Hyphae_segmentation

Image quality feature extraction, using them for shallow learning to train 2 different classifiers (good vs. bad images; semi-good vs. good image) to separate images into 3 groups. Use group specific parameters to segment and count hyphae.

Setup

Currently, we have hard-coded 2 classifiers and their associated parameters for image-processing. If experimental setup requires different setting --> changes need to be made in the binary_train.py and binary_process.py

process_batch.py is also experimental design (folder structure) specific

Change config.ini for current use:

[PATHS]
# Base directories
base_dir = /PATH/TO

# Training
training_data = %(base_dir)s/data/trainig_set/2025-10-29
model_output = %(base_dir)s/results/classifiers/train_2025-10-29

# Processing (for later use)
data_dir = /PATH/TO/data/
results_dir = %(base_dir)s/results/2025-10-23

# Feature extraction
feature_dir = %(base_dir)s/results/features

# Models to use for processing
bad_model = %(model_output)s/bad_detector_model.joblib
semi_model = %(model_output)s/semi_good_detector_model.joblib

[FEATURE_EXTRACTION]
# Image preprocessing
resize_height = 1440
resize_width = 1440

# Multi-scale parameters (comma-separated)
gaussian_scales = 0.7, 1.0, 1.6, 3.5, 5.0, 10.0
threshold_factors = 0.8, 1.0, 1.2

[CLASSIFIER]
# Random Forest parameters
n_estimators = 200
max_depth = 10
min_samples_split = 2
min_samples_leaf = 1
max_features = sqrt
random_state = 42
test_size = 0.2

feature_extraction.py is good for general re-use to extract features for image classifiers training.

Usage:
    python feature_extraction.py <input_folder> <output_file>

Example:
    python feature_extraction.py ./data/images ./features/extracted_features.tsv