• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.32 no.6
• /
• pp.1081-1090
• /
• 2022

The face recognition model is used for identity recognition of smartphones, providing convenience to many users. As a result, the security review of the DNN model is becoming important, with adversarial attacks present as a well-known vulnerability of the DNN model. Adversarial attacks have evolved to decision-based attack techniques that use only the recognition results of deep learning models to perform attacks. However, existing decision-based attack technique[14] have a problem that requires a large number of queries when generating adversarial examples. In particular, it takes a large number of queries to approximate the gradient. Therefore, in this paper, we propose a method of generating adversarial examples using orthogonal space sampling and dimensionality reduction sampling to avoid wasting queries that are consumed to approximate the gradient of existing decision-based attack technique[14]. Experiments show that our method can reduce the perturbation size of adversarial examples by about 2.4 compared to existing attack technique[14] and increase the attack success rate by 14% compared to existing attack technique[14]. Experimental results demonstrate that the adversarial example generation method proposed in this paper has superior attack performance.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.7
• /
• pp.996-1003
• /
• 2022

In this paper, we present a new method for determining the scan line and range for vision-based 1-D barcode recognition. This is a study on how to detect valid barcode representative points and directions by applying the DBSCAN clustering method based on block HOG (histogram of gradient) and determine scan lines and barcode crosslines based on this. In this paper, the minimum and maximum search techniques were applied to determine the cross-line range of barcodes based on the obtained scan lines. This can be applied regardless of the barcode size. This technique enables barcode recognition even by detecting only a partial area of the barcode, and does not require rotation to read the code after detecting the barcode area. In addition, it is possible to detect barcodes of various sizes. Various experimental results are presented to evaluate the performance of the proposed technique in this paper.

• International Journal of Naval Architecture and Ocean Engineering
• /
• v.13 no.1
• /
• pp.194-207
• /
• 2021

Vortex-Induced Vibration (VIV) is a major contributor to the fatigue damage of marine risers which are often arranged in an array configuration. In addition to helical strakes and fairings, studies have been strived in searching for possible VIV suppression techniques. Inspired by giant Saguaro Cacti, flexible cylinders of different cactus-shaped cross sections were tested in a water tunnel facility, and test results showed that cactus-like body shapes reduced VIV responses of a cylinder at no cost of significant increase of drag. A series of experiments were conducted on a pair of two tandem-arranged flexible cylinders and an array of four cylinders in a square configuration to investigate the effects of wake on the dynamic responses of cylinders and the VIV mitigation effectiveness of the cactus-like body shape. Results showed that the cylinders in a square configuration, either at the upstream or downstream positions, might have larger dynamic responses than those of a single cylinder. The cactus-like body shape could mitigate VIV responses of cylinders at upstream positions in an array configuration; however, similar to helical strakes, the mitigation efficiency was reduced on downstream cylinders. Note that the cactus-like cross-sectional shape investigated was not optimized for VIV suppression. The present study indicates that the modification of the cross-sectional shape of a cylinder to a well-designed cactus-like shape may be used as an alternative technique to mitigate the VIV of marine risers.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.12
• /
• pp.1890-1897
• /
• 2021

People counting is an important technology to provide application services such as smart home, smart building, smart car, etc. Due to the social distancing of COVID-19, the people counting technology attracted public attention. People counting system can be implemented in various ways such as camera, sensor, wireless, etc. according to service requirements. People counting system using WiFi AP uses WiFi CSI data that reflects multipath information. This technology is an effective solution implementing indoor with low cost. The conventional WiFi CSI-based people counting technologies have low accuracy that obstructs the high quality service. This paper proposes a deep learning people counting system based on WiFi CSI data. Data preprocessing using auto-encoder, data augmentation that transform WiFi CSI data, and a proposed deep learning model improve the accuracy of people counting. In the experimental result, the proposed approach shows 89.29% accuracy in 6 subjects.

• Journal of Platform Technology
• /
• v.10 no.3
• /
• pp.44-50
• /
• 2022

Research on facial age estimation is being actively conducted because it is used in various application fields. Facial images taken in various environments often have occlusions, and there is a problem in that performance of age estimation is degraded. Therefore, we propose age estimation method by creating an occluded part using image extrapolation technology to improve the age estimation performance of an occluded face image. In order to confirm the effect of occlusion in the image on the age estimation performance, an image with occlusion is generated using a mask image. The occluded part of facial image is restored using SpiralNet, which is one of the image extrapolation techniques, and it is a method to create an occluded part while crossing the edge of an image. Experimental results show that age estimation performance of occluded facial image is significantly degraded. It was confirmed that the age estimation performance is improved when using a face image with reconstructed occlusions using SpiralNet by experiments.

• KIPS Transactions on Computer and Communication Systems
• /
• v.11 no.8
• /
• pp.259-268
• /
• 2022

With prosperous usage of cloud services to improve management efficiency due to the explosive increase in data volume, various cryptographic techniques are being applied in order to preserve data privacy. In spite of the vast computing resources of cloud systems, decrease in storage efficiency caused by redundancy of data outsourced from multiple users acts as a factor that significantly reduces service efficiency. Among several approaches on privacy-preserving data deduplication over encrypted data, in this paper, the research results for improving efficiency of encrypted data deduplication using trusted execution environment (TEE) published in the recent USENIX ATC are analysed in terms of security and efficiency of the participating entities. We present a way to improve the stability of a key-managing server by integrating it with individual clients, resulting in secure deduplication without independent key servers. The experimental results show that the communication efficiency of the proposed approach can be improved by about 30% with the effect of a distributed key server while providing robust security guarantees as the same level of the previous research.

• Journal of the Semiconductor & Display Technology
• /
• v.22 no.1
• /
• pp.134-138
• /
• 2023

This study focuses on the analysis of the results of computational fluid dynamics simulations of mist-chemical vapor deposition for the growth of an epitaxial wafer in power semiconductor technology using artificial intelligence techniques. The conventional approach of predicting the uniformity of the deposited layer using computational fluid dynamics and design of experimental takes considerable time. To overcome this, artificial intelligence method, which is widely used for optimization, automation, and prediction in various fields, was utilized to analyze the computational fluid dynamics simulation results. The computational fluid dynamics simulation results were analyzed using a supervised deep neural network model for regression analysis. The predicted results were evaluated quantitatively using Euclidean distance calculations. And the Bayesian optimization was used to derive the optimal condition, which results obtained through deep neural network training showed a discrepancy of approximately 4% when compared to the results obtained through computational fluid dynamics analysis. resulted in an increase of 146.2% compared to the previous computational fluid dynamics simulation results. These results are expected to have practical applications in various fields.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.18 no.3
• /
• pp.89-100
• /
• 2023

Recently, on-device artificial intelligence (AI) solutions using mobile devices and embedded edge devices have emerged in various fields, such as computer vision, to address network traffic burdens, low-energy operations, and security problems. Although vision transformer deep learning models have outperformed conventional convolutional neural network (CNN) models in computer vision, they require more computations and parameters than CNN models. Thus, they are not directly applicable to embedded edge devices with limited hardware resources. Many researchers have proposed various model compression methods or lightweight architectures for vision transformers; however, there are only a few studies evaluating the effects of model compression techniques of vision transformers on performance. Regarding this problem, this paper presents a performance evaluation of vision transformers on embedded platforms. We investigated the behaviors of three vision transformers: DeiT, LeViT, and MobileViT. Each model performance was evaluated by accuracy and inference time on edge devices using the ImageNet dataset. We assessed the effects of the quantization method applied to the models on latency enhancement and accuracy degradation by profiling the proportion of response time occupied by major operations. In addition, we evaluated the performance of each model on GPU and EdgeTPU-based edge devices. In our experimental results, LeViT showed the best performance in CPU-based edge devices, and DeiT-small showed the highest performance improvement in GPU-based edge devices. In addition, only MobileViT models showed performance improvement on EdgeTPU. Summarizing the analysis results through profiling, the degree of performance improvement of each vision transformer model was highly dependent on the proportion of parts that could be optimized in the target edge device. In summary, to apply vision transformers to on-device AI solutions, either proper operation composition and optimizations specific to target edge devices must be considered.

• Advanced Industrial SCIence
• /
• v.2 no.3
• /
• pp.8-14
• /
• 2023

The majority of IoT devices already employ AIoT, however there are still numerous issues that need to be resolved before AI applications can be deployed. In order to more effectively distribute IoT edge resources, this paper propose a machine learning-based approach to managing IoT edge resources. The suggested method constantly improves the allocation of IoT resources by identifying IoT edge resource trends using machine learning. IoT resources that have been optimized make use of machine learning convolution to reliably sustain IoT edge resources that are always changing. By storing each machine learning-based IoT edge resource as a hash value alongside the resource of the previous pattern, the suggested approach effectively verifies the resource as an attack pattern in a distributed AIoT context. Experimental results evaluate energy efficiency in three different test scenarios to verify the integrity of IoT Edge resources to see if they work well in complex environments with heterogeneous computational hardware.

• International Journal of Computer Science & Network Security
• /
• v.23 no.9
• /
• pp.55-64
• /
• 2023

The Internet of Things (IoT) is the combination of the internet and various sensing devices. IoT security has increasingly attracted extensive attention. However, significant losses appears due to malicious attacks. Therefore, intrusion detection, which detects malicious attacks and their behaviors in IoT devices plays a crucial role in IoT security. The intrusion detection system, namely IDS should be executed efficiently by conducting classification and efficient feature extraction techniques. To effectively perform Intrusion detection in IoT applications, a novel method based on a Conventional Neural Network (CNN) for classification and an improved Genetic Algorithm (GA) for extraction is proposed and implemented. Existing issues like failing to detect the few attacks from smaller samples are focused, and hence the proposed novel CNN is applied to detect almost all attacks from small to large samples. For that purpose, the feature selection is essential. Thus, the genetic algorithm is improved to identify the best fitness values to perform accurate feature selection. To evaluate the performance, the NSL-KDDCUP dataset is used, and two datasets such as KDDTEST²¹ and KDDTEST⁺ are chosen. The performance and results are compared and analyzed with other existing models. The experimental results show that the proposed algorithm has superior intrusion detection rates to existing models, where the accuracy and true positive rate improve and the false positive rate decrease. In addition, the proposed algorithm indicates better performance on KDDTEST⁺ than KDDTEST²¹ because there are few attacks from minor samples in KDDTEST⁺. Therefore, the results demonstrate that the novel proposed CNN with the improved GA can identify almost every intrusion.