Satellite remote sensing has essentially been applied in agriculture such as monitoring paddy growth stages, cropping schedule, and fertilizer management over vast areas. Many methods have been devoted for detecting paddy growth stages by deriving a certain profiles of vegetation indices from high temporal resolution satellite data. In this paper, we introduce a time-variant method referred to as a heuristic decision tree (HDT) for detecting 3-classes of paddy growth stages from a limited number of MODIS images and generalize the result through machine learning to predict the growth stages in time-invariant manner. We compared our proposed method to two previous studies in  and . For the time-variant case, all methods are validated by 124 times of observation data among 50 reference points from field survey. After doing some adjustments for calculating the accuracy, the result shows that the accuracy for [8, 9], and our proposed method are 0.8951, 0.9375, and 0.9435 respectively. For the time-invariant case, we evaluate three Kernel-based Regularized (KR) classification methods, i.e. Principal Component Regression (KRPCR), Extreme Learning Machine (KR-ELM), and Support Vector Machine with radial basis function (RBF-SVM). All data samples are divided into training (25%) and testing (75%) sampling, and all models are trained and tested through 10-rounds random bootstrap re-sampling method to obtain more variety on hypothesis models during learning. The best model for each classifier method is defined as the one with the highest kappa coefficient during testing. The experimental results for 3-classes of paddy growth stages prediction show that the classification accuracy in testing for our proposed method of each learner are 0.8485, 0.8874, and 0.8252 respectively, whereas for previous study  are 0.6845, 0.7845, and 0.6949 respectively.
|Number of pages||10|
|Journal||Journal of Theoretical and Applied Information Technology|
|Publication status||Published - 1 Jan 2015|
- Kernel based learner
- Paddy growth stages
- Remote sensing