Surface roughness estimation by optimal tactile features for fruits and vegetables

A study was conducted to evaluate the surface roughness levels of fruits and vegetables by a novel tactile sensor. Firstly, the effective areas of the sensor were determined through the mechanical analysis with the ANSYS software, and the sensitive elements of polyvinylidene fluoride piezoelectric films and strain gauges were randomly arranged in these areas. When the sensor contacted with the surfaces of the fruits and vegetables, the signals produced by the sensitive elements were output and tactile features were obtained. Secondly, the D-score criterion was applied to evaluate the contribution of every tactile feature component in expressing the surface roughness levels. According to the value of D-score, the strategies of the sequential forward selection and equential forward floating selection were used to guide the optimization of feature components selection. Back propagation neural network model was applied to evaluate the performance of the optimal features. Finally, the experimental results revealed that the identification accuracy of the algorithm was up to 93.737%, which demonstrated that the optimal feature subsets possessed fewer dimensions while maintaining a high performance in expressing the surface roughness characteristics of the fruits and vegetables. The results also provided a basis for the optimized design of the tactile sensor.