• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.44 no.1
• /
• pp.125-132
• /
• 2024

Optical satellite images are being used for national security and collection of information, and their utilization is increasing. However, it acquires low-quality images that are not suitable for the user's requirement due to weather conditions and time constraints. In this paper, a deep learning-based conversion of image and colorization model referring to high-resolution SAR images was created to simulate the occluded area with clouds of optical satellite images. The model was experimented according to the type of algorithm applied and input data, and each simulated images was compared and analyzed. In particular, the amount of pixel value information between the input black-and-white image and the SAR image was similarly constructed to overcome the problem caused by the relatively lack of color information. As a result of the experiment, the histogram distribution of the simulated image learned with the Gray-scale image and the high-resolution SAR image was relatively similar to the original image. In addition, the RMSE value was about 6.9827 and the PSNR value was about 31.3960 calculated for quantitative analysis.

• Journal of the Korea Society of Computer and Information
• /
• v.29 no.2
• /
• pp.101-108
• /
• 2024

AI-OCR (Artificial Intelligence Optical Character Recognition) combines OCR technology with Artificial Intelligence to overcome limitations that required human intervention. To enhance the performance of AI-OCR, training on diverse data sets is essential. However, the recognition rate declines when image colors have similar brightness levels. To solve this issue, this study employs Homomorphic filtering as a preprocessing step to clearly differentiate color levels, thereby increasing text recognition rates. While Homomorphic filtering is ideal for text extraction because of its ability to adjust the high and low frequency components of an image separately using a gamma value, it has the downside of requiring manual adjustments to the gamma value. This research proposes a range for gamma threshold values based on tests involving image contrast, brightness, and entropy. Experimental results using the proposed range of gamma values in Homomorphic filtering suggest a high likelihood for effective AI-OCR performance.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.24 no.1
• /
• pp.1-8
• /
• 2024

On-device AI technology, which can operate AI models at the edge devices to support real-time processing and privacy enhancement, is attracting attention. As intelligent IoT is applied to various industries, services utilizing the on-device AI technology are increasing significantly. However, general deep learning models require a lot of computational resources for inference and learning. Therefore, various lightweighting methods such as quantization and pruning have been suggested to operate deep learning models in embedded edge devices. Among the lightweighting methods, we analyze how to lightweight and apply deep learning models to edge computing devices, focusing on pruning technology in this paper. In particular, we utilize dynamic and static pruning techniques to evaluate the inference speed, accuracy, and memory usage of a lightweight AI vision model. The content analyzed in this paper can be used for intelligent video control systems or video security systems in autonomous vehicles, where real-time processing are highly required. In addition, it is expected that the content can be used more effectively in various IoT services and industries.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.24 no.1
• /
• pp.31-36
• /
• 2024

Recently, mobile apps that utilize AI technologies are increasing. In the personal mobile environment, performance degradation may occur during the training phase of large AI workload due to limitations in memory capacity. In this paper, we extract memory reference traces of AI workloads and analyze their characteristics. From this analysis, we observe that AI workloads can cause frequent storage access due to weak temporal locality and irregular popularity bias during memory write operations, which can degrade the performance of mobile devices. Based on this observation, we discuss ways to efficiently manage memory write operations of AI workloads using persistent memory-based swap devices. Through simulation experiments, we show that the system architecture proposed in this paper can improve the I/O time of mobile systems by more than 80%.

• Smart Media Journal
• /
• v.12 no.11
• /
• pp.36-47
• /
• 2023

Industrial Control System(ICS), which controls facilities at major industrial sites, is increasingly connected to other systems through networks. With this integration and the development of intelligent attacks that can lead to a single external intrusion as a whole system paralysis, the risk and impact of security on industrial control systems are increasing. As a result, research on how to protect and detect cyber attacks is actively underway, and deep learning models in the form of unsupervised learning have achieved a lot, and many abnormal detection technologies based on deep learning are being introduced. In this study, we emphasize the application of preprocessing methodologies to enhance the anomaly detection performance of deep learning models on time series data. The results demonstrate the effectiveness of a Wavelet Transform (WT)-based noise reduction methodology as a preprocessing technique for deep learning-based anomaly detection. Particularly, by incorporating sensor characteristics through clustering, the differential application of the Dual-Tree Complex Wavelet Transform proves to be the most effective approach in improving the detection performance of cyber attacks.

• Journal of Korean Society of Disaster and Security
• /
• v.16 no.4
• /
• pp.129-136
• /
• 2023

A special offshore bridge with a high pylon has special structural features.Special offshore bridges have inspection blind spots that are difficult to visually inspect. To solve this problem, safety inspection methods using drones are being studied. In this study, image data of the pylon of a special offshore bridge was acquired using a drone. In addition, an artificial intelligence algorithm was developed to detect damage to the pylon surface. The AI algorithm utilized a deep learning network with different structures. The algorithm applied the stacking ensemble learning method to build a model that formed the ensemble and collect the results.

• Journal of the Korean Society of Radiology
• /
• v.81 no.6
• /
• pp.1305-1333
• /
• 2020

Deep learning has recently achieved remarkable results in the field of medical imaging. However, as a deep learning network becomes deeper to improve its performance, it becomes more difficult to interpret the processes within. This can especially be a critical problem in medical fields where diagnostic decisions are directly related to a patient's survival. In order to solve this, explainable artificial intelligence techniques are being widely studied, and an attention mechanism was developed as part of this approach. In this paper, attention techniques are divided into two types: post hoc attention, which aims to analyze a network that has already been trained, and trainable attention, which further improves network performance. Detailed comparisons of each method, examples of applications in medical imaging, and future perspectives will be covered.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.24 no.3
• /
• pp.177-182
• /
• 2024

In this paper, the purpose of this paper is to predict and prevent the risk of crowd concentration in advance for possible future crowd accidents based on the Itaewon crush accident in Korea on October 29, 2022. In the case of a single CCTV, the administrator can determine the current situation in real time, but since the screen cannot be seen throughout the day, objects are detected using YOLOv4, which learns images taken with CCTV angle, and safety accidents due to crowd concentration are prevented by notification when the number of clusters exceeds. The reason for using the YOLO v4 model is that it improves with higher accuracy and faster speed than the previous YOLO model, making object detection techniques easier. This service will go through the process of testing with CCTV image data registered on the AI-Hub site. Currently, CCTVs have increased exponentially in Korea, and if they are applied to actual CCTVs, it is expected that various accidents, including accidents caused by crowd concentration in the future, can be prevented.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.24 no.3
• /
• pp.87-94
• /
• 2024

We investigated the feasibility of utilizing Large Language Models (LLMs) to perform text-based games without training on game data in advance. We adopted ChatGPT-3.5 and its state-of-the-art, ChatGPT-4, as the systems that implemented LLM. In addition, we added the persistent memory feature proposed in this paper to ChatGPT-4 to create three game player agents. We used Zork, one of the most famous text-based games, to see if the agents could navigate through complex locations, gather information, and solve puzzles. The results showed that the agent with persistent memory had the widest range of exploration and the best score among the three agents. However, all three agents were limited in solving puzzles, indicating that LLM is vulnerable to problems that require multi-level reasoning. Nevertheless, the proposed agent was still able to visit 37.3% of the total locations and collect all the items in the locations it visited, demonstrating the potential of LLM.

• Journal of the Korean Society of Marine Environment & Safety
• /
• v.30 no.1
• /
• pp.82-88
• /
• 2024

As the modern maritime industry rapidly progresses through technological advancements, data processing technology is emphasized as a key driver of this development. Natural language processing is a technology that enables machines to understand and process human language. Through this methodology, we aim to develop a model that predicts the proportions of outcomes when entering new written judgments by analyzing the rulings of the Marine Safety Tribunal and learning the cause-providing ratios of previously adjudicated ship collisions. The model calculated the cause-providing ratios of the accident using the navigation applied at the time of the accident and the weight of key keywords that affect the cause-providing ratios. Through this, the accuracy of the developed model could be analyzed, the practical applicability of the model could be reviewed, and it could be used to prevent the recurrence of collisions and resolve disputes between parties involved in marine accidents.