Study of RNN with Its CNN-Based Hybridization for Temporal Remote Sensing Data Processing to Map Rabi Crops

Two widely used deep learning architectures are Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs) that have demonstrated remarkable success in various domains. This research paper explores the individual strengths of RNNs and CNNs in processing temporal data and investigates the benefits of the hybridization with CNN. The concept of hybrid models has become more popular as those models are capable to handle temporal dependencies along with spatial dependencies. In this research work, we have incorporated a novel approach of hybridization of RNN and CNN into the field of specific crop mapping. We also explored the advantages of this hybrid model along with the limitations. Additionally, we compared the results of RNN model with the hybrid model in order to gain insights about the accuracy. Our results concluded that hybrid models outperformed in handling both temporal and spatial dependencies, and provided better classification results. Yet, the combination of different layers in the hybrid model and the parameters that were optimized has a significant impact on the overall performance. A comparative study has also been conducted between RNN and the hybrid model for the same classes, i.e., mustard and wheat. In order to show the effectiveness of the proposed approach, Kappa, F-Score, and overall accuracy have also been calculated to represent the accuracy of the proposed approach, where the overall accuracy has been improved from 90 to 93%.