Enhanced Crop Classification Through Satellite-Image Processing

Crop classification using satellite imagery is pivotal for effective crop monitoring and management, providing essential insights for agricultural policy planning and optimizing water resources critical for crop growth and yield. Traditionally, crop classification has involved manually extracting vegetation indices and spectral data from satellite images, which are subsequently used in machine learning models. While these traditional methods have proven beneficial, they are often labor-intensive and limited to specific regions of interest. Recent advancements in deep learning have transformed the remote sensing landscape, offering more efficient and accurate methods for crop classification. Innovative approaches now integrate various deep learning spatial and temporal models to enhance classification performance. However, the quality of the satellite image is crucial in the robust performance of such models. Thus, in this study, an image enhancement technique, Deep Image Prior, is leveraged to enhance the quality of the satellite images. These enhanced images are then used to develop a crop classification model consisting of a convolutional autoencoder to extract the spatial features and a long short-term memory network to handle the temporal dynamics. From the results, it could be seen that image enhancement greatly improves the model’s performance.