• The Journal of the Acoustical Society of Korea
• /
• v.42 no.5
• /
• pp.395-401
• /
• 2023

Sound Event Detection (SED) aims to identify not only sound category but also time interval for target sounds in an audio waveform. It is a critical technique in field of acoustic surveillance system and monitoring system. Recently, various models have introduced through Detection and Classification of Acoustic Scenes and Events (DCASE) Task 4. This paper explored how to design optimal parameters of DenseNet based model, which has led to outstanding performance in other recognition system. In experiment, DenseRNN as an SED model consists of DensNet-BC and bi-directional Gated Recurrent Units (GRU). This model is trained with Mean teacher model. With an event-based f-score, evaluation is performed depending on parameters, related to model architecture as well as model training, under the assessment protocol of DCASE task4. Experimental result shows that the performance goes up and has been saturated to near the best. Also, DenseRNN would be trained more effectively without dropout technique.

• Convergence Security Journal
• /
• v.23 no.5
• /
• pp.101-106
• /
• 2023

Recently, cyberattacks are increasing in social engineering attacks using intelligent and continuous phishing sites and hacking techniques using malicious code. As personal security becomes important, there is a need for a method and a solution for determining whether a malicious URL exists using a web application. In this paper, we would like to find out each feature and limitation by comparing highly accurate techniques for detecting malicious URLs. Compared to classification algorithm models using features such as web flat panel DB and based URL detection sites, we propose an efficient URL anomaly detection technique.

• Journal of the Korea Society of Computer and Information
• /
• v.29 no.2
• /
• pp.31-41
• /
• 2024

Deep learning models show excellent performance in tasks such as image classification and object detection in the field of computer vision, and are used in various ways in actual industrial sites. Recently, research on improving robustness has been actively conducted, along with pointing out that this deep learning model is vulnerable to hostile examples. A hostile example is an image in which small noise is added to induce misclassification, and can pose a significant threat when applying a deep learning model to a real environment. In this paper, we tried to confirm the robustness of the edge-learning classification model and the performance of the adversarial example detection model using it for adversarial examples of various algorithms. As a result of robustness experiments, the basic classification model showed about 17% accuracy for the FGSM algorithm, while the edge-learning models maintained accuracy in the 60-70% range, and the basic classification model showed accuracy in the 0-1% range for the PGD/DeepFool/CW algorithm, while the edge-learning models maintained accuracy in 80-90%. As a result of the adversarial example detection experiment, a high detection rate of 91-95% was confirmed for all algorithms of FGSM/PGD/DeepFool/CW. By presenting the possibility of defending against various hostile algorithms through this study, it is expected to improve the safety and reliability of deep learning models in various industries using computer vision.

• ETRI Journal
• /
• v.46 no.3
• /
• pp.513-525
• /
• 2024

Deep neural networks trained on labeled medical data face major challenges owing to the economic costs of data acquisition through expensive medical imaging devices, expert labor for data annotation, and large datasets to achieve optimal model performance. The heterogeneity of diseases, such as Alzheimer's disease, further complicates deep learning because the test cases may substantially differ from the training data, possibly increasing the rate of false positives. We propose a reconstruction-based self-supervised anomaly detection model to overcome these challenges. It has a dual-subnetwork encoder that enhances feature encoding augmented by skip connections to the decoder for improving the gradient flow. The novel encoder captures local and global features to improve image reconstruction. In addition, we introduce an entropy-based image conversion method. Extensive evaluations show that the proposed model outperforms benchmark models in anomaly detection and classification using an encoder. The supervised and unsupervised models show improved performances when trained with data preprocessed using the proposed image conversion method.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.45 no.4
• /
• pp.100-106
• /
• 2008

In this paper we apply a discriminative weight training employing power spectral flatness measure (PSFM) to a statistical model-based voice activity detection (VAD) in various noise environments. In our approach, the VAD decision rule is expressed as the geometric mean of optimally weighted likelihood ratio test (LRT) based on a minimum classification error (MCE) method which is different from the previous works in th at different weights are assigned to each frequency bin and noise environments depending on PSFM. According to the experimental results, the proposed approach is found to be effective for the statistical model-based VAD using the LRT.

• Journal of Internet Technology
• /
• v.22 no.4
• /
• pp.903-911
• /
• 2021

As cyber-attacks on Cyber-Physical System (CPS) become more diverse and sophisticated, it is important to quickly detect malicious behaviors occurring in CPS. Since CPS can collect sensor data in near real time throughout the process, there have been many attempts to detect anomaly behavior through normal behavior learning from the perspective of data-driven security. However, since the CPS datasets are big data and most of the data are normal data, it has always been a great challenge to analyze the data and implement the anomaly detection model. In this paper, we propose and evaluate the Clustered Deep One-Class Classification (CD-OCC) model that combines the clustering algorithm and deep learning (DL) model using only a normal dataset for anomaly detection. We use auto-encoder to reduce the dimensions of the dataset and the K-means clustering algorithm to classify the normal data into the optimal cluster size. The DL model trains to predict clusters of normal data, and we can obtain logit values as outputs. The derived logit values are datasets that can better represent normal data in terms of knowledge distillation and are used as inputs to the OCC model. As a result of the experiment, the F1 score of the proposed model shows 0.93 and 0.83 in the SWaT and HAI dataset, respectively, and shows a significant performance improvement over other recent detectors such as Com-AE and SVM-RBF.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.17 no.4
• /
• pp.577-586
• /
• 2022

Detection and classification of steel surface defects are critical for product quality control in the steel industry. However, due to its low accuracy and slow speed, the traditional approach cannot be effectively used in a production line. The current, widely used algorithm (based on deep learning) has an accuracy problem, and there are still rooms for development. This paper proposes a method of steel surface defect detection combining EfficientNetV2 for image classification and YOLOv5 as an object detector. Shorter training time and high accuracy are advantages of this model. Firstly, the image input into EfficientNetV2 model classifies defect classes and predicts probability of having defects. If the probability of having a defect is less than 0.25, the algorithm directly recognizes that the sample has no defects. Otherwise, the samples are further input into YOLOv5 to accomplish the defect detection process on the metal surface. Experiments show that proposed model has good performance on the NEU dataset with an accuracy of 98.3%. Simultaneously, the average training speed is shorter than other models.

• International Journal of Computer Science & Network Security
• /
• v.23 no.11
• /
• pp.133-141
• /
• 2023

This work take deeper analysis of Adaptive Moment Estimation (Adam) and Adam with Weight Decay (AdamW) implementation in real world text classification problem (DGA Malware Detection). AdamW is introduced by decoupling weight decay from L2 regularization and implemented as improved optimizer. This work introduces a novel implementation of AdamW variant as AdamW+ by further simplifying weight decay implementation in AdamW. DGA malware detection LSTM models results for Adam, AdamW and AdamW+ are evaluated on various DGA families/ groups as multiclass text classification. Proposed AdamW+ optimizer results has shown improvement in all standard performance metrics over Adam and AdamW. Analysis of outcome has shown that novel optimizer has outperformed both Adam and AdamW text classification based problems.

• Convergence Security Journal
• /
• v.21 no.4
• /
• pp.69-75
• /
• 2021

Various image formats are being used when attempting to construct a malware detection model based on CNN. However, most previous studies emphasize only the detection or classification performance, and do not take into account the possible impact of image format on detection performance and resource usage. Therefore, in this paper, we analyze how the input image formats affect detection performance and resources usage when detecting android malware based on CNN. The dataset used in the experiment is the CICAndMal2017 Dataset. Subdataset extracted from the CICAndMal2017 Dataset were converted into images in four formats: BMP, JPG, PNG, and TIFF. We then trained our CNN model and measured malware detection performance and resource usage. As a result, there was no sifnificant difference between detection performance and the GPU/RAM usage, even if the image format changed. However, we found that the file size of the generated images varied by up to six times depending on the image format, and that significant differences occurred in the training time.

• International Journal of Advanced Culture Technology
• /
• v.9 no.3
• /
• pp.305-314
• /
• 2021

Currently, due to COVID-19, household waste has a lot of impact on the environment due to packaging of food delivery. In this paper, we design and implement Faster-RCNN, SSD, and YOLOv4 models for municipal waste detection and classification. The data set explores two types of plastics, which account for a large proportion of household waste, and the types of aluminum cans. To classify the plastic type and the aluminum can type, 1,083 aluminum can types and 1,003 plastic types were studied. In addition, in order to increase the accuracy, we compare and evaluate the loss value and the accuracy value for the detection of municipal waste classification using Faster-RCNN, SDD, and YoloV4 three models. As a final result of this paper, the average precision value of the SSD model is 99.99%, the average precision value of plastics is 97.65%, and the mAP value is 99.78%, which is the best result.