A new approach for crop type mapping in satellite images using hybrid deep capsule auto encoder

Abstract

The crop type prediction in high resolution satellite images is challenging due to the lack of ground truth labels. This reduces the crop prediction accuracy in satellite images. This work presents a new approach using hybrid deep capsule auto encoder for enhanced crop prediction in remote sensing images. Initially, the input images are pre-processed using a Refined Lee Filtering technique (RLF) to remove the speckle noise present in the images. Afterwards, spectral–spatial features with an extended morphological profile (EMP), extended attribute profiles (EAP), and hybrid wavelet features are extracted. Furthermore, a feature selection using a modified binary equilibrium optimizer (MBE) is presented to reduce the dimensionality of features. Finally, a hybrid deep capsule auto encoder maps the different crops in satellite images. Here, the adaptive atoms search optimization approach updates the optimized weights in the network. The implementation platform used in this work is PYTHON. The performance of the presented technique is examined with the Sentinal-2 dataset and Optical Radar dataset images. The experimental results of the presented technique provide improved performance than other existing approaches in terms of accuracy (98.78%), sensitivity (97.42%), F-score (97.62%), computational time, precision (95.09%), False Positive Rate (FPR) (0.83%), False Negative Rate (FNR) (2.58%), kappa coefficient (91.85%), and ROC curve with AUC is (0.98) for Sentinal-2 dataset and accuracy (98.57%) and kappa coefficient (91.85%) for optical Radar dataset.