Abstract: A solid waste identification method based on optical flow estimation for hyperspectral satellite remote sensing data is proposed in this paper.Firstly, this paper rethinks hyperspectral data from the perspective of sequence data, and proposes to introduce DeepFlow optical flow estimation technology to extract brightness change information of spectral dimension as a more discriminative spectral motion feature. Then, the extracted spectral motion features are combined with the original spectral features and input to the commonly used support vector machine for solid waste recognition to improve the recognition accuracy. Finally, a specific method of post-processing for solid waste identification is proposed to improve the identification effect. In this paper, the remote sensing data of “Zhuhai-1” hyperspectral satellite is used, and the experiment is carried out by taking Jiyuan demonstration area in Henan province as an example. The experimental results show that the proposed method can quickly and accurately identify the industrial solid wastes stacked in the open air in a wide range, and preliminarily lock the 11 regional risk points in Jiyuan demonstration area where there are solid wastes left and illegal stacking behaviors, and the accuracy is better than the traditional spectral feature extraction and classification methods. Thus, the time and workload are significantly saved for the later manual on-site investigation of solid waste and “waste clearance” action.

Key words: hyperspectral remote sensing, solid waste identification, optical flow estimation, spectral motion feature, Zhuhai-1