The Research On Edge Detection And Transfer Classification For Reclamation Remote Sensing Images

Remote sensing technology has become a powerful means of reclamation monitoring due to its rapid and accurate observation ability for large scale space exploration. But because of serious spectral aliasing, the automation degree of the remote sensing image interpretation is relatively low, which restricts the application of remote sensing in the reclamation dynamic monitoring.This paper, from the view of data characteristics of remote sensing image, mainly researches the edge detection and transfer classification for reclamation remote sensing information extraction technology.First of all, in view of the traditional edge detection algorithm in the application of remote sensing image affected by noise in the serious and easy to produce false edge, calculation ability of cellular automata and the fast pattern mining advantage of extreme learning machine are combined, so that the optimal transfer rules between spectrum and edge can be obtained. An extreme learning machine based cellular automata algorithm for remote sensing image edge detection is proposed to reduce the impact of false edges.Secondly, as remote sensing image classification algorithms lack sample reuse strategy, they are unable to deal with multitemporal remote sensing images who have the spectrum drift phenomenon between each image. In the framework of transfer learning, remote sensing image classification model can be extended to transfer classification algorithm. To improve the ability to adapt to different model domain, transfer learning of extreme learning machine is proposed. At the same time, in order to get rid of the restriction of the large scene image endmember spectral variability on transfer learning, transfer learning of improved Bayesian ARTMAP algorithm is proposed. The establishment of recycling strategy of remote sensing samples, improves the multitemporal remote sensing image classification accuracy and timeliness of reclamation monitoring.Finally, proposed algorithms are used to extract the land use/land cover information and the change information of 1987-2014 Liaohe estuarine wetland remote sensing image, so that analysis of the impact of reclamation activities for ecological service value is generated, and these results can provide references for further analysis and research.