Crop yield predictive modeling using optimized deep convolutional neural network: An automated crop management system

In agriculture, AI devices have developed tools that help farmers in achieving exact and controlled cultivation. Besides, it helps farmers with suitable assistance in auspicious harvesting, frequent change of crops, water management, kind of crops to be grown-up, optimum plantation, pest attack, and nourishment. To meet the requirements of the farmers, enhance yield production, and administration of agricultural activities, the proper method to predict crop yield is more significant. This paper proposes a new crop yield prediction method as part of the crop management system using deep learning (DL) models. The suggested model has three stages: "pre-processing, feature extraction, and yield prediction". Initially, data cleaning is done in the pre-processing stage and then the features like “soil parameters, Metrological Parameters and statistical and higher order statistical features” are determined. Here the soil parameters include “Soil Temperature (Morning, Afternoon, and Evening), Soil Moisture (Morning, Afternoon, Evening), Nitrogen (N) Phosphorus (P), Potassium (K), Water Holding capacity of the soil, Bulk density, Porosity. The Metrological Parameters include Air Temperature (Morning, Afternoon, Evening), Humidity (Morning, Afternoon, Evening), Light (Temperature (Morning, Afternoon, Evening), Precipitation (Morning, Afternoon, Evening), Wind speed (Morning, Afternoon, Evening), Dew point (Morning, Afternoon, Evening), Air pressure (Morning, Afternoon, Evening). maximum temperature and minimum temperature”. Later these features are specified to an “Optimized Deep Convolutional Neural Network (DCNN)" that gives the prediction of crop yield results with known information. Especially, to enhance the prediction accuracy of the classifier, the weights of DCNN are fine-tuned via Self Adaptive Whale Optimization Algorithm (SA-WOA). Finally, the proposed model is observed as superior to the traditional modellingerror analysis. However, the MAE value of the proposed model is 0.247, when the learning rate is 70.