Mobile Real-Time Grasshopper Detection and Data Aggregation Framework.

Area of Science:

• Agricultural Entomology
• Computer Vision
• Machine Learning

Background:

• Grasshoppers (Orthoptera: Acrididae) are significant global pests, causing widespread crop and ecosystem damage.
• Current manual detection methods for grasshoppers are labor-intensive, challenging, and lack scalability.
• There is a need for automated systems to improve the efficiency and accuracy of grasshopper scouting.

Purpose of the Study:

• To introduce MAESTRO, a novel deep learning framework for automated grasshopper detection.
• To develop a system deployable on both desktop and edge devices, including smartphones.
• To present a new open dataset (GHCID) for grasshopper detection research.

Main Methods:

• MAESTRO utilizes a state-of-the-art two-stage deep learning approach for insect detection in RGB images.
• The framework is designed for deployment on various platforms, including edge devices without internet connectivity.
• A new dataset, GHCID, comprising diverse grasshopper populations was created for training and validation.

Main Results:

• The stationary MAESTRO system achieved a detection performance of 78% with an analysis time of approximately 1000ms per image.
• The mobile application version of MAESTRO demonstrated a detection performance of 49% with a faster analysis time of around 400ms per image.
• The data-driven algorithms show potential for adaptation to other agricultural detection tasks, such as plant disease identification.

Conclusions:

• MAESTRO offers a promising automated solution for grasshopper detection, enhancing scouting productivity and data fidelity.
• The framework's adaptability to edge devices and potential for broader agricultural applications highlight its significance.
• This technology can significantly contribute to more effective control strategies for this critical global pest.