Surface textures classification with tactile sensor signals

Tactile sensing plays a critical role in managing environmental interactions precisely in robotics and prosthetics technologies. One important research topic in this field is surface characterization. To understand the texture properties of surfaces, human tissue recognition systems are modeled using artificial finger structures. In this modeling, internal measurement units are placed on surfaces, creating vibrations that generate various electrical signals. These signals are then analyzed and classified using pattern recognition techniques. This thesis utilizes two different datasets for surface characterization. In the first dataset, the Fractal Detrended Fluctuation Analysis (F-DFA) and Support Vector Machines (SVM) methods are applied. F-DFA removes specific slopes from the data to enable more accurate surface analysis. SVM provides high accuracy in classifying these data, achieving a classification accuracy of %96.16. In the second dataset, features are extracted using Field-Programmable Gate Arrays (FPGA) integrated into the artificial finger structure, and these features are classified using Long Short-Term Memory (LSTM) algorithms. LSTM effectively processes time series data, learning the dynamic properties of tactile data, and achieves a classification accuracy of %96.87. These studies contribute significantly to enhancing the potential of tactile sensing in robotics and prosthetics, advancing the development of surface characterization and analysis methods. Innovative techniques like artificial finger structures and FPGA-based feature extraction offer important steps towards improving accuracy and efficiency in this research area.