MicroPolluScan: A Deep Learning-Based Monitoring System for Classification of Microplastic Contamination in Fishponds

Microplastics in aquatic life are hazardous to sea life, food safety, and aquaculture. In this paper, we describe MicroPolluScan, an AI-based deep learning system which uses automated microscopic image analysis to detect, classify, and quantify microplastic contamination in fishponds. To facilitate the development, we employed an Agile Scrum–CRISP-DM paradigm, that is an architectural process based on an iterative software design methodology, along with structured data science methodology. Microscopic pictures of beads, fragments, and fibers were taken from fishponds in Lucena City, augmented with Roboflow, and trained in YOLOv5 and YOLOv8 networks on Google Colab. Comparison results demonstrated that YOLOv8 provided the best performance (precision = 0.82, recall = 0.85, F1 = 0.83, mAP = 0.89) because of the anchor-free detection head and C2f supporting mechanism. A fine-tuned model was implemented in a Flask-based web system, for real-time detection and visualization dashboards for practitioners of aquaculture. Software quality of the software evaluated following ISO/IEC 25010 has a mean of 3.80 (Strongly Agree) followed by a Cronbach’s α of 0.89 indicating instrument stability and system usability. Our results show that by combining computer vision and web technologies microplastic monitoring can be automated leading to rapid decision making. We envisage future research projects to provide more diversity on the dataset, incorporating IoT-based sensors, and applying explainable-AI models for improved transparency and ground application.