Jo Jun 16, 2025
Deep Neural Network (DNN) is the core area of Artificial Intelligence (AI), which deals with algorithms that mechanically mimic the biological structure and function of brains. The deep learning model consists of a number of neurons, activation functions, optimization algorithms, data addition tools, etc.
Deep learning based on artificial neural networks is well suited for modeling, classifying and recognizing complex data such as images, speeches, texts, language translation, etc. However, the use of deep neural networks requires a large amount of training data. In order to identify personal information, for example, you need to record a large number of pictures and voices of each person for training. The large amount of training data leads to a huge amount of computation, and therefore, they are distributed to the computers connected to the Internet worldwide, or high-performance computers equipped with GPUs are used.
Kim Sun Il, a researcher at the Faculty of Metal Engineering, has established a detection system to increase the detection speed and rate of amorphous body detection, using deep neural network that is widely used worldwide, and evaluated its performance. He configured the deep neural network suitably for amorphous body detection based on YOLOv4 that provides high speed and detection rate for object detection.
First, he modified the K-means clustering method used in the standard YOLOv4 to fit the database used to increase the detection rate. Then, he modified the structure of the model to increase the detection speed by reducing the number of overlapping layers to 16 and the number of parameters to 17 059 472.
With the proposed deep neural network, the detection rate is 98% and the time of processing the image of a frame is 0.04s, which demonstrates its high detection accuracy and real-time performance.
...
Jo Jun 15, 2025
Mercury damages the nervous system, brain, kidney and lungs, etc. This heavy metal exists as a vapor or a liquid metal. In the mercurous state, it exists as inorganic salt, and in the mercuric state, it may form either inorganic salt or organometal compound. Mercury does not break down in the environment, and thus it remains as a persistent pollutant. The development of rapid and cost-effective tools for analyzing mercury (Hg) concentrations in environmental samples is very important as the toxicity of mercury is lethal to humans.
Various analytical methods have been developed to monitor mercury concentrations in environmental samples. They include atomic absorption spectrophotometric determination (AAS), atomic fluorescence spectrophotometric determination (AFS), and inductively coupled plasma mass spectrophotometric determination (ICP-MS). However, these methods are expensive, laborious and time-consuming.
To overcome these drawbacks, methods for low-cost and rapid detection of mercury in environmental samples using silver nanoparticles and gold nanoparticles have been widely developed in recent years. Silver nanoparticles are synthesized using plant extracts such as water apples and spinach for colorimetric detection of Hg. These analytical methods transform analyte concentration into color changes, which have the potential for qualitative and quantitative applications through colorimetric or naked-eye detection. Thus, colorimetric methods using silver nanoparticles and gold nanoparticles have shown to be promising tools for monitoring the concentration of mercury in environmental samples.
Ri Song Ho, a section head at the General Assay Office, has prepared nano silver solution by green-chemistry method, and proposed a low-cost, rapid and easy-to-use analytical method for spectrophotometric determination of mercury.
For preparing stable nano silver solution, he used aqueous extract of apple as bioreductant. The prepared nano silver solution was brown, with a surface plasmon resonance peak at 420nm. The addition of Hg2+ ions then changed the silver nanoparticles into colorless ones. The color change (decrease of absorbance) was proportional to the concentration of Hg2+ ions.
At the time of UV-visible spectrophotometric determination, the detection limit of the proposed method was 0.1mg/L and the relative standard deviation (RSD) was below 4%.
...
Jo Jun 13, 2025
In the industrial field, production processes are becoming more complex and the requirements of the stability and reliability of operation and the quality of products are becoming stricter. Therefore, the study on fault diagnosis and estimation is of great significance and research on it is widely conducted.
There have been many studies on fault estimation using estimation of unknown input signals, and there are several such methods including augmented state estimation with unknown input. Such approaches are of practical significance because they can be effectively applied to fault diagnosis and fault tolerant control. Usually, fault estimation is carried out through three steps; fault detection, isolation and estimation. Only when the information on the size of fault is available, effective fault-tolerant control is possible
Jang Myong Jun, a researcher at the Faculty of Automation Engineering, has studied the robust fault estimation method for nonlinear discrete time systems by combining the general unknown input observation method and H∞ attenuation method, and proposed an observer algorithm guaranteeing the convergence of the observer and the required disturbance attenuation with respect to fault estimation error. In addition, he has solved fault estimation problems using linear matrix inequality (LMI), thus reducing calculation quantities.
The simulation results for a multi-tank system show that the proposed approach is effective.
...
Jo Jun 12, 2025
Rapid development of IT has brought a change in the traditional attendance system. The Automatic Attendance System (AAS) identifies students to take attendance, and saves its result in the database to generate necessary information such as various statistics.
To identify students, it is necessary to read the information of physical means such as RFID (Radio Frequency Identification) card, or the biometric information such as their fingerprints or faces.
AAS based on RFID needs all students to carry their cards and involves manual operation when they forget to bring or lose it. In the meantime, the AAS based on fingerprints is fast and simple, and it needs no physical RFID cards. However, fingerprint readers must be installed at every lecture room and it takes long to recognize students by their fingerprints one by one.
Therefore, the AAS based on face recognition are accepted to be efficient as cameras are installed in most schools.
Recently, great progress in the field of face recognition has led to new applications of this technology, and the development of various products has been brisk. In a paper, they presented a system, called FaceNet, that directly learns a mapping from face images to a compact Euclidean space where face similarities are represented as distances. Once this space has been produced, tasks such as face recognition, verification and clustering can be easily implemented using standard techniques with FaceNet embeddings as feature vectors. This approach has great application effect as it shows great face recognition performance for face images of only 128 bytes.
The multi-face recognition system detects and recognizes several faces from an image, which requires high-resolution cameras. However, it is too expensive for some schools, and it is impossible to recognize all faces in a large classroom such as a practical training room because it usually has a view angle of 40 degrees. Of course, a high-resolution camera with a wide angle lens is a solution, but it is more practical to utilize the existing cameras than purchasing expensive new ones.
Kim Myong Jin, a lecturer at the Faculty of Information Science and Technology, has implemented multi-face recognition by combining several cameras and proposed an AAS based on it. This is a system that recognizes students’ faces from the images captured by existing different types of cameras, and automatically registers their attendance. For face recognition, FaceNet which shows high performance is used.
This system enables automatic register of attendance by means of low-resolution cameras even in large classrooms. The experiments have proved that the proposed system guarantees the accuracy of 100% in well-lit conditions with suitable allocation of cameras.
...
Jo Jun 10, 2025
Electricity production is an important issue for our life. At present, power generation depends on fossil fuels. Excessive fossil fuel consumption by human activities has led to serious atmospheric and environmental problems including global warming, greenhouse gas emission, climate change, ozone layer depletion and acid rain. One of the energy generation methods to reduce the impact of natural disasters is thermoelectric conversion technology.
A thermoelectric generator produces electricity by using the thermoelectric effect. Thermoelectric generators consist of several sets of thermoelectric modules inserted between two heat exchangers. Each module is then composed of several tens to hundreds of pairs of thermoelectric couples connected together electrically in series and thermally in parallel, which directly convert some of the thermal energy that passes through them into electricity. However, the defects of thermoelectric technology are low energy-conversion efficiency and high material cost.
So far, different plans to get a higher degree of its efficiency have been suggested. Most of them, however, based on the main dimension of heat exchangers, focus on the influences of factors including temperature, velocity and flow direction in fluid, and the type of its internals, but not on the effect of the array of thermoelectric devices.
Kim Kwang Jin, a researcher at the Semiconductor Institute, has carried out a coupled fluid-solid numerical simulation of the heat exchanger internal flow field of a low-power thermoelectric generator using the waste heat from stove.
He carried out thermal field simulations in the heat exchanger with fins and without fins and an analysis of the shape of the inner fins, which were in good agreement with the results of the previous studies. Then, he determined the design parameters of the thermoelectric generator through the simulations according to the main dimensions, the number of inner fins and the arrangement of thermoelectric elements of the heat exchanger. He found that the performance of the heat exchanger is better, when the main dimensions of the heat exchanger is 250×200×30mm, the array of thermoelectric devices is linear and the internal topology of the heat exchanger is straight with 9 fins.
...
Jo Jun 9, 2025
The environment of virtual motion chambers must be designed to be arbitrarily controlled according to the purpose. Main environmental parameters include temperature, humidity, CO2 concentration, and oxygen concentration. In order to measure and control these parameters, environmental sensors should be installed in various parts of the chamber for real-time monitoring control.
Existing sensors usually consist of a transducer that maps physical quantities to electrical signals, an analog-to-digital converter (ADC), a microprocessor, and an output interface that maps them to standard analog or digital signals. To measure physical quantities, the systematic error component can be modified by calibrating the sensors.
For noise filtering, hardware methods (e.g., installing analog active and analog passive reference frequency filters at the input of ADC or installing discrete active reference frequency filters at the output) or program filters (average filters, moving average filters, Kalman filters and artificial neural network filters) can be used. The device active reference filter is bulky and expensive, and consumes a large amount of power, and the passive reference frequency filter is less efficient, so it is most reasonable to use a program filter.
Under operating conditions, the input of the sensor has continuous signals, which vary randomly with time, and the rate of change is limited by physical reasons. In this case, a distorted signal appears at the output of the filter. The dynamic distortion of effective signals by the average filter and the moving average filter depends on the maximum rate of change of input signals. Therefore, it is effective to use the average filter or moving average filter when the change rate of input signals is not large.
Han Chol Hak, a researcher at the Faculty of Automation Engineering, has proposed an improved additive-mean algorithm that suits the characteristics of objects to improve the accuracy of oxygen concentration measurements in environmental sensors, and demonstrated the superiority of its filtering effect through comparisons with other algorithms.
The improved average algorithm can be used effectively for oxygen concentration sensing in embedded systems as well as for processing signals with low measurement signal variation and high noise intensity.
...
