Signal Assessment Using ML for Evaluation of WSN Framework in Greenhouse Monitoring

Background and Objective: The deployment of a Wireless Sensor Network (WSN) provides a useful aid for monitoring greenhouse-like environments. WSN helps in achieving precision agriculture i.e. more yield can be produced with precise inputs. Before the deployment of a sensor network, it is necessary to explore the communication range of nodes. Communication signals are affected by losses due to stems, fruits, twigs, leaves, infrastructure material, etc. in a greenhouse. So as part of the deployment strategy, signal assessment is required in the greenhouse. Methods: This research work proposes a Machine Learning (ML) based signal assessment for the evaluation of WSN deployment in different structures of a tomato greenhouse. Signal strength is measured for a naturally ventilated greenhouse and a fan-pad ventilated greenhouse. Measurements for the naturally ventilated greenhouse are considered with two case scenarios i.e. with transmitter and receiver in the same lane and with transmitter and receiver in different lanes. Models are developed for measured values and evaluated in terms of correlation and error between measured and model formulated values. Results and Conclusion: For the naturally ventilated greenhouse case scenario 1, correlation increases from 91.83% to 95.42% as the degree increases from 2 to 7. Correlation for naturally ventilated greenhouse case scenario 2 rises from 72.51% at degree 2 to 90.09% at degree 10. For the fan-pad ventilated greenhouse, the model has a more complex fitting because of the spatial variability within the greenhouse. Correlation of the model increases from 79.39% to 84.06 % with an increase in degree from 2 to 11. For the naturally ventilated greenhouse, better correlation is achieved at lower degrees compared to the fan-pad ventilated greenhouse.