Proof of Concept: A TinyML-Based Image Classifier for Detecting Microplastics and Waste in Simulated Marine Environments

This study presents the development of an image classifier designed to identify microplastics and cardboard waste, with potential application in marine environments, using Tiny Machine Learning (TinyML) implemented on an ESP32-CAM module. A total of 229 images were collected, classified into mi-croplastics and non-microplastics, and preprocessing techniques such as scaling (to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$96\times 96$</tex> pixels) and normalization were applied to optimize the performance of the chosen model. The learning technique used was Transfer Learning, employing a model based on the MobileNetV1 architecture. The model was trained and validated, achieving an accuracy of 91.7% on the validation set under controlled conditions. This implementation demonstrates the initial feasibility of using TinyML on resource-constrained devices for plastic waste detection, laying the groundwork for future testing in real marine environments. The study provides a promising foundation for the development of environmental monitoring tools, although additional validation in field conditions is required to confirm its effectiveness in real aquatic ecosystems.