Detecting urban green space from landsat7 ETM+ data by using an unmixing algorithm of support vector machine

Xiufang Zhu,Chunyang He,Yaozhong Pan,Jinshui Zhang
DOI: https://doi.org/10.1109/IGARSS.2005.1525402
2005-07-25
Abstract:Xiufang Zhu, Chunyang He*, Yaozhong Pan, Jinshui Zhang Key Laboratory of Environmental Change and Natural Disaster, Ministry of Education of China College of Resources Science and Technology Beijing Normal University Beijing, China * Corresponding author: hcy@ires.cn Abstract—Getting green space information quickly and accurately provides a foundation for urban environment, estimating green space system planning and all-level decisionmaking. Cheap and medium/low resolution remote sensing data (CBERS, SPOT, TM, MODIS) have being usually used for a regional green space detection. However, medium/low resolution data have the prominent mixed-pixel problem. Hence unmixing algorithm has become very important in medium/low resolution remote sensing data analysis. In order to improve the identification of green space and the performance of unmixing, a new approach is extended by using the Support Vector Machine (SVM)-based algorithm, which combines the information of spectrum, texture and structure. After the validation and comparison with the Spectral Mixture Analysis (SMA), which is the most popular unmixing algorithm, results indicate that the approach based on SVM has higher performance and precision than that of SMA. SVM algorithm is a promising method for a regional green space detection based on medium/low resolution image.
What problem does this paper attempt to address?