In the complex industrial processes, the process data have the characteristics of imbalance and are incomplete due to the difficult-to-measure key variables, leading to the performance degradation of soft sensors. In order to deal with this problem, a novel noise injection supervised enhanced autoencoder virtual sample generation method is proposed. Firstly, in order to strengthen the mapping relationship between input and output and ensure the integrity of feature information, this method adds an enhancement layer to the encoding part of the autoencoder, and introduces label information for supervised constraint training in the decoding part. In order to increase the diversity of virtual samples, Gaussian noise is added to the features extracted from the hidden layer of the supervised enhanced autoencoder. Combine the generated virtual samples with the original small samples to enhance the performance of the soft sensing model. Unlike traditional virtual sample generation methods, the proposed NISEAE-VSG model can simultaneously generate useful input-output virtual samples. To verify the effectiveness of the proposed method, simulations were conducted using datasets of thermal power generation and polyethylene processes. The simulation results show that the proposed method generates virtual samples that are superior to other virtual sample generation methods and can effectively improve the accuracy of soft sensing modeling.