<p>Objects with varying degrees of hardness can withstand different maximum contact forces, and thus it is essential to identify hardness attributes of objects to prevent damage. This paper proposes an object hardness identification method for the precise grasping capabilities of a compliant gripper. In the pre-grasping stage, the compliant gripper can identify the hardness grade through the tactile data collected by the sensor, which helps it to apply appropriate contact force. Firstly, an experimental platform for tactile data acquisition is built, and the original tactile sequence dataset of grasping fruits, vegetables and other objects is collected by using MS1616S tactile sensor. Subsequently, based on the original dataset and the published tactile dataset, the classification model of object hardness identification based on deep learning is established. A Deep Belief Network (DBN) algorithm, renowned for its efficacy in handling nonlinear relationships, is employed to preprocess data dimension reduction and noise removal. The Least Squares Support Vector Machine (LSSVM) algorithm is then used to identify the hardness grade of the preprocessed tactile data, and compared with the K-Nearest Neighbor (KNN) algorithm. In these two datasets, the classification accuracy of DBN-LSSVM is above 98%. Finally, through the online identification experiment of grasping objects, the practicability of DBN-LSSVM classifier is verified, and the online identification accuracy is above 80%.</p>