Design of a Hybrid Fuzzy–K-Means System for Object Type Recognition Using Flex Sensor Data
Design of a Hybrid Fuzzy–K-Means System for Object Type Recognition Using Flex Sensor Data
Maryam Moazedi1 ghasem lotfi2
1) Assistant Professor, Department of Advanced Engineering, Faculty of Advanced Technologies, University of Mohaghegh Ardabili, Namin, Iran
2) MSc Student, Department of Advanced Engineering, Faculty of Advanced Technologies, University of Mohaghegh Ardabili, Namin, Iran
IOI Code
:
AAHD-ABGHCA
Publication :
National Conference on Information Technology, Nanotechnology, Artificial Intelligence and Technological Futures Studies(itnaf.ir)
Abstract :
Smart gloves equipped with flex sensors are among the effective tools for human–machine interaction and rehabilitation, and they have found widespread applications in various fields. Despite recent advances, challenges such as high cost, the need for user or operator calibration, errors in recognizing incomplete movements, and dependence on cloud processing still remain. Therefore, research and development in designing low‑cost sensors, robust algorithms based on fuzzy logic and machine learning, as well as edge-processing architectures, are essential for achieving practical, accurate, and accessible smart gloves. In this study, a fuzzy system based on K-Means clustering is proposed for recognizing the type of object held in the hand using five flex sensors. First, despite the availability of linguistic labels, the data are clustered using the K-Means algorithm to obtain cluster centers. Then, the dispersion or standard deviation of each cluster is calculated to design the membership functions. Fuzzy membership functions are defined based on these centers and standard deviations. Three membership functions are designed for each sensor. Fuzzy rules are formulated and evaluated using the AND operator. The output of each rule is a number between 0 and 2, which is mapped to the nearest label after defuzzification. By replacing the minimum operator with the average and then the product operator, the accuracy increased from 54.8% to 82.2% and finally to 86.2%. Considering the simplicity of the method, the limited number of sensors, and its capability for real-time implementation, the results show better performance compared with similar methods.
Keywords : hand gesture flex sensor fuzzy logic k means clustering gaussian membership function