Sungho Kim
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Sungho Kim has a total of 13 co-authors. Here are the top co-authors by number of shared publications:
- Su-Jin Shin
AI Autonomy Technology Center Agency for Defense Development Daejeon 34186 Republic of Korea
2 shared publications - Youngjung Kim
Agency for Defense Development, Daejeon, South Korea
2 shared publications - Chang Ouk Kim
Yonsei University
1 shared publication - Dong-Yeon Kim
Agency for Defense Development, Yuseong P.O. Box 35, Daejeon 34186, Republic of Korea
1 shared publication - Ha Jun Hwang
Department of Industrial Engineering, Yonsei University, Seoul 03722, Republic of Korea
1 shared publication - In-Su Hwang
Agency for Defense Development, Institute of Defense Advanced Technology Research, Daejeon 34186, Korea
1 shared publication
Latest Publications
- Open Access
Stepwise Soft Actor–Critic for UAV Autonomous Flight Control
Despite the growing demand for unmanned aerial vehicles (UAVs), the use of conventional UAVs is limited, as most of them require being remotely operated by a person who is not within the vehicle’s field of view. Recently, many studies have introduced reinforcement learning (RL) to address hurdles for the autonomous flight of UAVs. However, most previous studies have assumed overly simplified environments, and thus, they cannot be applied to real-world UAV operation scenarios. To address the limitations of previous studies, we propose a stepwise soft actor–critic (SeSAC) algorithm for efficient learning in a continuous state and action space environment. SeSAC aims to overcome the inefficiency of learning caused by attempting challenging tasks from the beginning. Instead, it starts with easier missions and gradually increases the difficulty level during training, ultimately achieving the final goal. We also control a learning hyperparameter of the soft actor–critic algorithm and implement a positive buffer mechanism during training to enhance learning effectiveness. Our proposed algorithm was verified in a six-degree-of-freedom (DOF) flight environment with high-dimensional state and action spaces. The experimental results demonstrate that the proposed algorithm successfully completed missions in two challenging scenarios, one for disaster management and another for counter-terrorism missions, while surpassing the performance of other baseline approaches.
Open AccessStepwise Soft Actor–Critic for UAV Autonomous Flight Control
Despite the growing demand for unmanned aerial vehicles (UAVs), the use of conventional UAVs is limited, as most of them require being remotely operated by a person who is not within the vehicle’s field of view. Recently, many studies have introduced reinforcement learning (RL) to address hurdles for the autonomous flight of UAVs. However, most previous studies have assumed overly simplified environments, and thus, they cannot be applied to real-world UAV operation scenarios. To address the limitations of previous studies, we propose a stepwise soft actor–critic (SeSAC) algorithm for efficient learning in a continuous state and action space environment. SeSAC aims to overcome the inefficiency of learning caused by attempting challenging tasks from the beginning. Instead, it starts with easier missions and gradually increases the difficulty level during training, ultimately achieving the final goal. We also control a learning hyperparameter of the soft actor–critic algorithm and implement a positive buffer mechanism during training to enhance learning effectiveness. Our proposed algorithm was verified in a six-degree-of-freedom (DOF) flight environment with high-dimensional state and action spaces. The experimental results demonstrate that the proposed algorithm successfully completed missions in two challenging scenarios, one for disaster management and another for counter-terrorism missions, while surpassing the performance of other baseline approaches.
- Open Access
GPS Spoofing Detection Method for Small UAVs Using 1D Convolution Neural Network
The navigation of small unmanned aerial vehicles (UAVs), such as quadcopters, significantly relies on the global positioning system (GPS); however, UAVs are vulnerable to GPS spoofing attacks. GPS spoofing is an attempt to manipulate a GPS receiver by broadcasting manipulated signals. A commercial GPS simulator can cause a GPS-guided drone to deviate from its intended course by transmitting counterfeit GPS signals. Therefore, an anti-spoofing technique is essential to ensure the operational safety of UAVs. Various methods have been introduced to detect GPS spoofing; however, most methods require additional hardware. This may not be appropriate for small UAVs with limited capacity. This study proposes a deep learning-based anti-spoofing method equipped with 1D convolutional neural network. The proposed method is lightweight and power-efficient, enabling real-time detection on mobile platforms. Furthermore, the performance of our approach can be enhanced by increasing training data and adjusting the network architecture. We evaluated our algorithm on the embedded board of a drone in terms of power consumption and inference time. Compared to the support vector machine, the proposed method showed better performance in terms of precision, recall, and F-1 score. Flight test demonstrated our algorithm could successfully detect GPS spoofing attacks.
Open AccessGPS Spoofing Detection Method for Small UAVs Using 1D Convolution Neural Network
The navigation of small unmanned aerial vehicles (UAVs), such as quadcopters, significantly relies on the global positioning system (GPS); however, UAVs are vulnerable to GPS spoofing attacks. GPS spoofing is an attempt to manipulate a GPS receiver by broadcasting manipulated signals. A commercial GPS simulator can cause a GPS-guided drone to deviate from its intended course by transmitting counterfeit GPS signals. Therefore, an anti-spoofing technique is essential to ensure the operational safety of UAVs. Various methods have been introduced to detect GPS spoofing; however, most methods require additional hardware. This may not be appropriate for small UAVs with limited capacity. This study proposes a deep learning-based anti-spoofing method equipped with 1D convolutional neural network. The proposed method is lightweight and power-efficient, enabling real-time detection on mobile platforms. Furthermore, the performance of our approach can be enhanced by increasing training data and adjusting the network architecture. We evaluated our algorithm on the embedded board of a drone in terms of power consumption and inference time. Compared to the support vector machine, the proposed method showed better performance in terms of precision, recall, and F-1 score. Flight test demonstrated our algorithm could successfully detect GPS spoofing attacks.
- Open Access
- 16 June 2021
- Published by MDPI AG in Applied Sciences
- Vol. 11 (12) , 5569
Coupling Denoising to Detection for SAR Imagery
Detecting objects in synthetic aperture radar (SAR) imagery has received much attention in recent years since SAR can operate in all-weather and day-and-night conditions. Due to the prosperity and development of convolutional neural networks (CNNs), many previous methodologies have been proposed for SAR object detection. In spite of the advance, existing detection networks still have limitations in boosting detection performance because of inherently noisy characteristics in SAR imagery; hence, separate preprocessing step such as denoising (despeckling) is required before utilizing the SAR images for deep learning. However, inappropriate denoising techniques might cause detailed information loss and even proper denoising methods does not always guarantee performance improvement. In this paper, we therefore propose a novel object detection framework that combines unsupervised denoising network into traditional two-stage detection network and leverages a strategy for fusing region proposals extracted from both raw SAR image and synthetically denoised SAR image. Extensive experiments validate the effectiveness of our framework on our own object detection datasets constructed with remote sensing images from TerraSAR-X and COSMO-SkyMed satellites. Extensive experiments validate the effectiveness of our framework on our own object detection datasets constructed with remote sensing images from TerraSAR-X and COSMO-SkyMed satellites. The proposed framework shows better performances when we compared the model with using only noisy SAR images and only denoised SAR images after despeckling under multiple backbone networks.
- 16 June 2021
- Published by MDPI AG in Applied Sciences
- Vol. 11 (12) , 5569
Open AccessCoupling Denoising to Detection for SAR Imagery
Detecting objects in synthetic aperture radar (SAR) imagery has received much attention in recent years since SAR can operate in all-weather and day-and-night conditions. Due to the prosperity and development of convolutional neural networks (CNNs), many previous methodologies have been proposed for SAR object detection. In spite of the advance, existing detection networks still have limitations in boosting detection performance because of inherently noisy characteristics in SAR imagery; hence, separate preprocessing step such as denoising (despeckling) is required before utilizing the SAR images for deep learning. However, inappropriate denoising techniques might cause detailed information loss and even proper denoising methods does not always guarantee performance improvement. In this paper, we therefore propose a novel object detection framework that combines unsupervised denoising network into traditional two-stage detection network and leverages a strategy for fusing region proposals extracted from both raw SAR image and synthetically denoised SAR image. Extensive experiments validate the effectiveness of our framework on our own object detection datasets constructed with remote sensing images from TerraSAR-X and COSMO-SkyMed satellites. Extensive experiments validate the effectiveness of our framework on our own object detection datasets constructed with remote sensing images from TerraSAR-X and COSMO-SkyMed satellites. The proposed framework shows better performances when we compared the model with using only noisy SAR images and only denoised SAR images after despeckling under multiple backbone networks.
- Open Access
- 24 August 2020
- Published by MDPI AG in Remote Sensing
- Vol. 12 (17) , 2734
Hierarchical Multi-Label Object Detection Framework for Remote Sensing Images
Detecting objects such as aircraft and ships is a fundamental research area in remote sensing analytics. Owing to the prosperity and development of CNNs, many previous methodologies have been proposed for object detection within remote sensing images. Despite the advance, using the object detection datasets with a more complex structure, i.e., datasets with hierarchically multi-labeled objects, is limited to the existing detection models. Especially in remote sensing images, since objects are obtained from bird’s-eye view, the objects are captured with restricted visual features and not always guaranteed to be labeled up to fine categories. We propose a hierarchical multi-label object detection framework applicable to hierarchically partial-annotated datasets. In the framework, an object detection pipeline called Decoupled Hierarchical Classification Refinement (DHCR) fuses the results of two networks: (1) an object detection network with multiple classifiers, and (2) a hierarchical sibling classification network for supporting hierarchical multi-label classification. Our framework additionally introduces a region proposal method for efficient detection on vain areas of the remote sensing images, called clustering-guided cropping strategy. Thorough experiments validate the effectiveness of our framework on our own object detection datasets constructed with remote sensing images from WorldView-3 and SkySat satellites. Under our proposed framework, DHCR-based detections significantly improve the performance of respective baseline models and we achieve state-of-the-art results on the datasets.
- 24 August 2020
- Published by MDPI AG in Remote Sensing
- Vol. 12 (17) , 2734
Open AccessHierarchical Multi-Label Object Detection Framework for Remote Sensing Images
Detecting objects such as aircraft and ships is a fundamental research area in remote sensing analytics. Owing to the prosperity and development of CNNs, many previous methodologies have been proposed for object detection within remote sensing images. Despite the advance, using the object detection datasets with a more complex structure, i.e., datasets with hierarchically multi-labeled objects, is limited to the existing detection models. Especially in remote sensing images, since objects are obtained from bird’s-eye view, the objects are captured with restricted visual features and not always guaranteed to be labeled up to fine categories. We propose a hierarchical multi-label object detection framework applicable to hierarchically partial-annotated datasets. In the framework, an object detection pipeline called Decoupled Hierarchical Classification Refinement (DHCR) fuses the results of two networks: (1) an object detection network with multiple classifiers, and (2) a hierarchical sibling classification network for supporting hierarchical multi-label classification. Our framework additionally introduces a region proposal method for efficient detection on vain areas of the remote sensing images, called clustering-guided cropping strategy. Thorough experiments validate the effectiveness of our framework on our own object detection datasets constructed with remote sensing images from WorldView-3 and SkySat satellites. Under our proposed framework, DHCR-based detections significantly improve the performance of respective baseline models and we achieve state-of-the-art results on the datasets.