geospectral 0.2.3

A Python package for augmenting geospatial raster data (GeoTIFFs) with various techniques including noise addition, patch mixing, and geometric transformations.

Geospectral

Overview

GeoSpectral is a high-performance multispectral image augmentation library designed for efficient geospatial data processing.

Features

• Raster Clipping: Extract regions from large GeoTIFF files using shapefile boundaries
• Multithreaded Support: Choose between threaded and non-threaded operations
• Performance Optimized:
  • Non-threaded clipping: ~7 seconds
  • Multithreaded clipping: ~2 seconds (4x faster)
• Automatic Worker Detection: Intelligent worker allocation based on system resources, or manual configuration

Getting Started

Import Packages

# Import Packages
from augmentation import clip_raster_threaded,clip_raster_serial
shape="data.shp"
tiff="data.tif"

)

Non-threaded clipping:

# Basic usage example
clip_raster_serial(
    shapefile_path=shape,
    tiff_file=tiff,
    output_dir="tiles",

)

Multithreaded clipping:

# Basic usage example
clip_raster_threaded(
    shapefile_path=shape,
    tiff_file=tiff,
    output_dir="tiles",

)

## Load filepaths after clipping

```python
filepath,filename=files_load("tiles")
print(filename)

This block of code will help to load all the absolute filepaths. It returns a tuple file location and file path.

Applying Augmentations

geometric_augmentations(
    image_path=filepath[0],
    aug_type=[],
    apply_all=True,
    noise=0.05,
    patchmix=True
)

This function provides a lot of features ranging from just simple geometric augmentation to a bit advanced features like adding gaussian noise ,applying patchmix .

Geometric Augmentations

This applies selected image transformations using NumPy:

• H → Horizontal flip
• V → Vertical flip
• R → Rotate 90°
• R2 → Rotate 180°
• R3 → Rotate 270°

You can pass specific augmentations in a list using aug_type (e.g., ["H", "R"]) and set apply_all=False to apply only those operations.

Set apply_all=True to apply all augmentations.

---

Gaussian Noise Augmentation

You can add Gaussian noise to the image by simply passing a noise value (e.g., noise=0.05) when calling geometric_augmentations.

Example:

geometric_augmentations(
    image_path=filepath[0],
    aug_type=[],
    apply_all=True,
    noise=0.05,
    patchmix=True
)

How it works:

• The GeoTIFF bands are optionally normalized to the range [0, 1]
• Random Gaussian noise is generated using the provided noise value as the standard deviation
• Noise is added to each pixel across all bands
• Values are clipped to keep them within valid range
• The noisy image is saved as a new GeoTIFF

---

---

PatchMix (patchmix=True)

PatchMix replaces one quadrant of the clean image with its noisy version to create hybrid samples. The image is split into four parts (top-left, top-right, bottom-left, bottom-right), and noise is inserted into one patch at a time.

File Naming

The output files show where noise was added:

• noise_tl.tif → top-left
• noise_tr.tif → top-right
• noise_bl.tif → bottom-left
• noise_br.tif → bottom-right

---

This helps models learn localized noise and improves robustness.

---

Run Locally

Clone the project:

git clone https://github.com/bdipesh3045/geospectral.git
cd geospectral

Install dependency:

pip install setuptools

Install the package:

pip install .

---

Thank you!

---

Thank you for using GeoSpectral! 🎉 If you have any questions or feedback, feel free to reach out at 📧 [e-mail] ✨

---