Using Supervised Classification Methods for the Analysis of Multi-spectral Signatures of Rice Varieties in Panama

Abstract

In this article supervised classification methods for the analysis of local Panamanian rice crops using Near-Infrared (NIR) spectral signatures are assessed. Neural network (Multilayer Perceptron-MLP) and Tree based (Decision Trees-DT and Random Forest-RF) algorithms are used as regression and supervised classification of the spectral signatures by rice varieties, against other crops and by plant phenology (days after planting). Also, satellite derived spectral signature is validated with a field collected spectral model. Results suggest that MLP networks, either for regression or classification, were more efficient (RMSE of 8.78 and 0.068, respectively) than either tree based methods to regress/classify the rice spectral signature (RMSE of 19.37,19.09 and 0.979, respectively). The validation made using satellite derived spectral signatures resulted in MLP models with RMSE of 0.216 and 7.318, respectively, leaving room for further improvement of the models. This work aimed to present a practical example of the employment of recent supervised classification algorithms for the determination of regression and classification models from reflectance spectral signatures in local rice varieties.

Comparing Two Methods of Leaf Area Index Estimation for Rice (Oryza sativa L.) Using In-Field Spectroradiometric Measurements and Multispectral Satellite Images

“…However, as this spectral signatures suffer from suffering from having a mayor reflectance component coming from the soil (due to small leaf area). Measurements of this type are described in length in Sanchez-Galan et al [35,36]).…”