the input and hidden layers are classified by using the K-means clustering method and the output of the hidden layer is obtained. In the latter part, the cross entropy error function is used for parameter learning between the hidden layer and the output layer. The final output layer is output as a normalized probability value between 0 and 1 using the softmax function. The advantage of using the softmax function is that it allows access and application of multiple class problems and stochastic interpretation. First of all, there is an advantage that one output value affects the remaining output value and its accompanying learning is accelerated. Also, to solve the overfitting problem, L2-normalization is applied. To prove the superiority of the proposed pattern classifier, we compare and analyze the classification rate with conventional radial basis function neural networks.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.6
• /
• pp.834-841
• /
• 2022

Recently, applying artificial intelligence technologies in various fields of production has drawn an upsurge of research interest due to the increase for smart factory and artificial intelligence technologies. A great deal of effort is being made to introduce artificial intelligence algorithms into the defect detection task. Particularly, detection of defects on the surface of metal has a higher level of research interest compared to other materials (wood, plastics, fibers, etc.). In this paper, we compare and analyze the speed and performance of defect classification by combining machine learning techniques (Support Vector Machine, Softmax Regression, Decision Tree) with dimensionality reduction algorithms (Principal Component Analysis, AutoEncoders) and two convolutional neural networks (proposed method, ResNet). To validate and compare the performance and speed of the algorithms, we have adopted two datasets ((i) public dataset, (ii) actual dataset), and on the basis of the results, the most efficient algorithm is determined.

• Journal of the Korean Society for information Management
• /
• v.40 no.4
• /
• pp.307-327
• /
• 2023

To analyze the impact of word embedding on book titles, this study utilized word embedding models (Word2vec, GloVe, fastText) to generate embedding vectors from book titles. These vectors were then used as classification features for automatic classification. The classifier utilized the k-nearest neighbors (kNN) algorithm, with the categories for automatic classification based on the DDC (Dewey Decimal Classification) main class 300 assigned by libraries to books. In the automatic classification experiment applying word embeddings to book titles, the Skip-gram architectures of Word2vec and fastText showed better results in the automatic classification performance of the kNN classifier compared to the TF-IDF features. In the optimization of various hyperparameters across the three models, the Skip-gram architecture of the fastText model demonstrated overall good performance. Specifically, better performance was observed when using hierarchical softmax and larger embedding dimensions as hyperparameters in this model. From a performance perspective, fastText can generate embeddings for substrings or subwords using the n-gram method, which has been shown to increase recall. The Skip-gram architecture of the Word2vec model generally showed good performance at low dimensions(size 300) and with small sizes of negative sampling (3 or 5).

• Journal of Intelligence and Information Systems
• /
• v.25 no.1
• /
• pp.163-177
• /
• 2019

As smartphones are getting widely used, human activity recognition (HAR) tasks for recognizing personal activities of smartphone users with multimodal data have been actively studied recently. The research area is expanding from the recognition of the simple body movement of an individual user to the recognition of low-level behavior and high-level behavior. However, HAR tasks for recognizing interaction behavior with other people, such as whether the user is accompanying or communicating with someone else, have gotten less attention so far. And previous research for recognizing interaction behavior has usually depended on audio, Bluetooth, and Wi-Fi sensors, which are vulnerable to privacy issues and require much time to collect enough data. Whereas physical sensors including accelerometer, magnetic field and gyroscope sensors are less vulnerable to privacy issues and can collect a large amount of data within a short time. In this paper, a method for detecting accompanying status based on deep learning model by only using multimodal physical sensor data, such as an accelerometer, magnetic field and gyroscope, was proposed. The accompanying status was defined as a redefinition of a part of the user interaction behavior, including whether the user is accompanying with an acquaintance at a close distance and the user is actively communicating with the acquaintance. A framework based on convolutional neural networks (CNN) and long short-term memory (LSTM) recurrent networks for classifying accompanying and conversation was proposed. First, a data preprocessing method which consists of time synchronization of multimodal data from different physical sensors, data normalization and sequence data generation was introduced. We applied the nearest interpolation to synchronize the time of collected data from different sensors. Normalization was performed for each x, y, z axis value of the sensor data, and the sequence data was generated according to the sliding window method. Then, the sequence data became the input for CNN, where feature maps representing local dependencies of the original sequence are extracted. The CNN consisted of 3 convolutional layers and did not have a pooling layer to maintain the temporal information of the sequence data. Next, LSTM recurrent networks received the feature maps, learned long-term dependencies from them and extracted features. The LSTM recurrent networks consisted of two layers, each with 128 cells. Finally, the extracted features were used for classification by softmax classifier. The loss function of the model was cross entropy function and the weights of the model were randomly initialized on a normal distribution with an average of 0 and a standard deviation of 0.1. The model was trained using adaptive moment estimation (ADAM) optimization algorithm and the mini batch size was set to 128. We applied dropout to input values of the LSTM recurrent networks to prevent overfitting. The initial learning rate was set to 0.001, and it decreased exponentially by 0.99 at the end of each epoch training. An Android smartphone application was developed and released to collect data. We collected smartphone data for a total of 18 subjects. Using the data, the model classified accompanying and conversation by 98.74% and 98.83% accuracy each. Both the F1 score and accuracy of the model were higher than the F1 score and accuracy of the majority vote classifier, support vector machine, and deep recurrent neural network. In the future research, we will focus on more rigorous multimodal sensor data synchronization methods that minimize the time stamp differences. In addition, we will further study transfer learning method that enables transfer of trained models tailored to the training data to the evaluation data that follows a different distribution. It is expected that a model capable of exhibiting robust recognition performance against changes in data that is not considered in the model learning stage will be obtained.