ABSTRACT
Convolutional Neural Networks (CNNs) have improved image feature extraction ability at various scales. However, it fails to match the input features with different importance or complexity into the appropriate feature extraction branches more effectively. To solve this problem, we propose a novel statistical characteristics-based Multi-Scale Image Feature Extraction (SC-MSIFE) scheme for CNN, which adaptively matches batch image feature subsets into the appropriate feature extraction branches. We calculate and aggregate the gray distribution statistics of features to characterize the complexity, importance and interdependencies of batch image feature subsets, respectively. Then, we reorder the batch image feature subsets according to the gained information. Finally, we match the complex and significant batch image feature subsets into multi-scale feature extraction branches, while inputting the batch image feature subsets with simple and unimportant features into fewer-scale feature extraction branches. Extensive simulation results demonstrate the effectiveness of our proposed approach compared to baselines in terms of improving classification accuracy.
- Krizhevsky A, Sutskever I, and Hinton G E, 2017. Imagenet Classification with Deep Convolutional Neural Networks. Commun. ACM 60, 6 (June 2017), 84–90. http://dx.doi.org/10.1145/3065386.Google ScholarDigital Library
- Sermanet P, Eigen D, Zhang X, Mathieu M, Fergus R, and Lecun Y, 2013. Overfeat: Integrated Recognition, Localization and Detection Using Convolutional Networks. arXiv:1312.6229. http://dx.doi.org/10.48550/arXiv.1312.6229.Google ScholarCross Ref
- Zeiler M D and Fergus R, 2014. Visualizing and Understanding Convolutional Networks. In 13th European Conference on Computer Vision (ECCV), Zurich, SWITZERLAND, 818-833. http://dx.doi.org/10.1007/978-3-319-10590-1_53.Google ScholarCross Ref
- Xu Y, Zhang L, Du B, and Zhang F, 2018. Spectral–Spatial Unified Networks for Hyperspectral Image Classification. IEEE Transactions on Geoscience and Remote Sensing 56, 10, 5893-5909. http://dx.doi.org/10.1109/TGRS.2018.2827407.Google ScholarCross Ref
- Simonyan K and Zisserman A, 2014. Very Deep Convolutional Networks for Large-Scale Image Recognition. CoRR abs/1409.1556.Google Scholar
- Szegedy C, Ioffe S, Vanhoucke V, and Alemi A A, 2016. Inception-V4, Inception-Resnet and the Impact of Residual Connections on Learning. ArXiv abs/1602.07261.Google Scholar
- Szegedy C, Wei L, Yangqing J, Sermanet P, Reed S, Anguelov D, Erhan D, Vanhoucke V, and Rabinovich A, 2015. Going Deeper with Convolutions. In 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 1-9. http://dx.doi.org/10.1109/CVPR.2015.7298594.Google ScholarCross Ref
- Szegedy C, Vanhoucke V, Ioffe S, Shlens J, and Wojna Z, 2016. Rethinking the Inception Architecture for Computer Vision. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2818-2826. http://dx.doi.org/10.1109/CVPR.2016.308.Google ScholarCross Ref
- He K, Zhang X, Ren S, and Sun J, 2016. Deep Residual Learning for Image Recognition. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 770-778. http://dx.doi.org/10.1109/CVPR.2016.90.Google ScholarCross Ref
- Huang G, Liu Z, Maaten L V D, and Weinberger K Q, 2017. Densely Connected Convolutional Networks. In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2261-2269. http://dx.doi.org/10.1109/CVPR.2017.243.Google ScholarCross Ref
- Yu F, Wang D, Shelhamer E, and Darrell T, 2018. Deep Layer Aggregation. In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2403-2412. http://dx.doi.org/10.1109/CVPR.2018.00255.Google ScholarCross Ref
- Gao S H, Cheng M M, Zhao K, Zhang X Y, Yang M H, and Torr P, 2021. Res2net: A New Multi-Scale Backbone Architecture. IEEE Transactions on Pattern Analysis and Machine Intelligence 43, 2, 652-662. http://dx.doi.org/10.1109/TPAMI.2019.2938758.Google ScholarDigital Library
- Lan R S, Sun L, Liu Z B, Lu H M, Pang C, and Luo X N, 2021. Madnet: A Fast and Lightweight Network for Single-Image Super Resolution. IEEE TRANSACTIONS ON CYBERNETICS 51, 3 (MAR), 1443-1453. http://dx.doi.org/10.1109/TCYB.2020.2970104.Google ScholarCross Ref
- Zhou W J, Fan X M, Yu L, and Lei J S, 2023. Misnet: Multiscale Cross-Layer Interactive and Similarity Refinement Network for Scene Parsing of Aerial Images. Ieee Journal of Selected Topics in Applied Earth Observations and Remote Sensing 16, 2025-2034. http://dx.doi.org/10.1109/jstars.2023.3243247.Google ScholarCross Ref
- Qi G Q, Zhang Y C, Wang K P, Mazur N, Liu Y, and Malaviya D, 2022. Small Object Detection Method Based on Adaptive Spatial Parallel Convolution and Fast Multi-Scale Fusion. Remote Sensing 14, 2 (Jan). http://dx.doi.org/10.3390/rs14020420.Google ScholarCross Ref
- Krizhevsky A, 2009. Learning Multiple Layers of Features from Tiny Images. In (2009).Google Scholar
- Lee C Y, Xie S, Gallagher P, Zhang Z, and Tu Z, 2014. Deeply-Supervised Nets. arXiv:1409.5185v2, 562-570. http://dx.doi.org/10.48550/arXiv.1409.5185.Google ScholarCross Ref
Recommendations
Multi-view Multi-task Feature Extraction for Web Image Classification
MM '14: Proceedings of the 22nd ACM international conference on MultimediaThe features used in many multimedia analysis-based applications are frequently of very high dimension. Feature extraction offers several advantages in highly dimensional cases, and many recent studies have used multi-task feature extraction approaches, ...
Dynamic image fine feature extraction method based on machine vision
ICASIT 2020: Proceedings of the 2020 International Conference on Aviation Safety and Information TechnologyIn the dynamic image feature extraction of machine vision, a single key frame is often used to extract, which affects the accuracy of parameters when obtaining features. Therefore, this paper proposes a method of dynamic image subtle feature extraction ...
Multi-Scale Spatial-Spectral Feature Extraction Based on Dilated Convolution for Hyperspectral Image Classification
ICIGP '23: Proceedings of the 2023 6th International Conference on Image and Graphics ProcessingSupervised classification algorithms for hyperspectral image (HSI) require a large number of labels for model training, which will bring labeling cost and computational burden. To solve this problem, a dual-branch multi-scale three-dimensional ...
Comments