Material recognition is most applicable in the field of recycling where the input objects are composed of various materials with different recycling methods. For real-world waste, the images are more complex than normal labeled footage [1]. Our study presents a novel approach to material classification using analysis of the object’s shape and surface temperature under an induced heat transfer state by a radiant heat source and heat sink provided by the Peltier devices. The Peltier devices are powered by a DC generator and controlled by a MOSFET and an Arduino circuit. The Peltier devices’ temperature was controlled to reveal the response to temperature change which reveals more variable to the heat capacity and conductivity of the object. The collected data were calculated by three heat transfer theories using Python. The first is classic heat transfer theory which gives the highest error but needs the fewest steps to calculate [2]. The second is the heat diffusion theory which increases the accuracy, but only heat conductivity can be calculated [3]. The third method is to simulate the object’s heat transfer in a computer and estimate the thermal property until the simulated surface temperature matches the real temperature. The results from these three methods are compared to the database of real material properties to estimate errors. The CPU time is also recorded for comparison. The errors from these methods and the applied conditions are also compared with the required computational power, as measured by CPU time. The third method has the lowest error of 16.2%, and the CPU time depends on the resolution of the thermal footage, but the third method has one to two times the CPU time needed. Our research aims to start with homogeneous materials in a controlled environment, then progress in complexity to real-world waste.