Superpixel-based plastic litter detection in UAV hyperspectral imaging using spectral-textural features

Plastic pollution has reached alarming levels worldwide, requiring innovative solutions for accurate detection and monitoring. Automated plastic litter detection using hyperspectral imagery acquired by unmanned aerial vehicles (UAVs) has become critical for environmental monitoring. This paper presents a novel spectral-textural machine learning framework that integrates super-pixel segmentation with spectral and textural feature extraction to improve the accuracy and efficiency of plastic litter detection. The workflow uses Simple Linear Iterative Clustering (SLIC) for preliminary segmentation, followed by the extraction and integration of statistical spectral features and several textural features derived from the Gray Level Co-occurrence Matrix (GLCM). Validated on a real shortwave infrared dataset, the proposed methodology outperforms single feature and pixel-based approaches, demonstrating significant practical applicability for plastic debris detection. To interpret the results, feature importance ranking and SHapley Additive exPlanations (SHAP) analysis were used to identify the most influential features driving plastic predictions. This research provides a robust, scalable solution for environmental monitoring that addresses the limitations of traditional methods and satellite-based techniques.