A Survey Of Recent Deep Learning Algorithms Used In Smart Farming
Lavika Goel, Aishwarya Mishra
Abstract
Agriculture is one of the dominant pillars of the economy for any nation, the prediction of agricultural yield at an early stage helps us to plan, control, and monitor supply chain management of related commodities. In this study, we have investigated many proposed methods to understand the best practices for crop yield prediction (CYP). We also draw conclusions regarding which factors are more effective and contribute to CYP in smart farming. The factors considered in the study are soil property, humidity, temperature, LAI, NDVI, EVI, NIRV, etc. These factors are used as inputs to intelligent techniques such as Neural Networks, Deep Learning, variants of deep learning like CNN, RNN, DQNN, LSTMs, Boosting, Bagging, k-NN, Random Forest, SVM, etc. for CYP. Statistical techniques have also been studied in this paper in detail for their use in CYP. From the analytical study, it is observed that those techniques which used the above-stated factors for CYP have greater accuracy of 90 % −99 %. It is also observed that CYP becomes difficult with all varieties of crops, the approaches which targeted specific crops were more accurate in predictions whereas generalized approaches were less accurate. We also studied different intelligent techniques with respect to a wide range of criteria like data source distribution, method of prediction for CYP, the accuracy of prediction range, and evaluation parameters considered for a given crop. The highest accuracy reported is 95.64% which uses deep learning techniques along with phenological information for CYP.