Abstract

Assessing nutrition in the soil and correctly predicting crop yields play important roles in precision agriculture, made possible recently by ML, DL and IoT. This analysis examines how soil nitrogen, phosphorus and potassium (NPK) content is measured, with a special focus on wireless and handheld NPK sensors. We also study using attention-based deep learning networks and optimization to recommend crops more effectively. Experimental systems and relevant literature were thoroughly examined, looking at models designed to use data on soil, weather and seasonal aspects. To support crop recommendation and yield estimation, Random Forest, k-NN, SVM, Logistic Regression and DL techniques including GRU, RNN and hybrid structures were all assessed. Researchers also looked at using edge computing, steganography and federated learning (FL) to design secure, distributed technology for making predictions. Work was done on using data to manage water in three ways: Active Learning (IQ strategy), estimating yields through microwave sensing and making use of Explainable AI (XAI). Based on over 60 studies and prototypes, GRU and CNN have been identified as widely used DL models and adding ensemble techniques and better optimization methods can greatly improve prediction outcomes. We finish by sharing ideas for research that can help build better, scalable and affordable smart agriculture systems for places like India.