Estimation of maize planting area using mixed field decomposition of multi-temporal TM images

Abstract

Mid-resolution image has been the main data used to estimate the crop plant area in large scale. There are two problems of spectra variation inside and spectral mixed on boundary of the farmland, which greatly reduce the reliability of crops per-pixel classification. Aimed to the percentage of maize planting area in the mixed parcel, the support vector machine (SVM) algorithm is introduced to develop mixed field decomposition model by using different proportion of train sample parcels. The positional accuracy and gross accuracy of maize planting area of mixed field decomposition with SVM algorithm are evaluated and contrasted to that of decision tree classification. The study showed that the positional accuracy and gross accuracy of estimating maize planting area based on SVM model using multi-source information could reach 84.06% and 98.43% respectively, which was far higher than that of single source information. Comparing the per-pixel classification of decision tree, estimation of maize planting area based on mixed field decomposition has higher gross and positional accuracy. It indicates that mixed field decomposition could establish efficient separating hyperplane of SVM among the percentage of maize in the mixed parcels, which makes the accuracy of plant area of maize reach reliable.