The application of intelligent manipulator has been widely used in medical

military

agricultural

assembly industries and other fields. As one of the important physical properties of objects

softness and hardness have a great influence on grasping control object by intelligent manipulator. Within depth learning framework

a soft and hard objects recognition method for tactile perception based on convolutional neural network（ CNN） is proposed. Thin-film pressure sensor is used to acquire data of finger pressing soft and bard objects

training and test data set are set up. Train network in caffer

so as to simulate tactile to identify soft and hard object. The experimental results show that the accuracy of recognition on soft and hard objects is 94. 52 %

which indicates that the CNN has a good classification effect for recognition of soft and hard objects.