Vegetation Species Determination Using Spectral Characteristics and Artificial Neural Network (SCANN)

Classification of vegetation according to their species composition is one of the most important tasks in the application of remote sensing in precision agriculture. To prepare an algorithm for such a mandate, there is a need for ground truth. Field operation is very costly and time consuming. Therefore, some other method must be developed, such as extracting information from the satellite images, which is comparatively cheaper and faster. In this study, we first introduced a simple method for Determination of the Vegetation Specie in full cover pixels (DVS) using their laboratory measured spectral reflectance curves. Then, based on these pixels, a hybrid method for vegetation field classification, which we call SCANN (Spectral Characteristics and Artificial Neural Network), is introduced. In this method, different vegetation spectral reflectance characteristics at the three extremes of green, red, and near-infrared along with an artificial neural network method were used. Comparing the results of DVS with those of field collected data showed near 100% accuracy. Based on the results of DVS, the results of SCANN showed an overall accuracy of more than 94%. This method is suggested for unsupervised classification using Hyperspectral images.