Land Use and Land Cover Change Detection Techniques: A Data-driven and Application Based Perspective

摘要

During past decades, land use and land cover change detection techniques have undergone substantial development. However, different scenarios and an integrated workflow linking remote sensing imagery and GIS are often neglected. As a result, we develop a land use and land cover change detection and extraction system and propose five scenarios considering data availability and different classification techniques, which are pre-classification thresholding for bi-temporal images, post-unsupervised or supervised classification for vector and image, post-unsupervised or supervised classification for bi-temporal images. In this process, multiple classifiers and cluster algorithms such as K-means, ISODATA, pixel-based MLC and object-oriented SVM are included. The result shows post supervised classification scenario presents superiority. However, it can be declared that there is not a single method or technique which has the capability to suffice all the condition. In the future, the classification methods can be more diversified to adjust different data input in different regions and to improve accuracy.