Monitoring crop conditions using machine learning techiques and IoT systems. Case study; Maize in Rwanda

Abstract

The internet of Things (IoT) with Artificial Intelligence (AI) promotes the development of precision agriculture. Maize is the major source of income in African continent, the main issue to the developing countries including Rwanda is inadequate harvest or low productivity due to lack of crops conditions and growth monitoring at all stages of maize crops which lead to food insecurity and famines in developing countries. Maize crops are affected by environment changes and infections in their growth and pests attack the crops and damage them; the farmers do not be aware of the problem earlier to take decision accordingly. However, using technologies like AI, IoT help in automating farming activities which removes the errors caused by manual agricultural tasks and help farmers to control crops environment such as Humidity, Pressure and Temperature, pests, and insects attack crops. The main aim of this research is to design and develop a prototype of IoT system for capturing maize crops images which that are analysed by ML techniques and detects whether the crops are infected, are health or attacked by pests to assist farmers in taking care of the crops in all stages for the purpose of increasing the productivity and fight for food insecurity. This work highlights the IoT systems and ML techniques that can be implemented to achieve mentioned goal. This research work used BMP280 sensor to sense the environmental crops field parameters (temperature, humidity, pressure), Raspberry Pi Camera was used to capture the crops images, and EfficientNet, Transfer Learning and InceptionV3 models of deep learning convolutional neural networks used to predict maize plant health status through image analysis/processing and classification. The predicted results shows that the transfer learning model performance achieved 79.7% accuracy from 50%, and its loss reduced from 100% to the 50%. InceptionV3 model of Neural Networks achieved the overall accuracy of 99.5% and loss of 2%, while EfficientNet model achieved the overall accuracy of 96% increased from74% and loss decreased from 83% to 13%. This research project was implemented in four stages including designing prototype, collecting data using developed prototype and dataset organization and predictive processing using ML techniques. In addition to crop growth monitoring and early detection of crops health conditions, the proposed solution will be playing great value in the process of mass collection of crops images data set to complement synthetic data