• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1997.10a
• /
• pp.271-274
• /
• 1997

This paper presents an adaptive quantization of image sequences using the Radial Basis Function Network(RBFN) which classifies interframe image blocks. The clssification algorithm consists of two steps. Blocks are classified into NA(No Activity), SA(Small Activity), VA(Verical Activity), and HA(Horizontal Activity) classes according to edges, image activity and AC anergy distribution. RBFN is trained using the classification results of the above algorithm, which are nonlinear classification features are acquired from the complexity and variability of difference blocks. Simulation result shows that the the adaptive quantization using the RBFN method produced better results better results than that of the sorting and MLP methods.

• 제어로봇시스템학회:학술대회논문집
• /
• 1991.10a
• /
• pp.280-285
• /
• 1991

Neural network control has many innovative potentials for fast, accurate and intelligent adaptive control. In this paper, a learning control architecture for the dynamic control of a robot manipulator is developed using inverse dynamic neurocontroller and linear neurocontroher. The inverse dynamic neurocontrouer consists of a MLP (multi-layer perceptron) and the linear neurocontroller consists of SLPs (single layer perceptron). Compared with the previous type of neurocontroller which is using an inverse dynamic neurocontroller and a fixed PD gain controller, proposed architecture shows the superior performance over the previous type of neurocontroller because linear neurocontroller can adapt its gain according to the applied task. This superior performance is tested and verified through the control of PUMA 560. Without any knowledge on the dynamic model, its parameters of a robot , (The robot is treated as a complete black box), the neurocontroller, through practice, gradually and implicitly learns the robot's dynamic properties which is essential for fast and accurate control.

• 제어로봇시스템학회:학술대회논문집
• /
• 1991.10a
• /
• pp.1022-1027
• /
• 1991

A new method to control a robot manipulator by neural networks is proposed. The controller is composed of both a PD controller and a neural network-based feedforward controller. MLP(multi-layer perceptron) neural network is used for the feedforward controller and trained by BP(back-propagation) learning rule. Error terms for BP learning rule are composed of the outputs of a PD controller and the acceleration errors of manipulator joints. We compare the proposed method with existing ones and contrast performances of them by simulation. Also, We discuss the real application of the proposed method in consideration of the learning time of the neural network and the time required for sensing the joint acceleration.

• ETRI Journal
• /
• v.22 no.3
• /
• pp.10-18
• /
• 2000

Although an n-th order cross-entropy (nCE) error function resolves the incorrect saturation problem of conventional error backpropagation (EBP) algorithm, performance of multilayer perceptrons (MLPs) trained using the nCE function depends heavily on the order of nCE. In this paper, we propose an adaptive learning rate to markedly reduce the sensitivity of MLP performance to the order of nCE. Additionally, we propose to limit error signal values at out-put nodes for stable learning with the adaptive learning rate. Through simulations of handwritten digit recognition and isolated-word recognition tasks, it was verified that the proposed method successfully reduced the performance dependency of MLPs on the nCE order while maintaining advantages of the nCE function.

• Proceedings of the Korean Society of Machine Tool Engineers Conference
• /
• 2003.10a
• /
• pp.125-130
• /
• 2003

This Paper describes a Nonlinear adoptive noise canceller using Neural Network for Machine Tools Controller System. Back-Propagation Learning Algorithm based MLP (Multi Layer Perceptron)is used an adaptive filters. In this Paper. it assume that the noise of primary input in the adaptive noise canceller is not the same characteristic as that of the reference input. Experimental results show that the neural network base noise canceller outperforms the linear noise canceller. Especially to make noise cancel close to realtime, Primary Input is divided by Unit and each divided pan is processed for very short time than all the processed data are unified to whole data.

• Journal of Institute of Control, Robotics and Systems
• /
• v.21 no.10
• /
• pp.982-987
• /
• 2015

The inspection machine in SMT (Surface Mount Technology) line detects the assembly defects such as missing, misalignment, loosing, or tombstone. We propose a new method to classify the defect types of chip components by processing the image of PCB. Two original images are obtained from horizontal lighting and vertical lighting. The image of the component is divided into two soldering regions and one packaging region. The features are extracted by appling the PCA (Principle Component Analysis) to each region. The MLP (Multilayer Perceptron) and SVM (Support Vector Machine) are then used to classify the defect types by learning. The experimental results are presented to show the usefulness of the proposed method.

• Proceedings of the IEEK Conference
• /
• 1999.11a
• /
• pp.456-459
• /
• 1999

In this paper, we propose a method of unconstrained handwritten numeral recognition using image dithering and multiple modular MLPs. The set of sample numeral patterns is subdivided into clusters which are extended by their radius. On each extended cluster, we constructed MLPs network as the expert recognizer of corresponding cluster. The gating network is also trained by an MLPs to weigh the outputs of expert MLPs. In training and test phase of the recognizer, we utilize the multiple dithered numeral images and the combination of the outputs for corresponding dithered images. Experimental results show that our recognition method works very well.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.32B no.5
• /
• pp.756-764
• /
• 1995

In this paper, we proposed MLP(MultiLayer Perceptron) based HMM's(Hidden Markov Models) with bounded state duration for isolated word recognition. The minimum and maximum state duration for each state of a HMM are estimated during the training phase and used as parameters of constraining state transition in a recognition phase. The procedure for estimating these parameters and the recognition algorithm using the proposed HMM's are also described. Speaker independent isolated word recognition experiments using a vocabulary of 10 city names and 11 digits indicate that recognition rate can be improved by adjusting the minimum state durations.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.32B no.7
• /
• pp.1010-1024
• /
• 1995

An intelligent trajectory control method that controls a direction and a average velocity for a prosthetic arm by force and direction estimations using EMG signals is proposed. 3 stage linear filters are used as a real time joint trajectory planner to minimize the impact to human body induced by arm motions and to reduce muscle fatigues. We use combination of MLP and fuzzy filter for a limb direction estimation and a time model of force for determining a cartesian trajectory control parameter. EMG signals are acquired by using a amputation simulator and 2 dimensional joystick motion. Simulation results of the proposed method show that the arm is effectively followed the desired trajectory by estimated foreces and directions. This method reduces the number of electrodes and attatched sites compared with the method using Hogan's impedance control.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.31B no.4
• /
• pp.175-182
• /
• 1994

In this study, we present unvoiced consonant recognition system using characteristic parameters of the phoneme of the each syllable. For the recognition, the characteristic parameters on the time domain such as ZCR, total energy of the consonant region and half region energy of the consonant region, and those on the frequency domain such as the frequency spectrum of the transition region are used. The objective unvoiced consonants in this study are /ㄱ/,/ㄷ/,/ㅂ/,/ㅈ/,/ㅋ/,/ㅌ/,/ㅍ/ and /ㅊ/. Each characteristic parameter of two regions extracted from these segmented unvoiced consonants are used for each recognition system of the region, independently, And complementing two outputs of each other system, the final output is to be produced. The recognition system is implemented using MLP which has learning ability. The recognition simulation results for 112 unvoiced consonant samples are that average recognition rates are 96.4$\%$ under 80$\%$ learning rates and 93.7$\%$ under 60$\%$ learning rates.