• Korean Journal of Remote Sensing
• /
• v.38 no.6_1
• /
• pp.1407-1422
• /
• 2022

This study was conducted to classify tree species and assess the classification accuracy, using SE-Inception, a classification-based deep learning model. The input images of the dataset used Worldview-3 and GeoEye-1 images, and the size of the input images was divided into 10 × 10 m, 30 × 30 m, and 50 × 50 m to compare and evaluate the accuracy of classification of tree species. The label data was divided into five tree species (Pinus densiflora, Pinus koraiensis, Larix kaempferi, Abies holophylla Maxim. and Quercus) by visually interpreting the divided image, and then labeling was performed manually. The dataset constructed a total of 2,429 images, of which about 85% was used as learning data and about 15% as verification data. As a result of classification using the deep learning model, the overall accuracy of up to 78% was achieved when using the Worldview-3 image, the accuracy of up to 84% when using the GeoEye-1 image, and the classification accuracy was high performance. In particular, Quercus showed high accuracy of more than 85% in F₁ regardless of the input image size, but trees with similar spectral characteristics such as Pinus densiflora and Pinus koraiensis had many errors. Therefore, there may be limitations in extracting feature amount only with spectral information of satellite images, and classification accuracy may be improved by using images containing various pattern information such as vegetation index and Gray-Level Co-occurrence Matrix (GLCM).

• Ecology and Resilient Infrastructure
• /
• v.9 no.4
• /
• pp.218-227
• /
• 2022

This study attempted classification through deep learning-based image training for land cover classification in river spaces which is one of the important data for efficient river management. For this purpose, land cover classification analysis with the RGB image of the target section based on the category classification index of major land cover map was conducted by using the learning outcomes from the result of labeling. In addition, land cover classification of the river spaces was performed by unsupervised and supervised classification from Sentinel-2 satellite images provided in an open format, and this was compared with the results of deep learning-based image classification. As a result of the analysis, it showed more accurate prediction results compared to unsupervised classification results, and it presented significantly improved classification results in the case of high-resolution images. The result of this study showed the possibility of classifying water areas and wetlands in the river spaces, and if additional research is performed in the future, the deep learning based image train method for the land cover classification could be used for river management.

• Journal of the Korea Computer Graphics Society
• /
• v.26 no.4
• /
• pp.1-8
• /
• 2020

Motion synthesis using physics-based controllers can generate a character animation that interacts naturally with the given environment and other characters. Recently, various methods using deep neural networks have improved the quality of motions generated by physics-based controllers. In this paper, we present a control policy learned by deep reinforcement learning (DRL) that enables Luxo, the mascot character of Pixar animation studio, to run towards a random goal location while imitating a reference motion and maintaining its balance. Instead of directly training our DRL network to make Luxo reach a goal location, we use a reference motion that is generated to keep Luxo animation's jumping style. The reference motion is generated by linearly interpolating predetermined poses, which are defined with Luxo character's each joint angle. By applying our method, we could confirm a better Luxo policy compared to the one without any reference motions.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.6
• /
• pp.813-826
• /
• 2022

AI-based image processing technologies in various fields have been widely studied. However, the lighter the board, the more difficult it is to reduce the weight of image processing algorithm due to a lot of computation. In this paper, we propose a method using deep learning for object recognition algorithm in lightweight embedded boards. We can determine the area using a deep neural network architecture algorithm that processes semantic segmentation with a relatively small amount of computation. After masking the area, by using more accurate deep learning algorithm we could operate object detection with improved accuracy for efficient neural network (ENet) and You Only Look Once (YOLO) toward executing object recognition in real time for lightweighted embedded boards. This research is expected to be used for autonomous driving applications, which have to be much lighter and cheaper than the existing approaches used for object recognition.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.22 no.3
• /
• pp.418-424
• /
• 2018

Detection and classification of underwater targets is an important issue for both military and non-military purposes. Recently, many performance improvements are being reported in the field of pattern recognition with the development of deep learning technology. Among the results, DBN showed good performance when used for pre-training of DNN. In this paper, DBN was used for the classification of underwater targets using active sonar, and the results are compared with that of the conventional BPNN. We synthesized active sonar target signals using 3-dimensional highlight model. Then, features were extracted based on FrFT. In the single aspect based experiment, the classification result using DBN was improved about 3.83% compared with the BPNN. In the case of multi-aspect based experiment, a performance of 95% or more is obtained when the number of observation sequence exceeds three.

• Journal of Korean Tunnelling and Underground Space Association
• /
• v.23 no.4
• /
• pp.253-263
• /
• 2021

Concrete structures are damaged by aging and external environmental factors. This type of damage is to appear in the form of cracks, to proceed in the form of spalling. Such concrete damage can act as the main cause of reducing the original design bearing capacity of the structure, and negatively affect the stability of the structure. If such damage continues, it may lead to a safety accident in the future, thus proper repair and reinforcement are required. To this end, an accurate and objective condition inspection of the structure must be performed, and for this inspection, a sensor technology capable of detecting damage area is required. For this reason, we propose a deep learning-based image processing algorithm that can detect spalling. To develop this, 298 spalling images were obtained, of which 253 images were used for training, and the remaining 45 images were used for testing. In addition, an improved loss function and data augmentation technique were applied to improve the detection performance. As a result, the detection performance of concrete spalling showed a mean intersection over union of 80.19%. In conclusion, we developed an algorithm to detect concrete spalling through a deep learning-based image processing technique, with an improved loss function and data augmentation technique. This technology is expected to be utilized for accurate inspection and diagnosis of structures in the future.

• The Journal of the Convergence on Culture Technology
• /
• v.7 no.1
• /
• pp.640-645
• /
• 2021

The deep learning based end-to-end TTS system consists of Text2Mel module that generates spectrogram from text, and vocoder module that synthesizes speech signals from spectrogram. Recently, by applying deep learning technology to the TTS system the intelligibility and naturalness of the synthesized speech is as improved as human vocalization. However, it has the disadvantage that the inference speed for synthesizing speech is very slow compared to the conventional method. The inference speed can be improved by applying the non-autoregressive method which can generate speech samples in parallel independent of previously generated samples. In this paper, we introduce FastSpeech, FastSpeech 2, and FastPitch as Text2Mel technology, and Parallel WaveGAN, Multi-band MelGAN, and WaveGlow as vocoder technology applying non-autoregressive method. And we implement them to verify whether it can be processed in real time. Experimental results show that by the obtained RTF all the presented methods are sufficiently capable of real-time processing. And it can be seen that the size of the learned model is about tens to hundreds of megabytes except WaveGlow, and it can be applied to the embedded environment where the memory is limited.

• International Journal of Advanced Culture Technology
• /
• v.4 no.4
• /
• pp.70-74
• /
• 2016

Electronic commerce (E-commerce) using virtual reality (VR) has the advantage of being able to purchase products without restrictions of time and location by overcoming the limitations of existing offline transactions. It is still a rudimentary but fast growing technology, and the use of E-commerce in VR is expanding. The barriers that consumers might face in utilizing E-commerce in VR is the relevance to Information and Communications Technologies (ICT) technology. Fundamentally, it requires Internet access and use through PCs or mobile devices such as smart phones. Because unlike off-line markets, it is difficult to determine the purchase patterns of customers, customer purchasing behavior analysis must be done using computer access records. In order to expand and develop E-commerce in VR in the future, learning ability should be improved through combining with artificial neural network by deep learning that is recently in the spotlight, and the ability to overcome errors need to be improved to enable use in various fields.

• Journal of Korea Multimedia Society
• /
• v.23 no.5
• /
• pp.650-657
• /
• 2020

The rapid growth of internet users and faster network speed are driving the new ICT services. ICT Technology has improved our way of thinking and style of life, but it has created security problems such as malware, ransomware, and so on. Therefore, we should research against the increase of malware and the emergence of malicious code. For this, it is necessary to accurately and quickly detect and classify malware family. In this paper, we analyzed and classified visualization technology, which is a preprocessing technology used for deep learning-based malware classification. The first method is to convert each byte into one pixel of the image to produce a grayscale image. The second method is to convert 2bytes of the binary to create a pair of coordinates. The third method is the method using LSH. We proposed improving the technique of using the entire existing malicious code file for visualization, extracting only the areas where important information is expected to exist and then visualizing it. As a result of experimenting in the method we proposed, it shows that selecting and visualizing important information and then classifying it, rather than containing all the information in malicious code, can produce better learning results.

• Journal of Korea Society of Industrial Information Systems
• /
• v.25 no.2
• /
• pp.11-18
• /
• 2020

Recently, a series of studies on virtual try-on (VTON) using images have been published. A comparison study analyzed representative methods, SCMM-based non-deep learning method, deep learning based VITON and CP-VITON, using costumes and user images according to the posture and body type of the person, the degree of occlusion of the clothes, and the characteristics of the clothes. In this paper, we tackle the problems observed in the best performing CP-VTON. The issues tackled are the problem of segmentation of the subject, pixel generation of un-intended area, missing warped cloth mask and the cost function used in the learning, and limited the algorithm to improve it. The results show some improvement in SSIM, and significantly in subjective evaluation.