Hierarchically Channel-Wise Attention Model for Clean and Polluted Water Images Classification
Water image classification is challenging because clean water images of ocean or river share almost the same properties with images of polluted water. Inspired by the significant power of Convolutional Neural Network (CNN) in extracting various features for classification, we intend to utilize CNN to classify clean and polluted water images based on quantity of channel-wise and multi-layer CNN-extracted features. Since not all features are informative for water image classification, a dynamic feature attention scheme that utilize the properties of channel-wise and multi-layer is necessary to achieve robust and accurate results. In this paper, we propose a novel hierarchically channel-wise attention model for clean and polluted water images classification. The proposed model dynamically modulates context with multi-layer feature maps in a local and global sense, constructing a representative combination of features to boost classification performance. Experimental results on a latest water image dataset (reporting 71.2% in accuracy) with several comparative methods demonstrate the effectiveness and robustness of the proposed model incorporating CNN for water image classification.
KeywordsHierarchical attention model Channel-wise CNN feature Multi-layer CNN feature Channel-wise attention module
This work was supported by National Key R&D Program of China under Grant 2018YFC0407901, the Natural Science Foundation of China under Grant Grant 61702160, Grant61672273 and Grant 61832008, the Science Foundation of Jiangsu under Grant BK20170892, and Key R&D Program Projects in Jiangsu Province under Grant BE2015707.
- 1.Hu, J., Shen, L., Sun, G.: Squeeze-and-excitation networks. arXiv preprint arXiv:1709.01507 (2017)
- 2.Khan, M., Wu, X., Xu, X., Dou, W.: Big data challenges and opportunities in the hype of industry 4.0. In: Proceedings of IEEE International Conference on Communications, pp. 1–6 (2017)Google Scholar
- 4.Mnih, V., Heess, N., Graves, A., Kavukcuoglu, K.: Recurrent models of visual attention. In: Proceedings of NIPS, pp. 2204–2212 (2014)Google Scholar
- 5.Prasad, M.G., Chakraborty, A., Chalasani, R., Chandran, S.: Quadcopter-based stagnant water identification. In: Proceedings of Fifth National Conference on Computer Vision, Pattern Recognition, Image Processing and Graphics (NCVPRIPG), pp. 1–4 (2015)Google Scholar
- 6.Qi, L., Chen, Y., Yuan, Y., Fu, S., Zhang, X., Xu, X.: A QoS-aware virtual machine scheduling method for energy conservation in cloud-based cyber-physical systems. World Wide Web, pp. 1–23 (2019)Google Scholar
- 9.Rankin, A.L., Matthies, L.H., Bellutta, P.: Daytime water detection based on sky reflections. In: Proceedings of ICRA, pp. 5329–5336 (2011)Google Scholar
- 10.Santana, P., Mendonça, R., Barata, J.: Water detection with segmentation guided dynamic texture recognition. In: Proceedings of IEEE International Conference on Robotics and Biomimetics (ROBIO), pp. 1836–1841 (2012)Google Scholar
- 11.Wu, X., Shivakumara, P., Zhu, L., Lu, T., Pal, U., Blumenstein, M.: Fourier transform based features for clean and polluted water image classification. In: Proceedings of International Conference on Pattern Recognition (2018)Google Scholar
- 12.Xu, X., Liu, Q., Zhang, X., Zhang, J., Qi, L., Dou, W.: A blockchain-powered crowdsourcing method with privacy preservation in mobile environment. In: IEEE Transactions on Computational Social Systems (2019)Google Scholar
- 13.Xu, X., Zhang, X., Gao, H., Xue, Y., Qi, L., Dou, W.: Become: blockchain-enabled computation offloading for iot in mobile edge computing. In: IEEE Transactions on Industrial Informatics (2019)Google Scholar