• Journal of Korean Neurosurgical Society
• /
• v.66 no.1
• /
• pp.53-62
• /
• 2023

Objective : Deep learning is a machine learning approach based on artificial neural network training, and object detection algorithm using deep learning is used as the most powerful tool in image analysis. We analyzed and evaluated the diagnostic performance of a deep learning algorithm to identify skull fractures in plain radiographic images and investigated its clinical applicability. Methods : A total of 2026 plain radiographic images of the skull (fracture, 991; normal, 1035) were obtained from 741 patients. The RetinaNet architecture was used as a deep learning model. Precision, recall, and average precision were measured to evaluate the deep learning algorithm's diagnostic performance. Results : In ResNet-152, the average precision for intersection over union (IOU) 0.1, 0.3, and 0.5, were 0.7240, 0.6698, and 0.3687, respectively. When the intersection over union (IOU) and confidence threshold were 0.1, the precision was 0.7292, and the recall was 0.7650. When the IOU threshold was 0.1, and the confidence threshold was 0.6, the true and false rates were 82.9% and 17.1%, respectively. There were significant differences in the true/false and false-positive/false-negative ratios between the anterior-posterior, towne, and both lateral views (p=0.032 and p=0.003). Objects detected in false positives had vascular grooves and suture lines. In false negatives, the detection performance of the diastatic fractures, fractures crossing the suture line, and fractures around the vascular grooves and orbit was poor. Conclusion : The object detection algorithm applied with deep learning is expected to be a valuable tool in diagnosing skull fractures.

• Journal of Advanced Navigation Technology
• /
• v.25 no.6
• /
• pp.557-561
• /
• 2021

A meatadata has become an essential element in order to recommend video content to users. However, it is passively generated by video content providers. In the paper, a method for automatically generating metadata was studied in the existing manual metadata input method. In addition to the method of extracting emotion tags in the previous study, a study was conducted on a method for automatically generating metadata for genre and country of production through movie audio. The genre was extracted from the audio spectrogram using the ResNet34 artificial neural network model, a transfer learning model, and the language of the speaker in the movie was detected through speech recognition. Through this, it was possible to confirm the possibility of automatically generating metadata through artificial intelligence.

• Journal of Korean Society on Water Environment
• /
• v.33 no.3
• /
• pp.326-333
• /
• 2017

The lakes and reservoirs have been reported as important sources of carbon emissions to the atmosphere in many countries. Although field experiments and theoretical investigations based on the fundamental gas exchange theory have proposed the quantitative amounts of Net Atmospheric Flux (NAF) in various climate regions, there are still large uncertainties at the global scale estimation. Mechanistic models can be used for understanding and estimating the temporal and spatial variations of the NAFs considering complicated hydrodynamic and biogeochemical processes in a reservoir, but these models require extensive and expensive datasets and model parameters. On the other hand, data driven machine learning (ML) algorithms are likely to be alternative tools to estimate the NAFs in responding to independent environmental variables. The objective of this study was to develop random forest (RF) and multi-layer artificial neural network (ANN) models for the estimation of the daily $CO_2$ NAFs in Daecheong Reservoir located in Geum River of Korea, and compare the models performance against the multiple linear regression (MLR) model that proposed in the previous study (Chung et al., 2016). As a result, the RF and ANN models showed much enhanced performance in the estimation of the high NAF values, while MLR model significantly under estimated them. Across validation with 10-fold random samplings was applied to evaluate the performance of three models, and indicated that the ANN model is best, and followed by RF and MLR models.

• Nuclear Engineering and Technology
• /
• v.53 no.7
• /
• pp.2334-2340
• /
• 2021

Radioactive particle tracking (RPT) is a minimally invasive nuclear technique that tracks a radioactive particle inside a volume of interest by means of a mathematical location algorithm. During the past decades, many algorithms have been developed including ones based on artificial intelligence techniques. In this study, RPT technique is applied in a simulated test section that employs a simplified mixer filled with concrete, six scintillator detectors and a¹³⁷Cs radioactive particle emitting gamma rays of 662 keV. The test section was developed using MCNPX code, which is a mathematical code based on Monte Carlo simulation, and 3516 different radioactive particle positions (x,y,z) were simulated. Novelty of this paper is the use of a location algorithm based on a deep learning model, more specifically a 6-layers deep rectifier neural network (DRNN), in which hyperparameters were defined using a Bayesian optimization method. DRNN is a type of deep feedforward neural network that substitutes the usual sigmoid based activation functions, traditionally used in vanilla Multilayer Perceptron Networks, for rectified activation functions. Results show the great accuracy of the DRNN in a RPT tracking system. Root mean squared error for x, y and coordinates of the radioactive particle is, respectively, 0.03064, 0.02523 and 0.07653.

• Annual Conference on Human and Language Technology
• /
• 1989.10a
• /
• pp.126-131
• /
• 1989

본 논문에서는 Error Back Propagation 학습을 이용해 한글 문자를 On-Line 인식하는 시스템을 제안한다. Pointing device의 궤적을 추적해 입력 패턴의 특징(feature)을 추출해 신경 회로망 입력으로 준다. 이때 사용하는 특징은 기본 획 (stroke)의 종류 및 획간의 상대적 위치 관계이다. 학습과정에서는 자소의 정의를 읽어 초성, 중성, 종성에 대해 각 획수마다 정의된 신경회로망의 weight를 조정한다. 인식 과정에서는 초성, 중성, 종성의 순으로 에러가 최소인 획수의 신경회로망 출력을 택하여 2 바이트 조합형 코드로 완성한다. 이로써 Intelligent Man-Machine Interface 시스템중 위치 및 크기에 무관한 전필 입력 시스템을 구현한다.

• Journal of the military operations research society of Korea
• /
• v.17 no.1
• /
• pp.146-158
• /
• 1991

In this paper, we deal with the TSP (Traveling Salesperson Problem) which is well-known as NP-complete optimization problem. the TSP is applicable to network routing. task allocation or scheduling. and VLSI wiring. Well known numerical methods such as Newton's Metheod. Gradient Method, Simplex Method can not be applicable to find Global Solution but the just give Local Minimum. Exhaustive search over all cyclic paths requires 1/2 (n-1) ! paths, so there is no computer to solve more than 15-cities. Heuristic algorithm. Simulated Annealing, Artificial Neural Net method can be used to get reasonable near-optimum with polynomial execution time on problem size. Therefore, we are able to select the fittest one according to the environment of problem domain. Three methods are simulated about symmetric TSP with 30 and 50-city samples and are compared by means of the quality of solution and the running time.

• Nuclear Engineering and Technology
• /
• v.52 no.6
• /
• pp.1334-1339
• /
• 2020

This paper presents an empirical and comprehensive forecasting analysis of the uranium price. Prices are generally difficult to forecast, and the uranium price is not an exception because it is affected by many external factors, apart from imbalances between demand and supply. Therefore, a systematic analysis of multiple forecasting methods and combinations of them along repeated forecast origins is a way of discerning which method is most suitable. Results suggest that i) some sophisticated methods do not improve upon the Naïve's (horizontal) forecast and ii) Unobserved Components methods are the most powerful, although the gain in accuracy is not big. These two facts together imply that uranium prices are undoubtedly subject to many uncertainties.

• Transactions of the Korean Society of Machine Tool Engineers
• /
• v.15 no.2
• /
• pp.87-92
• /
• 2006

This paper supplements shortcoming of radioactivity check by detecting defect of SWP weld zone using ultrasonic wave. Manufacture 2 stage robot detection systems that can follow weld bead of SWP by method to detect weld defects of SWP that shape of weld bead is complex for this as quantitative. Also, through signal processing ultrasonic wave defect signal system of GUI environment that can grasp easily existence availability of defect because do videotex compose. Ultrasonic wave signal of weld defects develops artificial intelligence style sightseeing system to enhance pattern recognition of weld defects and the classification rate using neural net. Classification of weld defects that do fan Planar defect and that do volume defect of by classify.

• The Transactions of the Korean Institute of Electrical Engineers A
• /
• v.51 no.3
• /
• pp.109-117
• /
• 2002

High accuracy of the load forecasting for power systems improves the security of the power system and generation cost. However, the forecasting problem is difficult to handle due to the nonlinear and the random-like behavior of system loads as well as weather conditions and variation of economical environments. So far. many studies on the problem have been made to improve the prediction accuracy using deterministic, stochastic, knowledge based and artificial neural net(ANN) method. In the conventional load forecasting method, the load forecasting maximum error occurred for the holidays on Saturday and Monday. In order to reduce the load forecasting error of the daily peak load for the holidays on Saturday and Monday, fuzzy concept and linear regression theory have been adopted into the load forecasting problem. The proposed algorithm shows its good accuracy that the average percentage errors are 2.11% in 1996 and 2.84% in 1997.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.18 no.3
• /
• pp.101-108
• /
• 2023

Optimizing the performance of deep neural networks on embedded systems is a challenging task that requires efficient compilers and runtime systems. We propose a TVM-based approach that consists of three steps: quantization, auto-scheduling, and ahead-of-time compilation. Our approach reduces the computational complexity of models without significant loss of accuracy, and generates optimized code for various hardware platforms. We evaluate our approach on three representative CNNs using ImageNet Dataset on the NVIDIA Jetson AGX Xavier board and show that it outperforms baseline methods in terms of processing speed.