ABSTRACT
For remote sensing scene classification (RSSC), exemplar-based class-incremental learning uses all the training data of the new classes and a small number of exemplars of the old classes to train the model in each incremental learning phase, which leads to the failure of fully preserving the historical knowledge as well as the problem of imbalance between the old and new data. In this paper, we compress the exemplars by downsampling the non-discriminative pixels (e.g., the background) to save more compressed exemplars under a fixed memory budget, so that the incremental learning model can alleviate the imbalance problem while better preserving the historical knowledge. Specifically, this compression is achieved by generating 0-1 masks of discriminative pixels from class activation map (CAM). In addition, convolutional block attention module (CBAM) is added to help CAM accurately acquire the locations of discriminative pixels. Furthermore, adaptive mask generation model called class-incremental masking (CIM) adaptively finds optimal thresholds for different classes in class-incremental learning and transforms the heat map of CAM into a 0-1 mask with class-specific thresholds. We conducted experiments on two publicly available remote sensing scene data and the experimental results show that the classification performance of the proposed method outstands that of several state-of-the-art methods.
- T. Jiraporn and I. Sarun. 2023. Plant Species Classification Using Leaf Edge Feature Combination with Morphological Transformations and SIFT Key Point, Journal of Image and Graphics, Vol. 11, No. 1, pp. 91-97. DOI:10.18178/joig.11.1.91-97.Google ScholarCross Ref
- K. Saminathan, B. Sowmiya, and M.Chithra Devi. 2023. Multiclass Classification of Paddy Leaf Diseases Using Random Forest Classifier. Journal of Image and Graphics, Vol. 11, No. 2, pp. 195-203. DOI:10.18178/joig.11.2.195-203.Google ScholarCross Ref
- H. A. Ahmed, H. M. Hama, S. I. Jalal, 2023. Deep Learning in Grapevine Leaves Varieties Classification Based on Dense Convolutional Network. Journal of Image and Graphics, Vol. 11, No. 1, pp. 98-103. DOI:10.18178/joig.11.1.98-103.Google ScholarCross Ref
- G. Cheng, X. Xie, J. Han, 2020. Remote Sensing Image Scene Classification Meets Deep Learning: Challenges, Methods, Benchmarks, and Opportunities. In IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 13, pp. 3735-3756. DOI: 10.1109/JSTARS.2020.3005403.Google ScholarCross Ref
- L. Bai, Q. Liu, C. Li, 2022. Remote Sensing Image Scene Classification Using Multiscale Feature Fusion Covariance Network With Octave Convolution. In IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1-14, Art no. 5620214.DOI: 10.1109/TGRS.2022.3160492.Google ScholarCross Ref
- M. McCloskey and N. J. Cohen. 1989. Catastrophic interference in connectionist networks: The sequential learning problem. In Psychology of Learning and Motivation. Amsterdam, The Netherlands: Elsevier, vol. 24, pp.109–165. DOI:10.1016/S0079-7421(08)60536-8.Google ScholarCross Ref
- D. Zhou, Q. Wang, Z. Qi. 2023. Deep class-incremental learning: A survey[J]. arXiv preprint arXiv:2302.03648. DOI:10.48550/arXiv.2302.03648.Google ScholarCross Ref
- J. Kirkpatrick, R. Pascanu, and N. Rabinowitz. 2017. Overcoming catastrophic forgetting in neural networks[J]. Proceedings of the national academy of sciences, 114(13): 3521-3526. DOI: 10.48550/arXiv.1612.00796.Google ScholarCross Ref
- Z. Li and D. Hoiem. 2018. Learning without Forgetting. In IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 40, no. 12, pp. 2935-2947, 1 Dec. DOI: 10.1109/TPAMI.2017.2773081.Google ScholarDigital Library
- A. A. Rusu, N. C. Rabinowitz, G. Desjardins. 2016. Progressive neural networks[J]. arXiv preprint arXiv:1606.04671. DOI: 10.48550/arXiv.1606.04671Google ScholarCross Ref
- B. Zhao, X. Xiao, G. Gan, 2020. Maintaining discrimination and fairness in class incremental learning[C]. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 13208-13217. DOI: 10.48550/arXiv.1911.07053.Google ScholarCross Ref
- S. A. Rebuffi, A. Kolesnikov, G. Sperl, 2017. icarl: Incremental classifier and representation learning[C]. In Proceedings of the IEEE conference on Computer Vision and Pattern Recognition. 2001-2010. https://doi.org/10.48550/arXiv.1611.07725.Google ScholarCross Ref
- Y. Liu, B. Schiele, and Q. Sun. 2021. RMM: Reinforced memory management for class-incremental learning[J]. Advances in Neural Information Processing Systems, 34: 3478-3490. DOI: 10.48550/arXiv.2301.05792.Google ScholarCross Ref
- S. Woo, J. Park, J. Y. Lee, 2018. Cbam: Convolutional block attention module[C]. In Proceedings of the European conference on computer vision (ECCV). 3-19. DOI: 10.48550/arXiv.1807.06521.Google ScholarDigital Library
- B. Zhou, A. Khosla, and A. Lapedriza. 2016. Learning deep features for discriminative localization[C]. In Proceedings of the IEEE conference on computer vision and pattern recognition. 2921-2929. DOI: 10.48550/arXiv.1512.04150.Google ScholarCross Ref
- F. Wang, D. Zhou., H. Ye, 2022. Foster: Feature boosting and compression for class-incremental learning[C]. In European conference on computer vision. Cham: Springer Nature Switzerland, 398-414. DOI: 10.48550/arXiv.2204.04662.Google ScholarDigital Library
- Z. Luo, Y. Liu, B. Schiele, 2023. Class-incremental exemplar compression for class-incremental learning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 11371-11380. DOI: 10.48550/arXiv.2303.14042.Google ScholarCross Ref
- Y. Yang and S. Newsam. 2010. Bag-of-visual-words and spatial extensions for land-use classification. In Proc. 18th SIGSPATIAL Int. Conf. Adv. Geograph. Inf. Syst., pp. 270–279. DOI:10.1145/1869790.1869829.Google ScholarDigital Library
- G. S. Xia, J.Hu, F. Hu, 2017. AID: A benchmark data set for performance evaluation of aerial scene classification. IEEE Trans. Geosci. Remote Sens., vol. 55, no. 7, pp. 3965–3981. DOI: 10.1109/TGRS.2017.2685945.Google ScholarCross Ref
- D. Ye, J. Peng, H. Li, 2022. Better Memorization, Better Recall: A Lifelong Learning Framework for Remote Sensing Image Scene Classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 60: 1-14. DOI: 10.1109/TGRS.2022.3190392.Google ScholarCross Ref
Index Terms
- A Class-incremental Learning Method based on Exemplar Compression for Remote Sensing Scene Classification
Recommendations
Subspace prototype learning for few-Shot remote sensing scene classification
Highlights- Proposed subspace prototype classifier for few-shot remote sensing scene classification.
AbstractRecently, few-shot remote sensing scene classification (FSRSSC)has attracted more and more researchers’ attention. The FSRSSC aims to solve the problem of how to distinguish novel categories when there are limited labeled samples ...
Feature Transformation for Cross-domain Few-Shot Remote Sensing Scene Classification
Pattern Recognition, Computer Vision, and Image Processing. ICPR 2022 International Workshops and ChallengesAbstractEffectively classifying remote sensing scenes is still a challenge due to the increasing spatial resolution of remote imaging and large variances between remote sensing images. Existing research has greatly improved the performance of remote ...
A Self-Training Hierarchical Prototype-based Ensemble Framework for Remote Sensing Scene Classification
Highlights- A novel semi-supervised ensemble framework with a cross-checking mechanism is proposed for remote sensing scene classification;
- An enhanced self-training hierarchical prototype-based classifier is introduced as the base learner;
- A ...
AbstractRemote sensing scene classification plays a critical role in a wide range of real-world applications. Technically, however, scene classification is an extremely challenging task due to the huge complexity in remotely sensed scenes, and the ...
Comments