Object classification system using temperature variation of smart finger device via machine learning

• Published in: Sensors and actuators. A. Physical. Vol. 356; p. 114338
• Main Authors: Park, Heon Ick, Cho, Tae Jin, Choi, In-Geol, Rhee, Min Suk, Cha, Youngsu
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 16.06.2023
• Subjects: Machine learning; Object detection; Temperature sensing; Temperature sensing; Machine learning; Object detection
• ISSN: 0924-4247; 1873-3069
• DOI: 10.1016/j.sna.2023.114338

In this study, we proposed smart finger devices (SFDs) for an object classification system using unimodal temperature sensors. Each SFD comprised a module with a flexible thermoelectric device (TED) and a resistance temperature detector (RTD) sensor embedded in a silicone finger cot mounted on a robot gripper. The stored Peltier heat on the TED of the SFD was transferred to the object when the robot gripper grasped it. The RTD sensor data obtained through a one-dimensional convolutional neural network (1D-CNN) distinguished materials with similar thermal conductivities. Through two preprocessing steps, the sensor data were fed into the designed classifier to identify ten selected objects. Finally, our configured classifier performed real-time recognition using unimodal temperature sensors. [Display omitted] •Smart finger device had module of a temperature sensor and a thermoelectric device.•A pair of smart finger devices used to classify ten objects.•The device achieved high accuracy despite small thermal conductivity differences.•A neural network with a small and fast architecture was adopted to process sensor data.