Vegetation Cover Classification from Satellite Images using AdaGrad with Convolutional Neural Network

Effective approaches for the vegetation-cover mapping are important for the management practices and the classification purposes. The satellites images are captured through the aerial have significant potential to support the application in various fields, particularly in robotics-related resolutions for keen urban planning. However, some challenges have been considered in this research, like inadequate model variance and constrained simplification capabilities. Hence, this research proposes the optimization function of AdaGrad with Convolutional Neural Network (AdaGrad-CNN) approach for the vegetation cover classification by using satellite images. Initially, this research collected the two standard datasets such as Sentinel-2 MultiSpectral Instrument (MSI) and NITRDrone to test and validate the proposed method. The sens2Cor approach is used for the preprocessing to remove the noises in the satellite images. The experimental results represented that the proposed AdaGrad-CNN approach attains the better overall accuracy of 98.21% and Kappa of 0.972 on Sentinel-2 MSI dataset as compared to the existing method like U-Net.