• Proceedings of the Korea Information Processing Society Conference
• /
• 2019.05a
• /
• pp.398-400
• /
• 2019

Faced with the increasing demand for image recognition on mobile devices, how to run convolutional neural network (CNN) models on mobile devices with limited computing power and limited storage resources encourages people to study efficient model design. In recent years, many effective architectures have been proposed, such as mobilenet_v1, mobilenet_v2 and PeleeNet. However, in the process of feature selection, all these models neglect some information of shallow features, which reduces the capture of shallow feature location and semantics. In this study, we propose an effective framework based on Feature Pyramid Networks to improve the recognition accuracy of deep and shallow images while guaranteeing the recognition speed of PeleeNet structured images. Compared with PeleeNet, the accuracy of structure recognition on CIFA-10 data set increased by 4.0%.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.1
• /
• pp.60-79
• /
• 2022

The U-Net architecture-based segmentation models attained remarkable performance in numerous medical image segmentation missions like skin lesion segmentation. Nevertheless, the resolution gradually decreases and the loss of spatial information increases with deeper network. The fusion of adjacent layers is not enough to make up for the lost spatial information, thus resulting in errors of segmentation boundary so as to decline the accuracy of segmentation. To tackle the issue, we propose a new deep learning-based segmentation model. In the decoding stage, the feature channels of each decoding unit are concatenated with all the feature channels of the upper coding unit. Which is done in order to ensure the segmentation effect by integrating spatial and semantic information, and promotes the robustness and generalization of our model by combining the atrous spatial pyramid pooling (ASPP) module and channel attention module (CAM). Extensive experiments on ISIC2016 and ISIC2017 common datasets proved that our model implements well and outperforms compared segmentation models for skin lesion segmentation.

• Smart Structures and Systems
• /
• v.32 no.2
• /
• pp.111-121
• /
• 2023

This study proposes an automated assembly performance evaluation method for prefabricated steel structures (PSSs) using machine learning methods. Assembly component images were segmented using a modified version of the receptive field pyramid. By factorizing channel modulation and the receptive field exploration layers of the convolution pyramid, highly accurate segmentation results were obtained. After completing segmentation, the positions of the bolt holes were calculated using various image processing techniques, such as fuzzy-based edge detection, Hough's line detection, and image perspective transformation. By calculating the distance ratio between bolt holes, the assembly performance of the PSS was estimated using the k-nearest neighbors (kNN) algorithm. The effectiveness of the proposed framework was validated using a 3D PSS printing model and a field test. The results indicated that this approach could recognize assembly components with an intersection over union (IoU) of 95% and evaluate assembly performance with an error of less than 5%.

• Journal of Institute of Control, Robotics and Systems
• /
• v.20 no.12
• /
• pp.1189-1195
• /
• 2014

This paper addresses a Moving Target Indication (MTI) algorithm which can be used for small Unmanned Aerial Vehicles (UAVs) equipped with image sensors. MTI is a system (or an algorithm) which detects moving objects. The principle of the MTI algorithm is to analyze the difference between successive image data. It is difficult to detect moving objects in the images recorded from dynamic cameras attached to moving platforms such as UAVs flying at low altitudes over a variety of terrain, since the acquired images have two motion components: 'camera motion' and 'object motion'. Therefore, the motion of independent objects can be obtained after the camera motion is compensated thoroughly via proper manipulations. In this study, the camera motion effects are removed by using wiener filter-based image registration, one of the non-parametric methods. In addition, an image pyramid structure is adopted to reduce the computational complexity for UAVs. We demonstrate the effectiveness of our method with experimental results on outdoor video sequences.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.15 no.5
• /
• pp.374-383
• /
• 1990

For a visual inspectio system to be viable, it must in general achieve the same or lower costs as those incurred by manual methods, it must have a reasonable inspection rate, and it must be reliable and maintainable : for the visual inspection system, inspection at high speed, flexibility for variety of products, and low cost are needed. In order to solve these problems, we propose the visual inspection algorithm by divide-and-conquer technique using the pyramid data structure and the development of simplified pattern matching method. By this method we can reduce the number of data required to compare with patterns and make the inspection time short.

• Journal of the Korean Society for Library and Information Science
• /
• v.45 no.4
• /
• pp.287-295
• /
• 2011

The Five Laws of Library Science was introduced into Japan in 1935. After World War II, it was widely disseminated among young librarians by the education for librarianship developed in the 1950's. Its concept, "Books for All," met a serious opposition from local authorities based on their long continued concept of "preservation." After they realized that people eagerly use local libraries, they interpreted that the duty of public library was "lending books." They introduced "Privatization" of libraries into quite a few local libraries in order to save money. For overcoming problems, it is important to disseminate the library concept among the public, the assembly members and government officials planning to change their attitude of disregarding libraries. As the fundamental library philosophy, the Five Laws of Library Science is important. In the commentary of the Five Laws written by the present writer, he included an illustration of triangle pyramid showing the structure of the Five Laws. He included here the spiral of study on the library and the library science, based on the Section 814 of the Five Laws. He welcomes discussions on his Japanese angle from different viewpoints in order to develop further understanding of the Five Laws.

• Proceedings of the Korean Vacuum Society Conference
• /
• 2012.08a
• /
• pp.184-184
• /
• 2012

Recently, InGaN/GaN multi-quantum well grown on GaN pyramid structures have attracted much attention due to their hybrid characteristics of quantum well, quantum wire, and quantum dot. This gives us broad band emission which will be useful for phosphor-free white light emitting diode. On the other hand, by using quantum dot emission on top of the pyramid, site selective single photon source could be realized. However, these structures still have several limitations for the single photon source. For instance, the quantum efficiency of quantum dot emission should be improved further. As detection systems have limited numerical aperture, collection efficiency is also important issue. It has been known that micro-cavities can be utilized to modify the radiative decay rate and to control the radiation pattern of quantum dot. Researchers have also been interested in nano-cavities using localized surface plasmon. Although the plasmonic cavities have small quality factor due to high loss of metal, it could have small mode volume because plasmonic wavelength is much smaller than the wavelength in the dielectric cavities. In this work, we used localized surface plasmon to improve efficiency of InGaN qunatum dot as a single photon emitter. We could easily get the localized surface plasmon mode after deposit the metal thin film because lnGaN/GaN multi quantum well has the pyramidal geometry. With numerical simulation (i.e., Finite Difference Time Domain method), we observed highly enhanced decay rate and modified radiation pattern. To confirm these localized surface plasmon effect experimentally, we deposited metal thin films on InGaN/GaN pyramid structures using e-beam deposition. Then, photoluminescence and time-resolved photoluminescence were carried out to measure the improvement of radiative decay rate (Purcell factor). By carrying out cathodoluminescence (CL) experiments, spatial-resolved CL images could also be obtained. As we mentioned before, collection efficiency is also important issue to make an efficient single photon emitter. To confirm the radiation pattern of quantum dot, Fourier optics system was used to capture the angular property of emission. We believe that highly focused localized surface plasmon around site-selective InGaN quantum dot could be a feasible single photon emitter.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2021.06a
• /
• pp.133-133
• /
• 2021

The performance of radar Quantitative Precipitation Estimation (QPE) using Long Short-Term Memory (LSTM) networks in hydrological applications depends on either the quality of data or the three-dimensional CAPPI structure from the weather radar. While radar data quality is controlled and enhanced by the more and more modern radar systems, the effect of CAPPI structure still has not yet fully investigated. In this study, three typical and important types of CAPPI structure including inverse-pyramid, cubic of grids 3x3, cubic of grids 4x4 are investigated to evaluate the effect of CAPPI structures on the performance of radar QPE using LSTM networks. The investigation results figure out that the cubic of grids 4x4 of CAPPI structure shows the best performance in rainfall estimation using the LSTM networks approach. This study give us the precious experiences in radar QPE works applying LSTM networks approach in particular and deep-learning approach in general.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.45 no.9
• /
• pp.807-815
• /
• 2017

Auxetic structures with negative Poisson's ratio can be used to achieve high mechanical properties in energy absorption and destruction toughness. In this paper, we aim to design an auxetic structure which has a high negative Poisson's ratio and a stiffness over 50N/mm by using an optimization method. Length(L), thickness(t) and angle(${\theta}_1$, ${\theta}_2$) of an auxetic pyramid are the design parameters, and also stress, Poisson's ratio and stiffness are thr reaction factors. We used Box-Behnken method and conducted 4 factors - 3 levels experiment design. Finite element models are analyzed by using Edison program CSD_EPLAST.

• Journal of Digital Convergence
• /
• v.19 no.5
• /
• pp.307-313
• /
• 2021

In this paper, a lightweight network with fewer parameters compared to the existing object detection method is proposed. In the case of the currently used detection model, the network complexity has been greatly increased to improve accuracy. Therefore, the proposed network uses EfficientNet as a feature extraction network, and the subsequent layers are formed in a pyramid structure to utilize low-level detailed features and high-level semantic features. An attention process was applied between pyramid structures to suppress unnecessary noise for prediction. All computational processes of the network are replaced by depth-wise and point-wise convolutions to minimize the amount of computation. The proposed network was trained and evaluated using the PASCAL VOC dataset. The features fused through the experiment showed robust properties for various objects through a refinement process. Compared with the CNN-based detection model, detection accuracy is improved with a small amount of computation. It is considered necessary to adjust the anchor ratio according to the size of the object as a future study.