Master Thesis: Determination of Particle Size Distributions through Real-Time Area, Mass, and Geometry Flow Characterization of Non-Ferrous Metals: A YOLO-Based Instance Segmentation Approach

The task included three parts: annotation, YOLO instance segmentation and particle weight regression

Dataset

In the dataset folder include:

• copper: the copper particle sample named with different size fraction as batch
• minerals: the mineral particle sample named with different size fraction as batch

Usage

Morphology operation

• morphology_processing.py: use basic mathematical operation, such as thresholding, closing and opening, to annotate particle images. Their annotated dataset were stored in the dataset_morph folder.

YOLO pipeline

In the yolo folder include:

• image_tiling.py: tiling the images to 512 by 512 dimension
• combined_images.py: combine both minerals_tiling and copper_tiling folder to one images folder
• sam2_mask_generator.py: particle annotation were stored in the dataset folder named copper and minerals
• SAM2_label.py: label each class and contour coordinate of particle images with yolo segmentation format
• dataset_split.py: split the dataset to train:val:test = 7:2:1 to folders train, val and test
• yolov5.py, yolov8.py, yolov9.py, yolov11.py, yolov12.py: instance segmentation model training

Regression pipeline

Direct run the notebook regression.ipynb in the regression folder

Dashboard App

In the dash folder:

• app.py: Integrate both YOLO pipeline and regression pipeline to show the particle detection, inference time (as metadata) and predicted weight dashboard in real-time.

---

Clone the repository

git clone https://git.rwth-aachen.de/Poyen-Chen/master_thesis.git

Add your files

cd existing_repo
git remote add origin https://git.rwth-aachen.de/Poyen-Chen/master_thesis.git
git branch -M main
git push -uf origin main