Quantification of Soil Properties from Hyperspectral Data for Sustainable Agriculture Using Deep Learning

The characterization of soil properties is critical for optimizing farming for sustainable agriculture. All the existing techniques for soil quantification do not take advantage of the sequential nature of Hyperspectral Data. This work focuses on proposing a Hybrid Framework that can quantitatively assess the soil properties from Hyperspectral data by extracting the essential features via Principal Component Analysis and Locality Preserving Projections. The extracted features are combined to form the Hybrid dataset which is then given as input to Long Short-Term Memory Networks, a deep learning-based framework which is typically used for sequential problems. The effectiveness of the Hybrid Framework is shown by comparing it with the existing regression models.