Design and prototyping of environmental conservation system based on embedded machine learning for precision farming

Abstract

Most of the existing precision agriculture solutions recommend the use of fertilizers as a remedy to poor soil fertility. Such solutions cause environmental degradation in the long run mainly due to the overuse of fertilizers. There is, therefore, a need for a system to ensure that farmers can practice precision farming in terms of a sustainable soil management approach to attain high yields while at the same time conserving the environment. In this research, a design and prototype of an embedded machine learning-based system to predict the best crop to grow with minimal use of fertilizers to conserve the environment is presented. The system senses different real-time soil parameters daily, integrates them with forecast weather information and uses embedded machine learning techniques to determine which crop would grow best under the existing conditions with minimal use of fertilizers. In addition to crop prediction, the system helps farmers to monitor the nutrient evolution of the soil so that action can be done in real time. The results are either displayed on the device or sent to the farmer’s mobile phone. This is a move from the existing solutions that depend on cloud analytics and do not consider the change of soil conditions overtime in making the predictions and decisions. The prototype was tested at STES Group in Rwanda, an innovation and start-ups support hub that provides a commercial smart farming system. The data collected was hosted on a virtual cloud provided by STES so that data can be stored for future use. The implementation of the proposed solution is expected to not only lead to high productivity and reduced costs but also conserve the environment.