Adaptive Multi-Scale Gaussian-Laplacian Pyramid with Gabor Filtering for Microplastics Detection

Microplastics, defined as plastic particles with characteristic dimensions smaller than 5 mm, have emerged as a major environmental pollutant, posing significant threats to marine ecosystems, wildlife, and human health. Accurate detection and classification of microplastics in environmental samples are therefore essential for monitoring their spatial distribution. This study proposes an adaptive Gaussian-Laplacian pyramid-based framework for multi-scale image decomposition to enhance microplastic detection in microscopic images. Unlike conventional methods that use a fixed number of pyramid levels, the proposed approach dynamically selects the most informative level for each image based on quantitative metrics such as variance, edge energy, and entropy. This adaptive selection ensures optimal feature extraction at the most relevant scale, improving detection accuracy for small and variably shaped microplastics. Detection is performed using a sliding window approach combined with a Support Vector Machine (SVM) classifier, followed by Non-Maximum Suppression (NMS) to eliminate duplicate detections. The system was evaluated on 574 microscopic images, achieving high detection sensitivity with an accuracy of 0.93 on Level 2 with a radius of 15 pixels. A comparative analysis with recent studies demonstrates that the proposed method offers superior scalability and balanced detection performance, particularly for small objects.