ParticleSeg3D: a scalable out-of-the-box deep learning segmentation solution for individual particle characterization from micro CT images in mineral processing and recycling
Minerals, metals, and plastics are indispensable for a modern society. Yet, their limited supply necessitates optimized extraction and recycling processes, which must be meticulously adapted to the material properties. Current imaging approaches perform material analysis on crushed particles imaged...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article (Journal) |
| Language: | English |
| Published: |
1 February 2024
|
| In: |
Powder technology
Year: 2024, Volume: 434, Pages: 1-13 |
| ISSN: | 0032-5910 |
| DOI: | 10.1016/j.powtec.2023.119286 |
| Online Access: | Verlag, lizenzpflichtig, Volltext: https://doi.org/10.1016/j.powtec.2023.119286 Verlag, lizenzpflichtig, Volltext: https://www.sciencedirect.com/science/article/pii/S0032591023010690 
|
| Author Notes: | Karol Gotkowski, Shuvam Gupta, Jose R.A. Godinho, Camila G.S. Tochtrop, Klaus H. Maier-Hein, Fabian Isensee |
| Summary: | Minerals, metals, and plastics are indispensable for a modern society. Yet, their limited supply necessitates optimized extraction and recycling processes, which must be meticulously adapted to the material properties. Current imaging approaches perform material analysis on crushed particles imaged with computed tomography (CT) using segmentation and mass characterization. However, their inability to reliably separate touching particles and need to annotate and retrain on new images, leaves untapped potential. By contrast, particle-level characterization unlocks better understanding of particle properties such as mass, appearance and structure. Here, we propose ParticleSeg3D, an instance segmentation method for particle-level characterization with strongly varying properties from CT images. Our approach is based on the powerful nnU-Net, introduces a particle size normalization, employs a border-core representation, and is trained with a diverse dataset. We demonstrate that ParticleSeg3D can be applied out-of-the-box to a large variety of materials without retraining, including materials and properties not present during training. |
|---|---|
| Item Description: | Gesehen am 04.11.2024 |
| Physical Description: | Online Resource |
| ISSN: | 0032-5910 |
| DOI: | 10.1016/j.powtec.2023.119286 |