Abstract
The availability of agricultural land use information allows decision makers and managers to establish short-term and to long-term plans for land conservation and sustainable use. The objective of this study was to develop a method for extraction of agricultural land use information based on remote sensing imagery. By combining particle swarm optimization (PSO), k-means clustering algorithm and minimum distance classifier, a PSO-k-means-based minimum distance classifier for agricultural land use classification was developed. Crop planting information was collected and divided into five classes: water bodies, paddy fields, bean fields, wheat fields and others (windbreak, roads, rare areas, and buildings, etc.). K-means, a widely used algorithm in pattern recognition for unsupervised classification, became a part of supervised classification by using PSO to find the optimal initial position vectors in a training sample pretreatment process. The optimal cluster of each subclass was finally used for minimum distance classification. The results obtained from Miyajimanuma wetland land use information extraction showed that merely using a small feature space composed of the first three principal components of a SPOT 5 image enabled classification accuracy of 93%.
