• ETRI Journal
• /
• v.46 no.1
• /
• pp.82-95
• /
• 2024

As automated essay scoring (AES) has progressed from handcrafted techniques to deep learning, holistic scoring capabilities have merged. However, specific trait assessment remains a challenge because of the limited depth of earlier methods in modeling dual assessments for holistic and multi-trait tasks. To overcome this challenge, we explore providing comprehensive feedback while modeling the interconnections between holistic and trait representations. We introduce the DualBERT-Trans-CNN model, which combines transformer-based representations with a novel dual-scale bidirectional encoder representations from transformers (BERT) encoding approach at the document-level. By explicitly leveraging multi-trait representations in a multi-task learning (MTL) framework, our DualBERT-Trans-CNN emphasizes the interrelation between holistic and trait-based score predictions, aiming for improved accuracy. For validation, we conducted extensive tests on the ASAP++ and TOEFL11 datasets. Against models of the same MTL setting, ours showed a 2.0% increase in its holistic score. Additionally, compared with single-task learning (STL) models, ours demonstrated a 3.6% enhancement in average multi-trait performance on the ASAP++ dataset.

• Nuclear Engineering and Technology
• /
• v.55 no.10
• /
• pp.3907-3912
• /
• 2023

Radiation portal monitors (RPMs) installed at airports and harbors to prevent illicit trafficking of radioactive materials generally use large plastic scintillators. However, their energy resolution is poor and radionuclide identification is nearly unfeasible. In this study, to improve isotope identification, a RPM system based on a multi-array plastic scintillator and convolutional neural network (CNN) was evaluated by measuring the spectra of radioactive sources. A multi-array plastic scintillator comprising an assembly of 14 hexagonal scintillators was fabricated within an area of 50 × 100 cm². The energy spectra of ¹³⁷Cs, ⁶⁰Co, ²²⁶Ra, and ⁴K (KCl) were measured at speeds of 10-30 km/h, respectively, and an energy-weighted algorithm was applied. For the CNN, 700 and 300 spectral images were used as training and testing images, respectively. Compared to the conventional plastic scintillator, the multi-arrayed detector showed a high collection probability of the optical photons generated inside. A Compton maximum peak was observed for four moving radiation sources, and the CNN-based classification results showed that at least 70% was discriminated. Under the speed condition, the spectral fluctuations were higher than those under dwelling condition. However, the machine learning results demonstrated that a considerably high level of nuclide discrimination was possible under source movement conditions.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.30 no.4
• /
• pp.527-536
• /
• 2020

It has been found that an attacker can extract the secret key embedded in a security device and recover the operation instruction using power consumption traces which are some kind of side channel information. Many profiling-based side channel attacks based on a deep learning model such as MLP(Multi-Layer Perceptron) method are recently researched. In this paper, we implemented a disassembler for operation instruction set used in the micro-controller AVR XMEGA128-D4. After measuring the template traces on each instruction, we automatically made the pre-processing process and classified the operation instruction set using a deep learning model CNN. As an experimental result, we showed that all instructions are classified with 87.5% accuracy and some core instructions used frequently in device operation are with 99.6% respectively.

• Journal of IKEEE
• /
• v.28 no.2
• /
• pp.234-238
• /
• 2024

This paper presents a study on the integration of 3D-stacked dual-tip RRAM with a CNN accelerator architecture, leveraging its low drive current characteristics and scalability in a 3D stacked configuration. The dual-tip structure is utilized in a parallel connection format in a synaptic array to implement multi-level capabilities. It is configured within a Network-on-chip style accelerator along with various hardware blocks such as DAC, ADC, buffers, registers, and shift & add circuits, and simulations were performed for the CNN accelerator. The quantization of synaptic weights and activation functions was assumed to be 16-bit. Simulation results of CNN operations through a parallel pipeline for this accelerator architecture achieved an operational efficiency of approximately 370 GOPs/W, with accuracy degradation due to quantization kept within 3%.

• Journal of the Korea Computer Graphics Society
• /
• v.23 no.3
• /
• pp.95-103
• /
• 2017

Horizon correction is a crucial stage for image composition enhancement. In this paper, we propose a deep learning based method for estimating the slanted angle of a photograph and correcting it. To estimate and correct the horizon direction, existing methods use hand-crafted low-level features such as lines, planes, and gradient distributions. However, these methods may not work well on the images that contain no lines or planes. To tackle this limitation and robustly estimate the slanted angle, we propose a convolutional neural network (CNN) based method to estimate the slanted angle by learning more generic features using a huge dataset. In addition, we utilize multiple adaptive spatial pooling layers to extract multi-scale image features for better performance. In the experimental results, we show our CNN-based approach robustly and accurately estimates the slanted angle of an image regardless of the image content, even if the image contains no lines or planes at all.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.11a
• /
• pp.347-350
• /
• 2020

본 논문에서는 영상 변환 기술인 이산웨이블릿변환(Discrete Wavelet Transform, DWT)를 딥러닝 기반의 네트워크로 구현한다. 딥러닝 기술 중에도 CNN 기반으로 네트워크를 설계하였으며, 본 DWT 네트워크는 해상도에 의존적이지 않은 계층들로만 구성된다. 데이터세트를 구성할 때 파이썬의 라이브러리를 사용하여 레이블 데이터세트를 구성한다. 128×128크기의 gray-scale 영상을 입력으로 사용하고 이에 대응하는 레이블 데이터세트를 구성하여 1-level DWT를 수행하는 네트워크의 학습을 진행한다. 역방향 변환도 네트워크 설계 후 데이터세트를 구성하여 학습을 진행한다. 학습이 완료된 1-level DWT 네트워크를 반복적으로 사용하여 Multi-level DWT 네트워크를 구성한다. 또한 양자화에 의한 간단한 영상압축 실험을 진행하여 DWT 네트워크의 성능과 압축 등의 응용분야에 활용할 수 있음을 보인다. 설계한 DWT 네트워크의 1-level 순방향 변환 성능은 42.18dB의 PSNR을 보였고, 1-level 역방향 변환 성능은 50.13dB의 PSNR을 보였다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.8
• /
• pp.3917-3941
• /
• 2019

The recent interest in artificial intelligence and machine learning has partly contributed to an interest in the use of such approaches for hyperspectral remote sensing (HRS) imagery classification, as evidenced by the increasing number of deep framework with deep convolutional neural networks (CNN) structures proposed in the literature. In these approaches, the assumption of obtaining high quality deep features by using CNN is not always easy and efficient because of the complex data distribution and the limited sample size. In this paper, conventional handcrafted learning-based multi features based on expert knowledge are introduced as the input of a special designed CNN to improve the pixel description and classification performance of HRS imagery. The introduction of these handcrafted features can reduce the complexity of the original HRS data and reduce the sample requirements by eliminating redundant information and improving the starting point of deep feature training. It also provides some concise and effective features that are not readily available from direct training with CNN. Evaluations using three public HRS datasets demonstrate the utility of our proposed method in HRS classification.

• ETRI Journal
• /
• v.41 no.6
• /
• pp.750-759
• /
• 2019

Presently, the extraction of hand-crafted features is still the dominant method in radar emitter recognition. To solve the complicated problems of selection and updation of empirical features, we present a novel automatic feature extraction structure based on deep learning. In particular, a convolutional neural network (CNN) is adopted to extract high-level abstract representations from the time-frequency images of emitter signals. Thus, the redundant process of designing discriminative features can be avoided. Furthermore, to address the performance degradation of a single platform, we propose the construction of an ensemble learning-based architecture for multi-platform fusion recognition. Experimental results indicate that the proposed algorithms are feasible and effective, and they outperform other typical feature extraction and fusion recognition methods in terms of accuracy. Moreover, the proposed structure could be extended to other prevalent ensemble learning alternatives.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.40 no.2
• /
• pp.91-108
• /
• 2022

In most cases, optical images have been used as training data of DL (Deep Learning) models for object detection, recognition, identification, classification, semantic segmentation, and instance segmentation. However, properties of 3D objects in the real-world could not be fully explored with 2D images. One of the major sources of the 3D geospatial information is DSM (Digital Surface Model). In this matter, characteristic information derived from DSM would be effective to analyze 3D terrain features. Especially, man-made objects such as buildings having geometrically unique shape could be described by geometric elements that are obtained from 3D geospatial data. The background and motivation of this paper were drawn from concept of the intrinsic image that is involved in high-level visual information processing. This paper aims to extract buildings after classifying terrain features by training DL model with DSM-derived information including slope, aspect, and SRI (Shaded Relief Image). The experiments were carried out using DSM and label dataset provided by ISPRS (International Society for Photogrammetry and Remote Sensing) for CNN-based SegNet model. In particular, experiments focus on combining multi-source information to improve training performance and synergistic effect of the DL model. The results demonstrate that buildings were effectively classified and extracted by the proposed approach.

• Journal of Internet Computing and Services
• /
• v.21 no.4
• /
• pp.117-126
• /
• 2020

A maritime object detection system is an intelligent assistance system to maritime autonomous surface ship(MASS). It detects automatically floating debris, which has a clash risk with objects in the surrounding water and used to be checked by a captain with a naked eye, at a similar level of accuracy to the human check method. It is used to detect objects around a ship. In the past, they were detected with information gathered from radars or sonar devices. With the development of artificial intelligence technology, intelligent CCTV installed in a ship are used to detect various types of floating debris on the course of sailing. If the speed of processing video data slows down due to the various requirements and complexity of MASS, however, there is no guarantee for safety as well as smooth service support. Trying to solve this issue, this study conducted research on the minimization of computation volumes for video data and the increased speed of data processing to detect maritime objects. Unlike previous studies that used the Hough transform algorithm to find the horizon and secure the areas of interest for the concerned objects, the present study proposed a new method of optimizing a binarization algorithm and finding areas whose locations were similar to actual objects in order to improve the speed. A maritime object detection system was materialized based on deep learning CNN to demonstrate the usefulness of the proposed method and assess the performance of the algorithm. The proposed algorithm performed at a speed that was 4 times faster than the old method while keeping the detection accuracy of the old method.