Abstract:
Objective Normalization method plays an important role in feature preprocessing phase not only in conventional machine learning domain but also in contemporary deep learning domain. Batch normalization (BN) is very successful, but its performance very depends on the sample size. Therefore, many researchers try to improve it when the sample size is inadequate through adding the sample size merely in the sample information space.
Methods This paper utilizes Bayes theory to integrate general information, prior information and sample information, and to offset the inadequate sample information. In this way, the mean value and the variance of sample can be estimated more precisely and more robust especially when the sample size is small, and the normalized feature will better fall into non-saturating region of activation function, which enables deep learning model to better describe original feature space.
Results The Top-1 test classification accuracy in NWPU-RESISC45 dataset is improved by 5.64% than that of BN method. Moreover, with the help of general information and prior information, the Bayes adjoint batch normalization （BABN) method is not sensitive to the sample size.
Conclusions The experiment results show that the proposed BABN method is feasible and effective, and performs better than BN method and other variants in the scene recognition of remote sensing image.
下载:
