• KIPS Transactions on Computer and Communication Systems
• /
• v.8 no.11
• /
• pp.251-262
• /
• 2019

In this paper, we propose a simulator that provides various algorithms of RBF networks on neuromorphic chips. To develop algorithms based on neuromorphic chips, the disadvantages of using simulators are that it is difficult to test various types of algorithms, although time is fast. This proposed simulator can simulate four times more types of network architecture than existing simulators, and it provides an additional a two-layer structure algorithm in particular, unlike RBF networks provided by existing simulators. This two-layer architecture algorithm is configured to be utilized for multiple input data and compared to the existing RBF for performance analysis and validation of utilization. The analysis showed that the two-layer structure algorithm was more accurate than the existing RBF networks.

• Smart Structures and Systems
• /
• v.29 no.1
• /
• pp.181-193
• /
• 2022

Data-driven structural health monitoring (SHM) of civil infrastructure can be used to continuously assess the state of a structure, allowing preemptive safety measures to be carried out. Long-term monitoring of large-scale civil infrastructure often involves data-collection using a network of numerous sensors of various types. Malfunctioning sensors in the network are common, which can disrupt the condition assessment and even lead to false-negative indications of damage. The overwhelming size of the data collected renders manual approaches to ensure data quality intractable. The task of detecting and classifying an anomaly in the raw data is non-trivial. We propose an approach to automate this task, improving upon the previously developed technique of image-based pre-processing on one-dimensional (1D) data by enriching the features of the neural network input data with multiple channels. In particular, feature engineering is employed to convert the measured time histories into a 3-channel image comprised of (i) the time history, (ii) the spectrogram, and (iii) the probability density function representation of the signal. To demonstrate this approach, a CNN model is designed and trained on a dataset consisting of acceleration records of sensors installed on a long-span bridge, with the goal of fault detection and classification. The effect of imbalance in anomaly patterns observed is studied to better account for unseen test cases. The proposed framework achieves high overall accuracy and recall even when tested on an unseen dataset that is much larger than the samples used for training, offering a viable solution for implementation on full-scale structures where limited labeled-training data is available.

• KIPS Transactions on Software and Data Engineering
• /
• v.11 no.9
• /
• pp.371-380
• /
• 2022

Convolutional neural networks are widely used to manipulate data arranged in a grid, such as images. A general convolutional neural network consists of a convolutional layers and a fully connected layers, and each layer contains a nonlinear activation functions. This paper proposes a combined parametric activation function to improve the performance of convolutional neural networks. The combined parametric activation function is created by adding the parametric activation functions to which parameters that convert the scale and location of the activation function are applied. Various nonlinear intervals can be created according to parameters that convert multiple scales and locations, and parameters can be learned in the direction of minimizing the loss function calculated by the given input data. As a result of testing the performance of the convolutional neural network using the combined parametric activation function on the MNIST, Fashion MNIST, CIFAR10 and CIFAR100 classification problems, it was confirmed that it had better performance than other activation functions.

• Asia-pacific Journal of Multimedia Services Convergent with Art, Humanities, and Sociology
• /
• v.7 no.3
• /
• pp.339-349
• /
• 2017

This paper proposes a four-stage algorithm for detecting lanes on a driving car. In the first stage, it extracts region of interests in an image. In the second stage, it employs a median filter to remove noise. In the third stage, a binary algorithm is used to classify two classes of backgrond and foreground of an input image. Finally, an image erosion algorithm is utilized to obtain clear lanes by removing noises and edges remained after the binary process. However, the proposed lane detection algorithm requires high computational time. To address this issue, this paper presents a parallel implementation of a real-time line detection algorithm on a multi-core architecture. In addition, we implement and simulate 8 different processing element (PE) architectures to select an optimal PE architecture for the target application. Experimental results indicate that 40×40 PE architecture show the best performance, energy efficiency and area efficiency.

• Journal of the Institute of Electronics Engineers of Korea SD
• /
• v.42 no.11
• /
• pp.9-16
• /
• 2005

This paper proposes a programmable PLL (phase locked loop) based clock generator supporting a wide-range-frequency input and output for high performance and low power SoC with multiple clock frequencies domains. The propose system reduces the locking time and obtains a wide range operation frequency by using a dual-charge pumps scheme. For low power operation of a chip, the locking processing circuits of the proposed PLL doesn't be working in the standby mode but the locking data are retained by the DAC. Also, a tracking ADC is designed for the fast relocking operation after stand-by mode exit. The programmable output frequency selection's circuit are designed for supporting a optimized DFS operation according to job tasks. The proposed PLL-based clock system has a relock time range of $0.85{\mu}sec{\sim}1.3{\mu}sec$($24\~26$cycle) with 2.3V power supply, which is fabricated on $0.35{\mu}m$ CMOS Process. At power-down mode, PLL power saves more than $95\%$ of locking mode. Also, the PLL using programmable divider has a wide locking range ($81MHz\~556MHz$) for various clock domains on a multiple IPs system.

• Journal of KIISE:Databases
• /
• v.37 no.1
• /
• pp.1-11
• /
• 2010

Recently, there has been a growing interest in the processing of continuous queries over multiple data streams. When the arrival rates of tuples exceed the memory capacity of the system, a load shedding technique is used to avoid the system becoming overloaded by dropping some subset of input tuples. In this paper, we propose an effective load shedding algorithm for multi-way windowed joins over multiple data streams. Most previous load shedding algorithms estimate the productivity of each tuple, i.e., the number of join output tuples produced by the tuple, based on its "join attribute value" and drop tuples with the lowest productivity. However, the productivity of a tuple cannot be accurately estimated from its join attribute value when the join attribute values are unique and do not repeat, or the distribution of the join attribute values changes over time. For these cases, we estimate the productivity of a tuple based on its "arrival order" on data streams, rather than its join attribute value. The proposed method can effectively estimate the productivity of a tuple even when the productivity of a tuple cannot be accurately estimated from its join attribute value. Through extensive experiments and analysis, we show that our proposed method outperforms the previous methods in terms of effectiveness and efficiency.

• Journal of Korea Multimedia Society
• /
• v.10 no.8
• /
• pp.1068-1077
• /
• 2007

This paper proposes a method that extracts characteristics of cracks such as length, thickness and direction from a concrete slab surface image with image processing techniques. These techniques extract the cracks from the concrete surface image in variable conditions including bad image conditions) using the ART2-based RBF network to recognize the dominant directions -45 degree, 45 degree, horizontal and vertical) of the extracted cracks from the automatically calculated specifications like the lengths, directions and widths of the cracks. Our proposed extraction algorithms and analysis of the concrete cracks used a Robert operation to emphasize the cracks, and a Multiple operation to increase the difference in brightness between the cracks and background. After these treatments, the cracks can be extracted from the image by using an iterated binarization technique. Noise reduction techniques are used three separate times on this binarized image, and the specifications of the cracks are extracted form this noiseless image. The dominant directions can be recognized by using the ART2-based RBF network. In this method, the ART2 is used between the input layer and the middle layer to learn, and the Delta learning method is used between the middle layer and the output layer. The experiments using real concrete images showed that the cracks were effectively extracted, and the Proposed ART2-based RBF network effectively recognized the directions of the extracted cracks.

• KIPS Transactions on Software and Data Engineering
• /
• v.5 no.12
• /
• pp.653-662
• /
• 2016

This paper describes a test case generation algorithm for Simulink/Stateflow models based on the Rapidly exploring Random Tree (RRT) algorithm that has been successfully applied to path finding. An important factor influencing the performance of the RRT algorithm is the metric used for calculating the distance between the nodes in the RRT space. Since a test case for a Simulink/Stateflow (SL/SF) model is an input sequence to check a specific condition (called a test target in this paper) at a specific status of the model, it is necessary to drive the model to the status before checking the condition. A status maps to a node of the RRT. It is usually necessary to check various conditions at a specific status. For example, when the specific status represents an SL/SF model state from which multiple transitions are made, we must check multiple conditions to measure the transition coverage. We propose a unique distance calculation metric, based on the observation that the test targets are gathered around some specific status such as an SL/SF state, named key nodes in this paper. The proposed metric increases the probability that an RRT is extended from key nodes by imposing penalties to non-key nodes. A test case generation algorithm utilizing the proposed metric is proposed. Three models of Electrical Control Units (ECUs) embedded in a commercial vehicle are used for the performance evaluation. The performances are evaluated in terms of penalties and compared with those of the algorithm using a typical RRT algorithm.

• The KIPS Transactions:PartA
• /
• v.12A no.6 s.96
• /
• pp.515-522
• /
• 2005

This paper presents a method for effectively modeling inlay/onlay prostheses restoring a tooth that are partially destroyed. An inlay/onlay is composed of internal surface adhering to an abutment, and external surface revealed to the outside sight. Internal surfaces are modeled using Minkowski sum expanding the grinded surface parts of abutments so that the internal surfaces can adhere to the abutments with closer contact. In modeling external surfaces, we exploit 3D mesh deformation techniques: DMFFD(direct manipulation free-form deformation)[19] and MWD(multiple wires deformation)[17] with three kinds of informations: standard teeth models, mesh data obtained by scanning a plaster cast of a patient's tooth, FGP(functionally guided plane) measuring the occlusion of the patients's teeth. The standard teeth models are used for building up the basic shapes of external surfaces, while the plaster fast and FGP data are used for reflecting the unique properties of adjacent md occlusal surfaces of the patients's teeth, which are slightly different to each other but very important for correct functioning. With these informations as input data, the adjacent and occlusal surfaces are automatically generated as mesh data using the techniques of DMFFD and m, respectively. Our method was implemented so that inlay/onlay prostheses fan be designed more accurately by visualizing the generated mesh models with requirements by dentists.

• Journal of KIISE:Databases
• /
• v.33 no.6
• /
• pp.563-577
• /
• 2006

A stream processor uses resource sharing method for efficient of limited resource in multiple continuous queries. The previous methods process aggregate queries to consist the level structure. So insert operation needs to reconstruct cost of the level structure. Also a search operation needs to search cost of aggregation information in each size of sliding windows. Therefore this paper uses linear structure for optimization of sliding window aggregations. The method comprises of making decision, generation and deletion of panes in sequence. The decision phase determines optimum pane size for holding accurate aggregate information. The generation phase stores aggregate information of data per pane from stream buffer. At the deletion phase, panes are deleted that are no longer used. The proposed method uses resources less than the method where level structures were used as data structures as it uses linear data format. The input cost of aggregate information is saved by calculating only pane size of data though numerous stream data is arrived, and the search cost of aggregate information is also saved by linear searching though those sliding window size is different each other. In experiment, the proposed method has low usage of memory and the speed of query processing is increased.