• Journal of Information Processing Systems
• /
• v.19 no.5
• /
• pp.673-687
• /
• 2023

Convolutional neural network-based deep learning technology is the most commonly used in image identification, but it requires large-scale data for training. Therefore, application in specific fields in which data acquisition is limited, such as in the military, may be challenging. In particular, the identification of ground weapon systems is a very important mission, and high identification accuracy is required. Accordingly, various studies have been conducted to achieve high performance using small-scale data. Among them, the ensemble method, which achieves excellent performance through the prediction average of the pre-trained models, is the most representative method; however, it requires considerable time and effort to find the optimal combination of ensemble models. In addition, there is a performance limitation in the prediction results obtained by using an ensemble method. Furthermore, it is difficult to obtain the ensemble effect using models with imbalanced classification accuracies. In this paper, we propose a transfer learning-based feature fusion technique for heterogeneous models that extracts and fuses features of pre-trained heterogeneous models and finally, fine-tunes hyperparameters of the fully connected layer to improve the classification accuracy. The experimental results of this study indicate that it is possible to overcome the limitations of the existing ensemble methods by improving the classification accuracy through feature fusion between heterogeneous models based on transfer learning.

• Imaging Science in Dentistry
• /
• v.52 no.2
• /
• pp.219-224
• /
• 2022

Purpose: This study aimed to evaluate the performance of transfer learning in a deep convolutional neural network for classifying implant fixtures. Materials and Methods: Periapical radiographs of implant fixtures obtained using the Superline (Dentium Co. Ltd., Seoul, Korea), TS III(Osstem Implant Co. Ltd., Seoul, Korea), and Bone Level Implant(Institut Straumann AG, Basel, Switzerland) systems were selected from patients who underwent dental implant treatment. All 355 implant fixtures comprised the total dataset and were annotated with the name of the system. The total dataset was split into a training dataset and a test dataset at a ratio of 8 to 2, respectively. YOLOv3 (You Only Look Once version 3, available at https://pjreddie.com/darknet/yolo/), a deep convolutional neural network that has been pretrained with a large image dataset of objects, was used to train the model to classify fixtures in periapical images, in a process called transfer learning. This network was trained with the training dataset for 100, 200, and 300 epochs. Using the test dataset, the performance of the network was evaluated in terms of sensitivity, specificity, and accuracy. Results: When YOLOv3 was trained for 200 epochs, the sensitivity, specificity, accuracy, and confidence score were the highest for all systems, with overall results of 94.4%, 97.9%, 96.7%, and 0.75, respectively. The network showed the best performance in classifying Bone Level Implant fixtures, with 100.0% sensitivity, specificity, and accuracy. Conclusion: Through transfer learning, high performance could be achieved with YOLOv3, even using a small amount of data.

• Korean Journal of Remote Sensing
• /
• v.39 no.5_4
• /
• pp.1111-1123
• /
• 2023

In recent times, semantic image segmentation methods using deep learning models have been widely used for monitoring changes in surface attributes using remote sensing imagery. To enhance the performance of various UNet-based deep learning models, including the prominent UNet model, it is imperative to have a sufficiently large training dataset. However, enlarging the training dataset not only escalates the hardware requirements for processing but also significantly increases the time required for training. To address these issues, transfer learning is used as an effective approach, enabling performance improvement of models even in the absence of massive training datasets. In this paper we present three transfer learning models, UNet-ResNet50, UNet-VGG19, and CBAM-DRUNet-VGG19, which are combined with the representative pretrained models of VGG19 model and ResNet50 model. We applied these models to building extraction tasks and analyzed the accuracy improvements resulting from the application of transfer learning. Considering the substantial impact of learning rate on the performance of deep learning models, we also analyzed performance variations of each model based on different learning rate settings. We employed three datasets, namely Kompsat-3A dataset, WHU dataset, and INRIA dataset for evaluating the performance of building extraction results. The average accuracy improvements for the three dataset types, in comparison to the UNet model, were 5.1% for the UNet-ResNet50 model, while both UNet-VGG19 and CBAM-DRUNet-VGG19 models achieved a 7.2% improvement.

• Geomechanics and Engineering
• /
• v.38 no.1
• /
• pp.29-44
• /
• 2024

This research investigates the effectiveness of data augmentation techniques in the automated analysis of B-scan images from ground-penetrating radar (GPR) using deep learning. In spite of the growing interest in automating GPR data analysis and advancements in deep learning for image classification and object detection, many deep learning-based GPR data analysis studies have been limited by the availability of large, diverse GPR datasets. Data augmentation techniques are widely used in deep learning to improve model performance. In this study, we applied four data augmentation techniques (geometric transformation, color-space transformation, noise injection, and applying kernel filter) to the GPR datasets obtained from a testbed. A deep learning model for GPR data analysis was developed using three models (Faster R-CNN ResNet, SSD ResNet, and EfficientDet) based on transfer learning. It was found that data augmentation significantly enhances model performance across all cases, with the mAP and AR for the Faster R-CNN ResNet model increasing by approximately 4%, achieving a maximum mAP (Intersection over Union = 0.5:1.0) of 87.5% and maximum AR of 90.5%. These results highlight the importance of data augmentation in improving the robustness and accuracy of deep learning models for GPR B-scan analysis. The enhanced detection capabilities achieved through these techniques contribute to more reliable subsurface investigations in geotechnical engineering.

• Journal of the Korea Institute of Military Science and Technology
• /
• v.27 no.4
• /
• pp.474-484
• /
• 2024

Reinforcement learning, which are also studied in the field of defense, face the problem of sample efficiency, which requires a large amount of data to train. Transfer learning has been introduced to address this problem, but its effectiveness is sometimes marginal because the model does not effectively leverage prior knowledge. In this study, we propose a stochastic initial state randomization(SISR) method to enable robust knowledge transfer that promote generalized and sufficient knowledge transfer. We developed a simulation environment involving a cooperative robot transportation task. Experimental results show that successful tasks are achieved when SISR is applied, while tasks fail when SISR is not applied. We also analyzed how the amount of state information collected by the agents changes with the application of SISR.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.3
• /
• pp.456-462
• /
• 2022

In this paper, we introduce a method of predicting time offset by applying LSTM, a deep learning model, to a precision time comparison technique based on measurement data extracted from code signals transmitted from GPS satellites to determine Universal Coordinated Time (UTC). First, we introduce a process of extracting time information from code signals received from a GPS satellite on a daily basis and constructing a daily time offset into one time series data. To apply the deep learning model to the constructed time offset time series data, LSTM, one of the recurrent neural networks, was applied to predict the time offset of a GPS satellite. Through this study, the possibility of time offset prediction by applying deep learning in the field of GNSS precise time transfer was confirmed.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.37 no.3
• /
• pp.119-127
• /
• 2019

In this study, we investigated the application of deep learning to the manufacturing process of traffic and road signs which are constituting the road layer in map production with 1 / 1,000 digital topographic map. Automated classification of road traffic sign images was carried out through construction of training data for images acquired by using transfer learning which is used in image classification of deep learning. As a result of the analysis, the signs of attention, regulation, direction and assistance were irregular due to various factors such as the quality of the photographed images and sign shape, but in the case of the guide sign, the accuracy was higher than 97%. In the digital mapping, it is expected that the automatic image classification method using transfer learning will increase the utilization in data acquisition and classification of various layers including traffic safety signs.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.4
• /
• pp.21-28
• /
• 2021

This study proposes a deep learning-based flat foot classification methodology using transfer learning. We used a transfer learning with VGG16 pre-trained model and a data augmentation technique to generate a model with high predictive accuracy from a total of 176 image data consisting of 88 flat feet and 88 normal feet. To evaluate the performance of the proposed model, we performed an experiment comparing the prediction accuracy of the basic CNN-based model and the prediction model derived through the proposed methodology. In the case of the basic CNN model, the training accuracy was 77.27%, the validation accuracy was 61.36%, and the test accuracy was 59.09%. Meanwhile, in the case of our proposed model, the training accuracy was 94.32%, the validation accuracy was 86.36%, and the test accuracy was 84.09%, indicating that the accuracy of our model was significantly higher than that of the basic CNN model.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.18 no.2
• /
• pp.329-335
• /
• 2023

Avocado, a superfood selected by Time magazine and one of the late ripening fruits, is one of the foods with a big difference between local prices and domestic distribution prices. If this sorting process of avocados is automated, it will be possible to lower prices by reducing labor costs in various fields. In this paper, we aim to create an optimal classification model by creating an avocado dataset through crawling and using a number of deep learning-based transfer learning models. Experiments were conducted by directly substituting a deep learning-based transfer learning model from a dataset separated from the produced dataset and fine-tuning the hyperparameters of the model. When an avocado image is input, the model classifies the ripeness of the avocado with an accuracy of over 99%, and proposes a dataset and algorithm that can reduce manpower and increase accuracy in avocado production and distribution households.

• Journal of Broadcast Engineering
• /
• v.25 no.6
• /
• pp.980-993
• /
• 2020

Deep-learning based style transfer has recently attracted great attention, because it provides high quality transfer results by appropriately reflecting the high level structural characteristics of images. This paper deals with the problem of providing more stable and more diverse style transfer results of such deep-learning based style transfer method. Based on the investigation of the experimental results from the wide range of hyper-parameter settings, this paper defines the problem of the stability and the diversity of the style transfer, and proposes a partial loss normalization method to solve the problem. The style transfer using the proposed normalization method not only gives the stability on the control of the degree of style reflection, regardless of the input image characteristics, but also presents the diversity of style transfer results, unlike the existing method, at controlling the weight of the partial style loss, and provides the stability on the difference in resolution of the input image.