MPSDYF: Design of an augmented Multimodal technique for Prediction of Soil quality to analyze crop Diseases, Yield and recommend Fertilization operations

Prediction of crop diseases, their yield, and recommendation of fertilization for improving the yield is a multidomain task that involves collective correlation of data samples, feature extraction & selection, classification of selected features into different yield types, post-processing for recommendations generation, and feedback-based continuous optimizations. Existing models that perform these tasks either have limited accuracy, or cannot be scaled for real-time heterogenous soil types. To overcome these issues, this text proposes design of an augmented multimodal technique for prediction of soil quality to analyze crop diseases, yield and recommend fertilization operations. The proposed model uses an augmented multimodal approach that combines multiple data sources, including soil properties, weather conditions, crop information, and satellite images. Analysis of the data samples is done via a combination of Vector Autoregressive Moving Average (VARMA) with Long-Short-Term Memory (LSTM) & Gated Recurrent Unit (GRU) to provide recommendations for fertilization operations based on soil quality and crop needs (or requirements). The proposed technique initially converts collected samples into 1D flattened vectors, and extracts colour, shape and texture feature sets. These feature sets are processed via the proposed fusion of VARMA LSTM & GRU processes. To validate performance of the model, it was tested on multiple datasets, and achieved 3.5% higher accuracy, 2.9% higher precision, 2.4% higher recall, and 8.3% lower delay in predicting soil quality and analysing its impact on crop diseases and yield levels. The results of this study can help farmers and agriculture practitioners make informed decisions about soil management and fertilization operations, leading to improved crop productivity and sustainability levels.