Agricultural Land Detection using Deep Learning Algorithms

The introduction of remote sensing technologies has made it possible to extract meaningful information from satellite images. One such application that is essential in managing natural disasters, estimating crop yields, and managing land management is agricultural land detection. This study compares Convolutional Neural Networks (CNN) and Autoencoders (AE) as two deep learning methods for detecting agricultural land in satellite images. Preparing the training and testing datasets and preprocessing the satellite images comes first. The CNN and AE models are then trained and tested on the datasets, and their performance is assessed using metrics like accuracy, recall, and F1-score. A qualitative study of the output images provided by both algorithms are analysed. According to findings, both the CNN and AE models are capable of accurately identifying agricultural land from satellite images, while the CNN model does so marginally more efficiently than the AE model. Both models were able to capture the key characteristics of agricultural land since their output images had a visually comparable appearance. Proposed work reveals that deep learning methods like CNN and AE may be employed well for satellite image-based agricultural land detection. Although both models are effective, the CNN model outperforms the AE model only barely. Proposed findings may be helpful in enhancing agricultural yield predictions, land management strategies, and disaster preparedness strategies.