Deep Learning-Based Crop Recommendation Using Soil and Environmental Parameters

Abstract

Crop identification is one of the priorities for increasing the productivity of agriculture, as well as the economic stability of the crops. Most farmers have traditional knowledge along with their experience, which leads to ineffective crop identification. In this context, this paper introduces a proper crop identification technique using deep learning techniques. The proposed system uses a Long Short-Term Memory network to determine the best crops according to the soil and environmental properties. The system is intended to be developed in two significant steps. At the beginning, the ANOVA test is used to determine the effect of the input variables on the system. Irrelevant variables reduce the efficiency of the system. Based on the test, the dominating properties of the crops, according to the environment, have been recognised as potassium, phosphorus, nitrogen, water, and rainfall. Then, in the second stage, the system is optimised based on several experiments involving variations in the number of layers, number of neurons, as well as the training epochs. The proposed approach provides promising insights for the development of an efficient decision-support system that assists the farmer in the selection of suitable crops.