There are two main problems with finding mi croplastics in environmental samples: (1) telling plastic particles apart from organic matter and (2) getting results that domain experts can understand and check. Existing methods are very accurate, but they don’t explain how they work and often don’t make a distinction between plastic and non-plastic. We put forward a two-stage framework that is easy to understand and combines YOLOv8 object detection with morphological feature extraction for pre-screening tasks. We explicitly include Algae as a negative class, which is different from previous work. This is because there is a lot of confusion in the real world between clear plastics and biological material. Our system gets 87.5% recall and 91.7% precision on a 4-class dataset. This is 19.5 points better than image-level baselines. An analysis of feature importance shows that the most useful geometric descriptors are equivalent diameter, eccentricity, and convex hull ratio. Our framework makes it possible to do cost-effective pre-screening that cuts down on FTIR testing by 60% while keeping the false negative rate below 13%. It does this by giving contamination index quantification and interpretable feature-based decisions.