Wheat acreage detection by extended support vector analysis with multi-temporal remote sensing images

Abstract

An Extended Support Vector Machine (ESVM), which could improve the conventional change detection methods by purifying thematic classification results and neglecting of the side effect by mixed pixel, is proposed for crop acreage measurement by multi-temporal remote sensing land cover change. The ESVM method combines the concept of hard and soft classification and divides the study objects into winter wheat, non-winter wheat and transitional areas. A case study was conducted at DaXing district of Beijing, China, and two remote sensing images from HJ-1/CCD were utilized in ESVM and conventional SVM methods to extract winter wheat area. The results showed that the crop distribution derived from the ESVM reflects reality more accurately than that from SVM. Crops acreage classified from ESVM gave lower RMSE than that from SVM in all window size. This research is therefore demonstrated a new winter wheat detection method which combines hard and soft classification with multi-temporal images.