Project Biocycling

Detecting contaminants in compost fertilizers.

Paper: Detection of Foreign Objects in Compost Fertilizer

This is a collection of datasets containing images of compost fertilizer or digestate that has been contamiated with small foreign objects, such as pieces of plastic foils, shards of hardplastic, pieces of glass, aluminum, etc.

The images always contain a background material of sieves and clean compost fertilizer or digestate. Taken directly from a variety of composting plants in switzerland. On top of this material a known set of foreign objects has been placed (sometimes partially covered by the background material).

For each sample images two images were taken. First, a hyperspectral image in the near-infrared range (Specim FX-17) and a RGB-image in the visual spectrum with a 6k Baumer RGB-Camera. The images are spatially aligned and can be mapped onto each other.

This dataset allows to train object detection or image segmentation models to detect foreign materials in compost fertilizer. Additionally transfer learning from one modality to the other is possible.