Development of a Raspberry Pi Based Rain Predictor for Smart Retractable Roof Systems using Convolutional Neural Networks: A Study Based on Daylight Conditions

Abstract

Technological advancements have made it possible to develop intelligent retractable roofs. These are used widely in home balconies, soccer stadiums, and smart greenhouses, to mention a few. However, the current retractable roof systems are fitted with rain sensors that need to be in direct contact with the rain droplets before they can send a signal to the control circuit to close the retractable roof. With such sensors, when the rainfall rate is high, they may be disturbing to property supposedly protected by the retractable roof when the rain is detected. For instance, nurturing of the crops in greenhouses is controlled by adhering to the required amounts of the soil nutrients and water concentration. So, the current sensors for rain detection that are installed in smart retractable roof systems for greenhouses may allow excess water concentrations than required as they need to be in direct contact with the rain droplets before the roof can close. The purpose of this study is to develop a Raspberry Pi camera-based system using Convolutional Neural Networks (CNN) to predict the possibility of rainfall by examining the cloud condition. The advantage of this method is that the rain is detected early before it falls hence preventing monitored environments from rainfall damages. The custom CNN model was trained on images of the clouds with two classes. The first class was formed by images of clear clouds while the second class was formed by grey to dark clouds that form before the rainfall. The model's performance was improved by fine tuning it's learning to a performance that reduced misclassifications while increasing the classification accuracy from 84% (at a learning rate of 0.01) to 97% (at a learning rate of 0.00001).