NS0720/pollen-svm-ifar: v1.0.0 - IFAR pipeline (pollen detection -> features -> SVM)

This release provides the full reproducible pipeline used in our IFAR submission for classifying 400× magnification pollen microscopy images. Pipeline overview: Detect circular pollen candidates (Hough + DBSCAN) and export coarse (x, y, r) metadata. Crop fixed 512×512 images centered at (x, y). (Manual step) Assign expert labels (type: cedar / cypre / dust / cedar_brst / cypre_brst). Estimate pollen center coordinates (cx, cy) from the largest contour on the 512×512 crop (robust to missing/failed internal signals). Extract 10 handcrafted features and z-score normalized features to create the final training table. Compute permutation importance for model interpretability. Validate that reducing the feature set (11 → 9) does not materially degrade performance. Evaluate performance using nested cross-validation (5×5) with out-of-fold predictions. An example dataset of ten 400× magnification images is included in pollen_picture_folder/ to reproduce the pipeline execution. For detailed instructions and expected outputs at each step, see README.md.