Resource-Constrained Intelligent Trap: Fruit Flies Surveillance Framework with TinyML Integration

• Main Authors: Nguyen, Quan Minh, Le, Vu Thanh, Lai, Minh Nhat, Vo, Hien Bich
• Format: Conference Proceeding
• Language: English
• Subjects: AIoT; Crops; edge computing; Electron traps; EmbeddedML; ESP32; FOMO; fruit fly; Productivity; Random access memory; Spraying; Surveillance; Tiny machine learning; TinyML
• ISSN: 2836-4392
• DOI: 10.1109/ICCE62051.2024.10634657

The threat posed by fruit flies to fruit crops requires innovative solutions to protect the harvest. Our research introduces an advanced solution to safeguard fruit crops from the menace of fruit flies using the ESP-EYE microcontroller and the FOMO algorithm with Edge Impulse AI Platform. Our intelligent insect trap is equipped with advanced features and demonstrates excellent performance. With a maximum RAM usage of 2.4Mb and an inference time of 5.694 seconds, the device achieves an impressive 96 percent accuracy based on the Intersection over the Union threshold (IoU) of 0.5. This effective solution not only ensures professional detection and count capabilities but also provides timely notifications to farmers about potential threats. It enables the use of targeted pesticides use in regions with high fruit flies, minimizing widespread spraying, and ensuring efficient crop protection. Furthermore, the trap runs smoothly without human intervention, thanks to its automatic replacement feature for fly-stained adhesive traps. This sustainable approach, adaptable to various types of pests, improves resource utilization and productivity, underscoring its significant value in agriculture and other related fields.