Circular Chromatography of Pfeiffer – Soil Quality Analysis with Computer Vision

This repository gathers datasets and scripts related to the project on soil quality analysis using Pfeiffer’s circular chromatography combined with computer vision techniques. The main objective is to provide a structured database along with preprocessing and data augmentation tools, enabling the application of automated analysis methods such as digital morphometry, chromatographic pattern classification, and neural network–based segmentation. The repository is organized into two main sections: Image datasets – available at different stages of preparation (raw, renamed, processed, augmented, and segmented). Python scripts – designed for automating tasks such as renaming, border cropping, data augmentation, and segmentation with machine learning. Datasets 1. Raw images of Pfeiffer’s circular chromatography This dataset contains the original images obtained from the circular chromatography of soil samples collected in different locations and dates. Format: JPG Quantity: 35 images Purpose: Initial basis for visual inspection and computational analysis. 2. Renamed images of Pfeiffer’s circular chromatography The raw images were renamed to standardize organization and facilitate association with collection metadata. Format: JPG Quantity: 35 images Purpose: Improve traceability between samples and field data. 3. Processed images of Pfeiffer’s circular chromatography Preprocessed images with outer borders removed, keeping only the relevant central chromatographic region. Format: JPG Quantity: 35 images Objective: Eliminate peripheral noise and standardize the area of interest. 4. Augmented images of Pfeiffer’s circular chromatography Images generated through data augmentation techniques to expand the dataset for training computer vision models. Applied transformations: Rotations (90°, 180°, 270°) Horizontal and vertical flips Brightness variations Contrast variations Format: JPG Quantity: 350 augmented images 5. Segmented images of Pfeiffer’s circular chromatography Images resulting from automatic segmentation, isolating the central chromatographic region. Format: JPG Quantity: 350 images Purpose: Morphological and statistical analysis, and training of classification algorithms. Python Scripts Renaming script Automates the renaming of images to ensure standardized naming conventions based on predefined criteria. Batch processing of .jpg / .png files Application of standardized names Facilitates association with collection metadata Border cropping script Automatically removes outer borders of chromatographic images, keeping only the central region. Symmetrical cropping of all borders Batch image processing Automatic saving in the destination directory Data augmentation script Applies data augmentation transformations to expand the dataset. Rotations, flips, brightness and contrast adjustments Automatic folder processing Generation and saving of multiple variations per image Segmentation script with U²-Net and Optuna Performs automatic segmentation of chromatographic images using U²-Net, with hyperparameter optimization supported by Optuna. Adjustable parameters: resizing, blur, threshold, Canny edge detection Mask generation for segmented images Evaluation through IoU (Intersection over Union) Saving of best parameter configurations in JSON format Requirements: Python 3.7+ OpenCV, Pillow, PyTorch, Optuna Pre-trained U²-Net model (u2net_portrait.pth), available at:https://github.com/xuebinqin/U-2-Net/tree/master Overall, this repository integrates datasets and code, enabling the workflow from raw image acquisition to advanced preprocessing, augmentation, and segmentation. It provides a foundation for studies on classification, segmentation, and morphometric analysis of soil quality using Pfeiffer’s circular chromatography.