• ETRI Journal
• /
• v.44 no.4
• /
• pp.560-572
• /
• 2022

H²RDC (heuristic handover based on RCC-DTSK-C), a heuristic algorithm based on a highly interpretable deep Takagi-Sugeno-Kang fuzzy classifier, is proposed for suppressing the mobile heterogeneous networks problem of frequent handover and handover ping-pong in the multibase-station scenario. This classifier uses a stack structure between subsystems to form a deep classifier before generating a base station (BS) priority sequence during the handover process, and adaptive handover hysteresis is calculated. Simulation results show that H²RDC allows user equipment to switch to the best antenna at the optimal time. In high-BS density load and mobility scenarios, the proposed algorithm's handover success rate is similar to those of classic algorithms such as best connection (BC), self tuning handover algorithm (STHA), and heuristic for handover based on AHP-TOPSIS-FUZZY (H²ATF). Moreover, the handover rate is 83% lower under H²RDC than under BC, whereas the handover ping-pong rate is 76% lower.

• Journal of Biomedical Engineering Research
• /
• v.38 no.1
• /
• pp.25-31
• /
• 2017

Testing TB in chest X-ray images is a typical method to diagnose presence and magnitude of PTB lesion. However, the method has limitation due to inter-reader variability. Therefore, it is essential to overcome this drawback with automatic interpretation. In this study, we propose a novel method for detection of PTB using SegNet, which is a deep learning architecture for semantic pixel wise image labelling. SegNet is composed of a stack of encoders followed by a corresponding decoder stack which feeds into a soft-max classification layer. We modified parameters of SegNet to change the number of classes from 12 to 2 (TB or none-TB) and applied the architecture to automatically interpret chest radiographs. 552 chest X-ray images, provided by The Korean Institute of Tuberculosis, used for training and test and we constructed a receiver operating characteristic (ROC) curve. As a consequence, the area under the curve (AUC) was 90.4% (95% CI:[85.1, 95.7]) with a classification accuracy of 84.3%. A sensitivity was 85.7% and specificity was 82.8% on 431 training images (TB 172, none-TB 259) and 121 test images (TB 63, none-TB 58). This results show that detecting PTB using SegNet is comparable to other PTB detection methods.

• KIPS Transactions on Software and Data Engineering
• /
• v.12 no.2
• /
• pp.83-90
• /
• 2023

Blurred image is an important factor in lowering image recognition rates in Computer vision. This mainly occurs when the camera is unstablely out of focus or the object in the scene moves quickly during the exposure time. Blurred images greatly degrade visual quality, weakening visibility, and this phenomenon occurs frequently despite the continuous development digital camera technology. In this paper, it replace the modified building module based on the Deep multi-patch neural network designed with convolution neural networks to capture details of input images and Attention techniques to focus on objects in blurred images in many ways and strengthen the image. It measures and assigns each weight at different scales to differentiate the blurring of change and restores from rough to fine levels of the image to adjust both global and local region sequentially. Through this method, it show excellent results that recover degraded image quality, extract efficient object detection and features, and complement color constancy.

• Journal of the Korean Institute of Electrical and Electronic Material Engineers
• /
• v.36 no.3
• /
• pp.226-232
• /
• 2023

As a potential alternative to flash memory, HfO₂/Al₂O₃ stacks appear to be a viable option as charge capture layers in charge trapping memories. The paper undertakes a review of HfO₂/Al₂O₃ stacks as charge trapping layers, with a focus on comparing the number, thickness, and post-deposition heat treatment and γ-ray and white x-ray treatment of such stacks. Compared to a single HfO₂ layer, the memory window of the 5-layered stack increased by 152.4% after O₂ annealing at ±12 V. The memory window enlarged with the increase in number of layers in the stack and the increase in the Al/Hf content in the stack. Furthermore, our comparison of the treatment of HfO₂/Al₂O₃ stacks with varying annealing temperatures revealed that an increased annealing temperature resulted in a wider storage window. The samples treated with O₂ and subjected to various γ radiation intensities displayed superior resistance. and the memory window increased to 12.6 V at ±16 V for 100 kGy radiation intensity compared to the untreated samples. It has also been established that increasing doses of white x-rays induced a greater number of deep defects. The optimization of stacking layers along with post-deposition treatment condition can play significant role in extending the memory window.

• Journal of the Microelectronics and Packaging Society
• /
• v.12 no.2 s.35
• /
• pp.111-119
• /
• 2005

Stack specimen with three dimensional interconnection structure through Cu via of $75{\mu}m$ diameter, $90{\mu}m$ height and $150{\mu}m$ pitch was successfully fabricated using subsequent processes of via hole formation with Deep RIE (reactive ion etching), Cu via filling with pulse-reverse electroplating, Si thinning with CMP, photolithography, metal film sputtering, Cu/Sn bump formation, and flip chip bonding. Contact resistance of Cu/Sn bump and Cu via resistance could be determined ken the slope of the daisy chain resistance vs the number of bump joints of the flip chip specimen containing Cu via. When flip- chip bonded at $270^{\circ}C$ for 2 minutes, the contact resistance of the Cu/Sn bump joints of $100{\times}100{\mu}m$ size was 6.7m$\Omega$ and the Cu via resistance of $75{\mu}m$ diameter, $90{\mu}m$ height was 2.3m$\Omega$.

• The Journal of the Convergence on Culture Technology
• /
• v.6 no.4
• /
• pp.789-795
• /
• 2020

Many deaths have been reported due to the worldwide pandemic of COVID-19. In order to prevent the further spread of COVID-19, it is necessary to quickly and accurately read images of suspected patients and take appropriate measures. To this end, this paper introduces a deep learning-based COVID-19 chest X-ray reading technique that can assist in image reading by providing medical staff whether a patient is infected. First of all, in order to learn the reading model, a sufficient dataset must be secured, but the currently provided COVID-19 open dataset does not have enough image data to ensure the accuracy of learning. Therefore, we solved the image data number imbalance problem that degrades AI learning performance by using a Stacked Generative Adversarial Network(StackGAN++). Next, the DenseNet-based classification model was trained using the augmented data set to develop the reading model. This classification model is a model for binary classification of normal chest X-ray and COVID-19 chest X-ray, and the performance of the model was evaluated using part of the actual image data as test data. Finally, the reliability of the model was secured by presenting the basis for judging the presence or absence of disease in the input image using Grad-CAM, one of the explainable artificial intelligence called XAI.

• Geophysics and Geophysical Exploration
• /
• v.26 no.4
• /
• pp.199-210
• /
• 2023

The physical properties of rocks in reservoirs change after CO₂ injection, we modeled a reservoir with a transition zone within which the physical properties change linearly. The function of the Wolf reflection coefficient consists of the velocity ratio of the upper and lower layers, the frequency, and the thickness of the transition zone. This function can be used to estimate the thickness of a reservoir or seafloor transition zone. In this study, we propose a method for predicting the thickness of the transition zone using deep learning. To apply deep learning, we modeled the thickness-dependent Wolf reflection coefficient on an artificial transition zone formation model consisting of sandstone reservoir and shale cap rock and generated time-frequency spectral images using the continuous wavelet transform. Although thickness estimation performed by comparing spectral images according to different thicknesses and a spectral image from a trace of the seismic stack did not always provide accurate thicknesses, it can be applied to field data by obtaining training data in various environments and thus improving its accuracy.

• Tunnel and Underground Space
• /
• v.21 no.6
• /
• pp.518-525
• /
• 2011

In this study, the natural ventilation pressure resulting from the large altitude difference which is a characteristic of high radioactive waste repository and the caloric value of the heat emitted by wastes was calculated and based on the results, natural ventilation quantities were calculated. A high radioactive waste repository can be considered as being operated through closed cycle thermodynamic processes similar to those of thermal engines. The heat produced by the heating of high radioactive wastes in the underground repository is added to the surrounding air, and the air goes up through the upcast vertical shaft due to the added heat while working on its surroundings. Part of the heat added by the work done by the air can be temporarily changed into mechanical energy to promote the air flow. Therefore, if a sustained and powerful heat source exists in the repository, the heat source will naturally enable continued cyclic flows of air. Based on this assumption, the quantity of natural ventilation made during the disposal of high radioactive wastes in a deep geological layer was mathematically calculated and based on the results, natural ventilation pressure of $74{\sim}183$Pa made by the stack effect was identified along with the resultant natural ventilation quantity of $92.5{\sim}147.7m^3/s$. The result of an analysis by CFD was $82{\sim}143m^3/s$ which was very similar to the results obtained by the mathematical method.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.9
• /
• pp.2505-2528
• /
• 2023

The field of facial expression recognition (FER) has been actively researched to improve human-computer interaction. In recent years, deep learning techniques have gained popularity for addressing FER, with numerous studies proposing end-to-end frameworks that stack or widen significant convolutional neural network layers. While this has led to improved performance, it has also resulted in larger model sizes and longer inference times. To overcome this challenge, our work introduces a novel lightweight model architecture. The architecture incorporates three key factors: Depth-wise Separable Convolution, Residual Block, and Attention Modules. By doing so, we aim to strike a balance between model size, inference speed, and accuracy in FER tasks. Through extensive experimentation on popular benchmark FER datasets, our proposed method has demonstrated promising results. Notably, it stands out due to its substantial reduction in parameter count and faster inference time, while maintaining accuracy levels comparable to other lightweight models discussed in the existing literature.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.3
• /
• pp.1188-1203
• /
• 2020

The human expression contains a lot of information that can be used to detect complex conditions such as pain and fatigue. After deep learning became the mainstream method, the traditional feature extraction method no longer has advantages. However, in order to achieve higher accuracy, researchers continue to stack the number of layers of the neural network, which makes the real-time performance of the model weak. Therefore, this paper proposed an expression recognition framework based on densely concatenated convolutional neural networks to balance accuracy and latency and apply it to humanoid robots. The techniques of feature reuse and parameter compression in the framework improved the learning ability of the model and greatly reduced the parameters. Experiments showed that the proposed model can reduce tens of times the parameters at the expense of little accuracy.