• Journal of Information Processing Systems
• /
• v.18 no.5
• /
• pp.677-687
• /
• 2022

We propose a detection algorithm based on tree-structured convolutional neural networks (TsCNNs) that finds pornography, propaganda, or other inappropriate content on a social media network. The algorithm sequentially applies the typical convolutional neural network (CNN) algorithm in a tree-like structure to minimize classification errors in similar classes, and thus improves accuracy. We implemented the detection system and conducted experiments on a data set comprised of 6 ordinary classes and 11 inappropriate classes collected from the Korean military social network. Each model of the proposed algorithm was trained, and the performance was then evaluated according to the images and videos identified. Experimental results with 20,005 new images showed that the overall accuracy in image identification achieved a high-performance level of 99.51%, and the effectiveness of the algorithm reduced identification errors by the typical CNN algorithm by 64.87 %. By reducing false alarms in video identification from the domain, the TsCNNs achieved optimal performance of 98.11% when using 10 minutes frame-sampling intervals. This indicates that classification through proper sampling contributes to the reduction of computational burden and false alarms.

• The Journal of the Convergence on Culture Technology
• /
• v.10 no.4
• /
• pp.575-580
• /
• 2024

Detecting the status of websites, normal or phishing, is necessary to defend against intelligent phishing attacks. We propose a machine learning-based classification to predict the status of websites. First, we collect information about 'URL', convert it into numerical data, and remove outliers. Second, we apply VIF(Variance Inflation Factors) to understand the correlation and independence between variables. Finally, we develop a phishing website detection model with machine learning-based classifications, which predicts website status. In the test datasets, Random Forest showed the best performance, with precision of 93.74%, recall of 92.26%, and accuracy of 93.14%. In the future, we expect to apply our model to detect various phishing crimes.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.15 no.1
• /
• pp.1-15
• /
• 2016

This study aims to develop a model by which city bus drivers who are likely to cause an accident can be figured out based on the information about their actual driving records. For this purpose, from the information about the actual driving records of the drivers who have caused an accident and those who have not caused any, significance variables related to traffic accidents are drawn, and the accuracy between models is compared for the classification models developed, applying a discriminant analysis and logistic regression analysis. In addition, the developed models are applied to the data on other drivers' driving records to verify the accuracy of the models. As a result of developing a model for the classification of drivers who are likely to cause an accident, when deceleration ($X_{deceleration}$) and acceleration to the right ($Y_{right}$) are simultaneously in action, this variable was drawn as the optimal factor variable of the classification of drivers who had caused an accident, and the prediction model by discriminant analysis classified drivers who had caused an accident at a rate up to 62.8%, and the prediction model by logistic regression analysis could classify those who had caused an accident at a rate up to 76.7%. In addition, as a result of the verification of model predictive power of the models showed an accuracy rate of 84.1%.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.43 no.3 s.309
• /
• pp.76-83
• /
• 2006

A main objective of audio equalizer is for user to tailor acoustic frequency response to increase sound comfort and example applications of audio equalizer includes large-scale audio system to portable audio such as mobile MP3 player. Up to now, all the audio equalizer requires manual setting to equalize frequency bands to create suitable sound quality for each genre of music. In this paper, we propose an intelligent audio graphic equalizer system that automatically classifies the music genre using music content analysis and then the music sound is boosted with the given frequency gains according to the classified musical genre when playback. In order to reproduce comfort sound, the musical genre is determined based on two-step hierarchical algorithm - coarse-level and fine-level classification. It can prevent annoying sound reproduction due to the sudden change of the equalizer gains at the beginning of the music playback. Each stage of the music classification experiments shows at least 80% of success with complete genre classification and equalizer operation within 2 sec. Simple S/W graphical user interface of 3-band automatic equalizer is implemented using visual C on personal computer.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.20 no.2
• /
• pp.189-194
• /
• 2010

In the recommendation system that recommends services to a specific user by using the estimation value of other users for users' recommendation service, collaborative filtering methods are widely used. But such recommendation systems have problems that exact classification is not possible because a specific user is classified to already classified group in the course of clustering and inexact result can be recommended in case of big errors in users' estimation values. In this paper, in order to increase estimation accuracy, the researchers suggest a recommendation system that applies collaborative filtering after reclassifying on the basis of a specific user's classification items and then finding and correcting the estimation values of the users beyond the critical value of time. This system uses a method where a specific user is not classified to already classified group in the course of clustering but a group is reorganized on the basis of the specific user. In addition, the researchers correct estimation information by cutting off the subordinate 10% from the trimmed mean of samples and then applies weight over time to the remaining data. As the result of an experiment, the suggested method demonstrated about 14.9%'s more accurate estimation result in case of using MAE than general collaborative filtering method.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.17 no.4
• /
• pp.563-568
• /
• 2007

This paper presents a methodology for predicting nonlinear time series based on the neural network with weighted fuzzy membership functions (NEWFM). The degree of classification intensity is obtained by bounded sum of weighted fuzzy membership functions extracted by NEWFM, then weighted average defuzzification is used for predicting nonlinear time series. The experimental results demonstrate that NEWFM has the classification capability of 92.22% against the target class of GDP. The time series created by NEWFM model has a relatively close approximation to the GDP which is a typical business cycle indicator, and has been proved to be a useful indicator which has the turning point forecasting capability of average 12 months in the peak point and average 6 months in the trough point during 5th to 8th cyclical period. In addition, NEWFM measures the efficiency of the economic indexes by the feature selection and enables the users to forecast with reduced numbers of 7 among 10 leading indexes while improving the classification rate from 90% to 92.22%.

• Journal of Korea Multimedia Society
• /
• v.11 no.11
• /
• pp.1491-1500
• /
• 2008

This paper proposes an intelligent surveillance system to recognize suspicious patterns of the human behavior by using the Hidden Markov Model. First, the method finds foot area of the human by motion detection algorithm from image sequence of the surveillance camera. Then, these foot locus form observation series of features to learn the HMM. The feature that is position of the human foot is changed to each code that corresponds to a specific label among 16 local partitions of image region. Therefore, specific moving patterns formed by the foot locus are the series of the label numbers. The Baum-Welch algorithm of the HMM learns each suspicious and specific pattern to classify the human behaviors. To recognize the inputted human behavior pattern in a test image, the probabilistic comparison between the learned pattern of the HMM and foot series to be tested decides the categorization of the test pattern. The experimental results show that the method can be applied to detect a suspicious person prowling in corridor.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.19 no.2
• /
• pp.435-442
• /
• 2015

In this paper, we propose a tracking and recognition of pedestrian/vehicle for intelligent multi-visual surveillance system. The intelligent multi-visual surveillance system consists of several fixed cameras and one calibrated PTZ camera, which automatically tracks and recognizes the detected moving objects. The fixed wide-angle cameras are used to monitor large open areas, but the moving objects on the images are too small to view in detail. But, the PTZ camera is capable of increasing the monitoring area and enhancing the image quality by tracking and zooming in on a target. The proposed system is able to determine whether the detected moving objects are pedestrian/vehicle or not using the SVM. In order to reduce the tracking error, an improved camera calibration algorithm between the fixed cameras and the PTZ camera is proposed. Various experimental results show the effectiveness of the proposed system.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.13 no.10
• /
• pp.2180-2188
• /
• 2009

This paper is study on real-time hand gesture recognition system based on vision for intelligent robot control. We are proposed a recognition system using PCA and BP algorithm. Recognition of hand gestures consists of two steps which are preprocessing step using PCA algorithm and classification step using BP algorithm. The PCA algorithm is a technique used to reduce multidimensional data sets to lower dimensions for effective analysis. In our simulation, the PCA is applied to calculate feature projection vectors for the image of a given hand. The BP algorithm is capable of doing parallel distributed processing and expedite processing since it take parallel structure. The BP algorithm recognized in real time hand gestures by self learning of trained eigen hand gesture. The proposed PCA and BP algorithm show improvement on the recognition compared to PCA algorithm.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.20 no.3
• /
• pp.337-342
• /
• 2010

The rise of music resources has led to a parallel rise in the need to manage thousands of songs on user devices. So users are tend to build play-list for manage songs. However the manual selection of songs for creating play-list is bothersome task. This paper proposes an auto play-list recommendation system considering user's context of use and preference. This system has two separate systems: mood and emotion classification system and music recommendation system. Users need to choose just one seed song for reflection their context of use and preference. The system recommends songs before the current song ends in order to fill up user play-list. User also can remove unsatisfied songs from recommended song list to adapt user preferences of the system for the next recommendation precess. The generated play-lists show well defined mood and emotion of music and provide songs that user preferences are reflected.