Edge AI Pest Detector

The Pest Detector Demonstrator showcases how edge AI can revolutionise agricultural pest management. Conventional pest control relies on manual scouting or cloud-based image analysis, both of which incur delays and high costs. Researchers recently developed a portable IoT device integrating a Tiny-LiteNet convolutional neural network and a Raspberry Pi 5 for on-device processing. The system uses a high-definition camera to capture plant images and processes them locally to detect pests and diseases with 98.6 % accuracy and 80 ms inference time. The model fits in 1.2 MB and has 1.48 million parameters, demonstrating the feasibility of running sophisticated AI on resource-constrained hardware. Low power consumption enables farmers to deploy the detector in remote fields where connectivity and electricity are scarce. Data can be transmitted via GSM/GPRS for cloud storage but only after local inference, reducing bandwidth usage. SmartLight’s demonstrator builds on this research. We integrate the Tiny-LiteNet model into an ESP32-S3 microcontroller, further reducing power draw. The device runs from a small solar panel and monitors crops for signs of pests such as aphids and leaf miners. When a pest is detected, the system triggers an audio or LED alert and logs the event. Farmers receive periodic summaries via Wi-Fi HaLow or LoRaWAN. Field trials show that early detection allows targeted pesticide application, reducing chemical use and protecting beneficial insects. We envisage deploying networks of these detectors to create pest-risk heatmaps for farms.