The invention relates to the technical field of machine learning, The invention discloses an object hardness identification method based on a Gaussian mixture noise generation countermeasure network,which comprises the following steps of: training a Gaussian mixture noise generation countermeasure network with a small-scale tactile data marking a hardness level as a true value, inputting the Gaussian mixture noise into the Gaussian mixture noise generation countermeasure network, and obtaining large-scale generated samples; Taking the parameters of the discriminator of the Gaussian mixture noise generation type antagonistic network as the initial values of the parameters of the hardness identification network, pre-training the hardness identification network by using the large-scale generated samples, re-training the hardness identification network by using the tactile data marking the hardness level, and determining the parameters of the hardness identification network; inputting Thetactile data to be predicted a into a hardness recognition network to obtain a hardness grade of the tactile data to be predicted. The invention has high hardness identification accuracy.