• Journal of Sensor Science and Technology
• /
• v.29 no.1
• /
• pp.19-26
• /
• 2020

Electrocardiogram (ECG) classification has become an essential task of modern day wearable devices, and can be used to detect cardiovascular diseases. State-of-the-art Artificial Intelligence (AI)-based ECG classifiers have been designed using various artificial neural networks (ANNs). Despite their high accuracy, ANNs require significant computational resources and power. Herein, three different ANNs have been compared: multilayer perceptron (MLP), convolutional neural network (CNN), and spiking neural network (SNN) only for the ECG classification. The ANN model has been developed in Python and Theano, trained on a central processing unit (CPU) platform, and deployed on a PYNQ-Z2 FPGA board to validate the model using a Jupyter notebook. Meanwhile, the hardware accelerator is designed with Overlay, which is a hardware library on PYNQ. For classification, the MIT-BIH dataset obtained from the Physionet library is used. The resulting ANN system can accurately classify four ECG types: normal, atrial premature contraction, left bundle branch block, and premature ventricular contraction. The performance of the ECG classifier models is evaluated based on accuracy and power. Among the three AI algorithms, the SNN requires the lowest power consumption of 0.226 W on-chip, followed by MLP (1.677 W), and CNN (2.266 W). However, the highest accuracy is achieved by the CNN (95%), followed by MLP (76%) and SNN (90%).

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.29 no.5C
• /
• pp.618-626
• /
• 2004

In this paper, we describe a robot endeffector recognition system based on a Modular Neural Networks (MNN). The proposed recognition system can be used for vision system which track a given object using a sequence of images from a camera unit. The main objective to achieve with the designed MNN is to precisely recognize the given robot endeffector and to minimize the processing time. Since the robot endeffector can be viewed in many different shapes in 3- D space, a MNN structure, which contains a set of feedforwared neural networks, can be more attractive in recognizing the given object. Each single neural network learns the endeffector with a cluster of training patterns. The training MNN patterns for a neural network share the similar characteristics so that they can be easily trained. The trained UM is les s sensitive to noise and it shows the better performance in recognizing the endeffector. The recognition rate of MNN is enhanced by 14％ over the single neural network. A vision system with the MNN can precisely recognize the endeffector and place it at the center of a display for a remote operator.

• Journal of the Korean Society for Precision Engineering
• /
• v.24 no.2 s.191
• /
• pp.47-57
• /
• 2007

The vibration environment essentially companied by vehicle operation on the road is determined by the shape of road surface, which is called profile. In general, the profile and severity of unpaved road is an important issue in the reliability of durability test for vehicles. In order to maintain severity of unpaved road, it is necessary to develop profilometer system. We developed profilometer system which is composed of data processing computer, power unit, air compressor and sensors. This paper focuses on the severity management of unpaved test courses using neural networks. This paper presents the maintenance range for cross-country course in CPG(Chang-won Proving Ground) and the evaluation of similarity degree between unpaved roads.

• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.11
• /
• pp.483-490
• /
• 2021

In order to learn an existing song and create a new song using an artificial neural network, it is necessary to convert the song into numerical data that the neural network can recognize as a preprocessing process, and one-hot encoding has been used until now. In this paper, we proposed a note embedding method using notes as a basic unit and a bar embedding method that uses the bar as the basic unit, and compared the performance with the existing one-hot encoding. The performance comparison was conducted based on quantitative evaluation to determine which method produced a song more similar to the song composed by the composer, and quantitative evaluation methods used in the field of natural language processing were used as the evaluation method. As a result of the evaluation, the song created with bar embedding was the best, followed by note embedding. This is significant in that the note embedding and bar embedding proposed in this paper create a song that is more similar to the song composed by the composer than the existing one-hot encoding.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.40 no.6
• /
• pp.56-68
• /
• 2003

Difficult problems In implementing digital neural network hardware are the extension of synapses and the programmability for relocating neurons. In this paper, the structure of a new hardware is proposed for solving these problems. Our structure based on traditional SIMD can be dynamically and easily reconfigured connections of network without synthesizing and mapping original design for each use. Using additional modular processing unit the numbers of neurons find synapses increase. To show the extensibility of our structure, various models of neural networks : multi-layer perceptrons and Kohonen network are formed and tested. The performance comparison with software simulation shows its superiority in the aspects of performance and flexibility.

• Proceedings of the KIEE Conference
• /
• 1999.07b
• /
• pp.546-549
• /
• 1999

For the safty driving of an automobile which is become individual requisites, a new Neural Network algorithm which recognized the load vehicles in real time is proposed. The proposed neural network technique is the real time computation method through the inter-node diffusion. In the network, a node corresponds to a state in the quantized input space. Each node is composed of a processing unit and fixed weights from its neighbor nodes as well as its input terminal. The most reliable algorithm derived for real time recognition of vehicles, is a dynamic programming based algorithm based on sequence matching techniques that would process the data as it arrives and could therefore provide continuously updated neighbor information estimates. Through several simulation experiments, real time reconstruction of the nonlinear image information is processed 1-D LIPN hardware has been composed and various experiments with static and dynamic signals have been implmented.

• Proceedings of the KIEE Conference
• /
• 2003.07d
• /
• pp.2130-2133
• /
• 2003

For real-time recognizing of the load vehicles a new Neural Network algorithm is proposed. The proposed neural network technique is the real time computation method through the inter-node diffusion. In the network, a node corresponds to a state in the quantized input space. Each node is composed of a Processing unit and fixed weights from its neighbor nodes as well as its input terminal. The most reliable algorithm derived for real time recognition of vehicles, is a dynamic programming based algorithm based on sequence matching techniques that would process the data as it arrives and could therefore provide continuously updated neighbor information estimates. Through severa simulation experiments, real time reconstruction nonlinear image information is Processed. 1-D hardware has been composed and various experi with static and dynamic signals have implemented.

• International Journal of Internet, Broadcasting and Communication
• /
• v.11 no.4
• /
• pp.76-85
• /
• 2019

We introduce a vision-based object detection method for real-time video surveillance system in low-end edge computing environments. Recently, the accuracy of object detection has been improved due to the performance of approaches based on deep learning algorithm such as Region Convolutional Neural Network(R-CNN) which has two stage for inferencing. On the other hand, one stage detection algorithms such as single-shot detection (SSD) and you only look once (YOLO) have been developed at the expense of some accuracy and can be used for real-time systems. However, high-performance hardware such as General-Purpose computing on Graphics Processing Unit(GPGPU) is required to still achieve excellent object detection performance and speed. To address hardware requirement that is burdensome to low-end edge computing environments, We propose sub-frame analysis method for the object detection. In specific, We divide a whole image frame into smaller ones then inference them on Convolutional Neural Network (CNN) based image detection network, which is much faster than conventional network designed forfull frame image. We reduced its computationalrequirementsignificantly without losing throughput and object detection accuracy with the proposed method.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.17 no.6
• /
• pp.367-374
• /
• 2022

The acceleration of neural networks has become an important topic in the field of computer vision. An accelerator is absolutely necessary for accelerating the lightweight model. Most accelerator-supported operators focused on direct convolution operations. If the accelerator does not provide GEMM operation, it is mostly replaced by CPU operation. In this paper, we proposed an optimization technique for 2-stage tiling-based GEMM routines on VTA. We improved performance of the matrix multiplication routine by maximizing the reusability of the input matrix and optimizing the operation pipelining. In addition, we applied the proposed technique to the DarkNet framework to check the performance improvement of the matrix multiplication routine. The proposed GEMM method showed a performance improvement of more than 2.4 times compared to the non-optimized GEMM method. The inference performance of our DarkNet framework has also improved by at least 2.3 times.

• Journal of Information Processing Systems
• /
• v.19 no.6
• /
• pp.791-802
• /
• 2023

The manual inspection of photovoltaic (PV) panels to meet the requirements of inspection work for large-scale PV power plants is challenging. We present a hot spot detection and positioning method to detect hot spots in batches and locate their latitudes and longitudes. First, a network based on the YOLOv3 architecture was utilized to identify hot spots. The innovation is to modify the RU_1 unit in the YOLOv3 model for hot spot detection in the far field of view and add a neural network residual unit for fusion. In addition, because of the misidentification problem in the infrared images of the solar PV panels, the DeepLab v3+ model was adopted to segment the PV panels to filter out the misidentification caused by bright spots on the ground. Finally, the latitude and longitude of the hot spot are calculated according to the geometric positioning method utilizing known information such as the drone's yaw angle, shooting height, and lens field-of-view. The experimental results indicate that the hot spot recognition rate accuracy is above 98%. When keeping the drone 25 m off the ground, the hot spot positioning error is at the decimeter level.