|
Design and development of an in-pipe mobile robot for pipeline inspection with AI defect detection system |
|
Azamat Nurlanovich Yeshmukhametov, Azamat Kenzhekhan, Akbayan Bakytzhannova, Sultan Omirbayev(Nazarbayev University, Kazakhstan), Yelnur Tuiubayev, Madiyar Daniyalov(Satbayev University, Kazakhstan) |
| Design and development of an in-pipe mobile robot for pipeline inspection with AI defect detection system
|
|
|
Attak Detection in RSU-OBU Communication using Deep Neural Network |
|
Youngwoo An, Yongsoon Eun(DGIST, Korea) |
| This paper proposes three types of neural network-based attack detectors for Internet of Vehicles (IoV) networks. The proposed attack detectors consist of Long Short-Term Memory (LSTM) layers. The fault detector is implemented in the Road Side Unit (RSU) that communicates with On Board Unit (OBU) installed in the autonomous vehicles.
We consider the multiplicative attacks in the RSU-OBU communication messages. The training and validation data sets
are generated by using an automated driving toolbox in MATLAB. Attack detector performance is evaluated using the
validation data set that is not included in the training data sets. Performance comparison of the three detectors are given. |
|
|
PIGD-TL: Physics-informed Generative Dynamics with Transfer Learning |
|
Minseok Jang, Jeongseok Hyun, Tuan Anh Nguyen, Jae-Woo Lee(Konkuk University, Korea) |
| Our research explores the integration of Physics-Informed Machine Learning (PIML) and Generative Adversarial Networks (GANs) to improve flight dynamics models for Electric Vertical Takeoff and Landing (eVTOL) aircraft. We highlight the impact of transfer learning, which leverages pre-existing models to enhance efficiency and reduce computational costs. This method proves especially valuable in PIML applications, enabling versatile model deployment across similar systems. Our work introduces groundbreaking models that generate physically realistic scenarios, thus elevating the reliability and performance of digital twins in predicting system dynamics. |
|
|
Comparative Performance of CNN and Transformer Architectures in Infrared Small Target Detection |
|
Lammi Choi(Seoul National University, Korea), Won Young Chung, In Ho Lee, Chan Gook Park(Seoul national university, Korea) |
| IR small target detection is tough due to low resolution, noise, and target-background similarity. Deep learning, especially CNNs and Transformers, enhances IR detection. This study evaluates Swin UNet, a Transformer-based model, for IR detection. Comparative simulations with CNNs (UNet, UNet++, UNet 3+) show comparable UNet 3+ and Swin UNet performance. This suggests Transformer's parity with CNNs, even with limited data. |
|
|
Exploring Object Detection Techniques in SAR Imagery:Addressing Limited Datasets and Leveraging Data Augmentation |
|
Haekang Song, Sungho Kim(Yeungnam University, Korea) |
| This paper explores the application of deep learning models to synthetic aperture radar (SAR) data for object detection. The challenges posed by limited benchmark datasets and the unique characteristics of SAR data are being discussed. Data augmentation techniques are examined as a solution to address the scarcity of labeled SAR data, along with potential pitfalls. In particular, leveraging diffusion models for SAR data augmentation is proposed, which can capture complex data distributions and generate diverse images. By enhancing the performance of deep learning-based object detectors, this research aims to advance SAR image analysis and contribute to the field of SAR-based object detection |
|
|
Cluster-based Object Detection System with Scalable Performance for Autonomous Driving |
|
Hongsuk Kim, Yongseong Lee(Kookmin University, Korea), Jangho Shin(Hyundai Motor Company, Korea), Jong-Chan Kim(Kookmin University, Korea) |
| Due to the unprecedented computing requirement of autonomous driving applications, in-vehicle computing architecture is going under tremendous changes. However, since the computing power of a single node is inherently limited, we propose to employ the cluster-based computing architecture. As an initial effort, we develop a prototype object detection system with three computing nodes and a GigE camera. For each node to select its assigned images in a round-robin manner, a distributed consensus scheme is proposed, which demonstrates a linear scalability of its frame rate with no additional delays. |
|