A Cost-Effective Prototype for Multimodal IoT-Based Agriculture Monitoring

IoT sensors are deployed on the field for various agriculture data mining tasks. Some of these tasks include automated plant species classification, weed detection, and pest identification, which allow agriculturalists to make faster decisions about corrective measures to take for maximized crop quality and quantity. As the availability and diversity of these sensors grow, so will the opportunities to automate more complex tasks. This will result in a deluge of agricultural time series and image data. Simultaneously, studies regarding low-cost IoT implementations for agriculture monitoring are at their infancy. To address the diversity of agriculture data and their increasing deployment costs, we propose an in-situ agriculture monitoring system that integrates a microcontroller camera paired with temperature, humidity, gas, and GPS sensors while also including embedded computing devices for GPU-enabled data transfer. The GPU-enabled data transfer functionality is critical for the system to receive and store agriculture time series and images at low-bit rates to stably run on web servers while minimizing energy and communication bandwidth constraints. This functionality also allows developers the option of deploying AI models including but not limited to any of the abovementioned agriculture data mining tasks. Compared to previous works in literature, this system provides a relatively cost-effective prototype for multimodal IoT-based agriculture monitoring in addition to more extensive hardware and software details accessible to users.