A MACHINE LEARNING AND INTERNET OF THINGS-BASED SMART IRRIGATION SYSTEM

Precision irrigation is essential for sustainable agriculture as water resources decrease. Our intelligent irrigation architecture has been designed to provide crops with accurate and optimal watering. A system makes utilization of a heterogeneous sensor network that is spread throughout an experimental 100-acre maize field and includes temperature sensors, rain gauges with tipping buckets and soil moisture probes. Using Arduino microcontrollers, analogue signals from the census are converted to digital signals, after which sent over Lora WAN to our cloud IoT platform. Neural network models with LSTM that are optimized on a GPU compute cluster and implemented in Python are used to integrate sensor data streams with satellite data and weather forecasts. To anticipate the daily crop water requirement (CWR), the sensor fusion LSTM models are trained using five years' worth of historical ground-truth data from the site. . Actuators that are automated and linked to water distribution lines can open or close valves to precisely regulate irrigation according to the estimated CWR. We have developed a reacts web dashboard that enables farmers to view sensor feeds, manually adjust valves, and verify model recommendations in order to fine-tune the AI models on go. Every week, the AI system retrains itself using streaming data to accommodate for unforeseen weather disturbances. Our intelligent architecture that is cloud-connected guarantees scalability among farms and offers growers interpretability via an interactive online interface. In future, our automated process will enable the simultaneous, precise supply of fertiliser, herbicides, and water for sustainable agriculture.