Bayesian Transfer Learning for Object Detection in Remote Sensing Images

Abstract:

In the literature of object detection in optical remote sensing images, a popular pipeline is first modifying an off-the- shelf deep neural network, then initializing the modified network by pretrained weights on a source data set, and finally fine-tuning the network on a target data set. The procedure works well in practice but might not make full use of underlying knowledge implied by pretrained weights. In this article, we propose a novel method, referred to as Fisher regularization, for efficient knowledge transferring. Based on Bayes’ theorem, the method stores underlying knowledge into a Fisher information matrix and fine-tunes parameters based on the knowledge. The proposed method would not introduce extra parameters and less sensitive to hyperparameters than classical weight decay. Experiments on NWPUVHR-10 and DOTA data sets show that the proposed method is effective and works well with different object detectors.

Bibtex:

@ARTICLE{9066887,
  author={Changsheng {Zhou} and Jiangshe {Zhang} and Junmin {Liu} and Chunxia {Zhang} and Guang {Shi} and Junying {Hu}},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  title={Bayesian transfer learning for object detection in optical remote sensing images},
  year={2020},
  volume={0},
  number={0},
  pages={1-15}
  }