Objective prediction of leather mechanical tactile properties based on structured light 3D reconstruction and deep learning

To objectify the tactile feelings, a deep learning method based on line laser measurement and auxiliary labelling is proposed to evaluate the tactile properties of different types of leather. The leather deformation is measured via the projected line laser stripe on the surface, and multidimensional mechanical feature vectors are derived from the load sequence and deformation data to characterize the leather’s mechanical response. The tactile properties of different leather samples are judged by manual touch, and the result is used for labelling the samples. A dual-task neural network is trained to map mechanical features to tactile properties, where the main task is to predict the tactile properties of each leather sample, and predicting the certain regions within a sample is an auxiliary task. The model is evaluated using feature similarity analysis, showing that different regions from a sample have high similarity in the feature space, whereas different samples exhibit low similarity. Experimental results demonstrate clear feature clustering for regions of the same leather and certain separation across different leathers, supporting the model’s effectiveness in tactile property prediction.