• Journal of the Korea Society of Computer and Information
• /
• v.26 no.10
• /
• pp.19-26
• /
• 2021

The performance of deep learning-based malware detection and classification models depends largely on how to construct a feature set to be applied to training. In this paper, we propose an approach to select the optimal feature set to maximize detection performance for CNN-based Android malware detection. The features to be included in the feature set were selected through the Chi-Square test algorithm, which is widely used for feature selection in machine learning and deep learning. To validate the proposed approach, the CNN model was trained using 36 characteristics selected for the CICANDMAL2017 dataset and then the malware detection performance was measured. As a result, 99.99% of Accuracy was achieved in binary classification and 98.55% in multiclass classification.

• Journal of Korean Society for Quality Management
• /
• v.50 no.3
• /
• pp.459-471
• /
• 2022

Purpose: The purpose of this study was to increase prediction accuracy for an anomaly interval identified using an artificial intelligence-based time series anomaly detection technique by establishing a pre-processing process. Methods: Significant variables were extracted by applying feature selection techniques, and anomalies were derived using the TadGAN time series anomaly detection algorithm. After applying machine learning and deep learning methodologies using normal section data (excluding anomaly sections), the explanatory power of the anomaly sections was demonstrated through performance comparison. Results: The results of the machine learning methodology, the performance was the best when SHAP and TadGAN were applied, and the results in the deep learning, the performance was excellent when Chi-square Test and TadGAN were applied. Comparing each performance with the papers applied with a Conventional methodology using the same data, it can be seen that the performance of the MLR was significantly improved to 15%, Random Forest to 24%, XGBoost to 30%, Lasso Regression to 73%, LSTM to 17% and GRU to 19%. Conclusion: Based on the proposed process, when detecting unsupervised learning anomalies of data that are not actually labeled in various fields such as cyber security, financial sector, behavior pattern field, SNS. It is expected to prove the accuracy and explanation of the anomaly detection section and improve the performance of the model.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.12
• /
• pp.3889-3903
• /
• 2022

To address the problem of low detection accuracy due to training noise caused by mislabeling when Tri-training for network intrusion detection (NID), we propose a Tri-training algorithm based on cross entropy and K-nearest neighbors (TCK) for network intrusion detection. The proposed algorithm uses cross-entropy to replace the classification error rate to better identify the difference between the practical and predicted distributions of the model and reduce the prediction bias of mislabeled data to unlabeled data; K-nearest neighbors are used to remove the mislabeled data and reduce the number of mislabeled data. In order to verify the effectiveness of the algorithm proposed in this paper, experiments were conducted on 12 UCI datasets and NSL-KDD network intrusion datasets, and four indexes including accuracy, recall, F-measure and precision were used for comparison. The experimental results revealed that the TCK has superior performance than the conventional Tri-training algorithms and the Tri-training algorithms using only cross-entropy or K-nearest neighbor strategy.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2021.05a
• /
• pp.538-540
• /
• 2021

In this paper, we proposed a smart target detection system that detects and recognizes a designated target to provide relative motion information when performing a target detection mission of a drone. The proposed system focused on developing an algorithm that can secure adequate accuracy (i.e. mAP, IoU) and high real-time at the same time. The proposed system showed an accuracy of close to 1.0 after 100k learning of the Google Inception V2 deep learning model, and the inference speed was about 60-80[Hz] when using a high-performance laptop based on the real-time performance Nvidia GTX 2070 Max-Q. The proposed smart target detection system will be operated like a drone and will be helpful in successfully performing surveillance and reconnaissance missions by automatically recognizing the target using computer image processing and following the target.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.33 no.4
• /
• pp.297-304
• /
• 2015

The change detection algorithms, based on remotely sensed satellite imagery, can be applied to various applications, such as the hazard/disaster analysis and the land monitoring. However, unchanged areas sometimes detected as the changed areas due to various errors in relief displacements and noise pixels, included in the original multi-temporal dataset at the application of unsupervised change detection algorithm. In this research, the object-based changed detection for the high-spatial resolution satellite images is applied by using the IR-MAD (Iteratively Reweighted- Multivariate Alteration Detection), which is one of those representative change detection algorithms. In additionally, we tried to increase the accuracy of change detection results with using the additional information, based on the cross-sharpening method. In the experiment, we used the KOMPSAT-2 satellite sensor, and resulted in the object-based IR-MAD algorithm, representing higher changed detection accuracy than that by the pixel-based IR-MAD. Also, the object-based IR-MAD, focused on cross-sharpened images, increased in accuracy of changed detection, compared to the original object-based IR-MAD. Through these experiments, we could conclude that the land monitoring and the change detection with the high-spatial-resolution satellite imagery can be accomplished efficiency by using the object-based IR-MAD algorithm.

• Imaging Science in Dentistry
• /
• v.34 no.1
• /
• pp.1-5
• /
• 2004

Purpose: To evaluate the diagnostic accuracy of occlusal and proximal caries detection using CCD, CMOS, PSP and film system. Materials and Methods : 32 occlusal and 30 proximal tooth surfaces were radiographed under standardized conditions using 3 digital systems; CCD (CDX-2000HQ, Biomedysis Co., Seoul, Korea), CMOS (Schick, Schick Inc., Long Island, USA), PSP (Digora/sup (R)/FMX, Orion Co./Soredex, Helsinki, Finland) and I film system (Kodak Insight, Eastman Kodak, Rochester, USA). 5 observers examined the radiographs for occlusal and proximal caries using a 5-point confidence scale. The presence of caries was validated histologically and radiographically. Diagnostic accuracy was evaluated using ROC curve areas (Az). Results: Analysis using ROC curves revealed the area under each curve which indicated a diagnostic accuracy. For occlusal caries, Kodak Insight film had an Az of 0.765, CCD one of 0.730, CMOS one of 0.742 and PSP one of 0.735. For proximal caries, Kodak Insight film had an Az of 0.833, CCD one of 0.832, CMOS one of 0.828 and PSP one of 0.868. No statistically significant difference was noted between any of the imaging modalities. Conclusion: CCD, CMOS, PSP and film performed equally well in the detection of occlusal and proximal dental caries. CCD, CMOS and PSP-based digital images provided a level of diagnostic performance comparable to Kodak Insight film.

• Journal of Environmental Health Sciences
• /
• v.41 no.5
• /
• pp.299-304
• /
• 2015

Objectives: A people-counting system measures real-time occupancy through motion detection. Accurate people-counting can be used to calculate suitable ventilation demands. This study determined the optimum motion threshold for a people-counting system. Methods: In a closed room with two occupants moving constantly, different thresholds were tested for the accuracy of a people-counting system. The experiments were conducted at 150, 300, 450 and 600 lux. These levels of brightness included the illumination levels of most public indoor areas. The experiments were repeated with three types of clothing coloration. Results: Overall, a threshold of 16 provided the lowest mean error percentage for the people-counting system. Brightness and clothing color did not have a significant impact on the results. Conclusion: A people-counting system could be used with threshold of 16 for most indoor environments.

• Journal of Electrical Engineering and Technology
• /
• v.10 no.2
• /
• pp.670-675
• /
• 2015

Medical imaging is one of the most powerful tools for gaining information about internal organs and tissues. It is a challenging task to develop sophisticated image analysis methods in order to improve the accuracy of diagnosis. The objective of this paper is to develop a Computer Aided Diagnostics (CAD) scheme for Brain Tumour detection from Magnetic Resonance Image (MRI) using active contour models and to investigate with several approaches for improving CAD performances. The problem in clinical medicine is the automatic detection of brain Tumours with maximum accuracy and in less time. This work involves the following steps: i) Segmentation performed by Fuzzy Clustering with Level Set Method (FCMLSM) and performance is compared with snake models based on Balloon force and Gradient Vector Force (GVF), Distance Regularized Level Set Method (DRLSE). ii) Feature extraction done by Shape and Texture based features. iii) Brain Tumour detection performed by various tree classifiers. Based on investigation FCMLSM is well suited segmentation method and Random Forest is the most optimum classifier for this problem. This method gives accuracy of 97% and with minimum classification error. The time taken to detect Tumour is approximately 2 mins for an examination (30 slices).

• Nuclear Engineering and Technology
• /
• v.54 no.6
• /
• pp.2213-2220
• /
• 2022

The maximum dose delivery at the end of the beam range provides the main advantage of using proton therapy. The range of the proton beam, however, is subject to uncertainties, which limit the clinical benefits of proton therapy and, therefore, accurate in vivo verification of the beam range is desirable. For the beam range verification in spot scanning proton therapy, a prompt gamma detection system, called as gamma electron vertex imaging (GEVI) system, is under development and, in the present study, the performance of the GEVI system in spot scanning proton therapy was predicted with Geant4 Monte Carlo simulations in terms of shift detection sensitivity, accuracy and precision. The simulation results indicated that the GEVI system can detect the interfractional range shifts down to 1 mm shift for the cases considered in the present study. The results also showed that both the evaluated accuracy and precision were less than 1-2 mm, except for the scenarios where we consider all spots in the energy layer for a local shifting. It was very encouraging results that the accuracy and precision satisfied the smallest distal safety margin of the investigated beam energy (i.e., 4.88 mm for 134.9 MeV).

• Proceedings of the Korea Water Resources Association Conference
• /
• 2023.05a
• /
• pp.175-175
• /
• 2023

Detection and monitoring of wildfires are essential for limiting their harmful effects on ecosystems, human lives, and property. In this research, we propose a novel method running in the Google Earth Engine platform for identifying and characterizing burnt regions using a hybrid of Sentinel-1 (C-band synthetic aperture radar) and Sentinel-2 (multispectral photography) images. The 2022 Uljin wildfire, the severest event in South Korean history, is the primary area of our investigation. Given its documented success in remote sensing and land cover categorization applications, we select the Random Forest (RF) method as our primary classifier. Next, we evaluate the performance of our model using multiple accuracy measures, including overall accuracy (OA), Kappa coefficient, and area under the curve (AUC). The proposed method shows the accuracy and resilience of wildfire identification compared to traditional methods that depend on survey data. These results have significant implications for the development of efficient and dependable wildfire monitoring systems and add to our knowledge of how machine learning and remote sensing-based approaches may be combined to improve environmental monitoring and management applications.