Insect Image Dataset for TinyML: Locust, Grasshopper, and Other Classes

This dataset provides labeled images for insect classification in a TinyML and edge AI context. It is designed to support the development and evaluation of lightweight deep learning models for real-time deployment on resource-constrained microcontrollers such as ESP32.

The dataset is composed of three classes:(1) Locust,(2) Grasshopper,(3) Other (non-target objects, background scenes, or ambiguous samples).

Images were collected and curated from multiple public sources and real-world agricultural environments, followed by cleaning and labeling to ensure consistency and relevance for embedded vision tasks. The "Other" class includes challenging samples such as partial occlusions, motion blur, and visually similar non-target objects, improving model robustness.

The dataset is suitable for classification and lightweight object detection tasks, particularly in smart agriculture applications such as pest monitoring and early infestation detection.

It is intended for use in TinyML workflows, including model training, quantization (e.g., INT8), and deployment on low-power edge devices.

Keywords: TinyML, insect classification, edge AI, embedded systems, agriculture, pest detection.