How m0dulo/pattern_recognition works
Project type and purpose
• This is a research-style, command‐line pipeline for surface-defect image analysis.
• It loads BMP images, extracts texture features using GLCM in MATLAB, then classifies them via (a) a MATLAB K-nearest‐neighbor script and (b) a PyTorch-based multilayer perceptron.
File structure
• dataset/ – raw image data (patches & scratches).
• features/ – CSVs holding extracted features (PCA, rescaling parameters).
• src/ –
– MATLAB scripts (.m) for feature extraction (get_glgm.m, glgm_norm.m, rescaling.m, write.m, load_all_features.m, main.m)
– MATLAB classification (knn.m + run_knn.bat) using config_knn.json
– Python classification (classification.py + run_mlp.bat) using config_mlp.json and PyTorch
– Utility scripts & configs (config_*.json)
• README, LICENSE, .gitignore – project metadata
Summarize README insights
• Core pipeline:
1. GLCM texture feature extraction
2. Classification via KNN & MLP (PyTorch)
• References for algorithms and optimizer (Adam)
Main components & interactions
A. Data Storage
– dataset/patches & dataset/scratches (input BMP files)
B. Feature Extraction Module (MATLAB)
– main.m calls get_features.m
– get_glgm.m computes co-occurrence matrix features
– glgm_norm.m normalizes them, rescaling.m applies scaling
– write.m & load_all_features.m aggregate results into features/*.csv
C. Configuration
– config_knn.json: KNN hyperparameters
– config_mlp.json: MLP architecture, learning rate, optimizer settings
D. Classification Module
1. MATLAB KNN branch
• knn.m reads features CSV + config_knn.json → predicts labels
• run_knn.bat launches MATLAB script
2. Python MLP branch
• classification.py (PyTorch) loads features CSV + config_mlp.json → trains/tests model
• run_mlp.bat launches Python script
E. External Dependencies
– MATLAB runtime/toolboxes
– Python interpreter + PyTorch
Architectural patterns & principles
• Pipeline pattern: clear, linear stages (ingest → extract → classify)
• Modularity: separate folders & scripts for each stage
• Config-driven: JSON files control algorithm parameters
• Dual-environment execution: MATLAB for extraction & KNN; Python/PyTorch for MLP