Modelling of energy productivity prediction systems of shallots classification growth phase system using convolutional neural network

Abstract

The energy productivity of agriculture is the effectiveness of energy consumed to carry out agricultural processes. In that case, energy prediction is needed in the future to achieve optimal agricultural yields for the farmers, make a strategy of the environmental health based on the energy field, especially non-renewable energy like fuel and consider the impact on the environment like pesticides, fertilizer and irrigation. This study used function modelling to predict energy productivity during shallot cultivation in Hutajulu, Sumatera Utara. The closest model equation is the Cobb-Douglas Model, a general model for predicting the production yield of a farm. Furthermore, a mathematical model was developed with Convolutional Neural Network (CNN). The system classified the image data based on the phases of shallot growth using CNN to process data in images. The model prediction used Ordinary Least Square (OLS) Regression to develop the Cobb-Douglas Mathematics model. It then continued with CNN modelling stages: Data Setup, Load Data, Preprocess Image, Model Development, Train Model, Model Evaluation, Test Model and Save Model. At the image classification stage, based on the growth phase, CNN was used with a sequential model with three hidden layers. The activation function used was ReLU and sigmoid to divide the class of growth phase into two based on binary digits (0.1). The energy productivity (Yi, kg MJ-1) model was lnYI = 4.177 - 0.0152lnX1 - 0.05lnX2 + 0.02lnX3 + 0.098lnX3 - 0.882lnX3 - 0.027 lnXer with an accuracy of 100% using Jupyter Notebook. The CNN model produces a percentage accuracy for the image classification of 95.3% with 100 epochs.