Metadata-Version: 2.4
Name: icepatchnet
Version: 0.1.0
Summary: Ice-PatchNet: A patch (segmented images) based CNN workflow
Author: Sai Vikas Amaraneni, Maloy Kumar Devnath, Shalini Srinivasa, Chhaya Kulkarni, Sudip Chakraborty, Vandana P. Janeja
License: iHARP
Project-URL: Homepage, https://github.com/vikasamaraneni333/icepatchnet
Project-URL: Source, https://github.com/vikasamaraneni333/icepatchnet
Project-URL: Issues, https://github.com/vikasamaraneni333/icepatchnet/issues
Keywords: icepatchnet,Ice-PatchNet,patch based CNN,Convolutional Neural Network,segmentation of images,image prediction,spatio temporal data mining,geospatial,geoai,climate,cryosphere,machine learning,data science
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Science/Research
Classifier: Topic :: Scientific/Engineering
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: Topic :: Scientific/Engineering :: Information Analysis
Classifier: Topic :: Scientific/Engineering :: GIS
Classifier: Topic :: Scientific/Engineering :: Atmospheric Science
Classifier: Topic :: Software Development :: Libraries :: Python Modules
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3 :: Only
Classifier: Operating System :: OS Independent
Requires-Python: >=3.8
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: numpy
Requires-Dist: torch
Requires-Dist: opencv-python
Requires-Dist: matplotlib
Requires-Dist: tqdm
Dynamic: license-file

# Predicting Antarctic Sea Ice with Scalable Deep Learning
## Ice-PatchNet
Ice-PatchNet is a novel, scalable deep learning framework designed to predict daily Sea Ice Extent (SIE) over the Antarctic region. By using a patch-based segmentation approach, the model can efficiently capture localized spatiotemporal features.

## Performance and Scope

- Target Region: Optimized for the Antarctic and Arctic polar regions.
- Resolution: Handles satellite imagery at a 25km spatial resolution.
- Predictive Windows: Supports lead-time predictions for 1-day, 7-day, and 14-day horizons.

### Important Version info
This version of Ice-PatchNet has only been tested on the images of sea ice extent over the Arctic region and the Antarctic region.

----------------------------------------------------------------------------

## Method Overview

1. **Image Preprocessing:** High-resolution SIE images are converted to grayscale to reduce complexity and computational overhead.
2. **Patch Segmentation:** The images are divided into a user-defined number of non-overlapping patches. This enables an equal distribution of SIE features and improves the model's ability to distinguish between land, open water, and ice.
3. **Spatiotemporal Feature Extraction:** Each patch is treated as a single unit and is processed through a deep convolutional neural network (3 layers, $3 \times 3$ kernels) with ReLU activation to identify localized patterns.
4. **Patches Reassembly:** Predicted patches are re-aligned using a global counter to reconstruct the full geographic extent without spatial distortion.

------------------------

## Citation
If you utilize this framework or the patch-based methodology in your research, please cite our ICDM '25 paper:
```
Amaraneni, S. V., Devnath, M. K., Srinivasa, S., Kulkarni, C., Chakraborty, S., & Janeja, V. P. (2025). Predicting Antarctic sea ice with scalable deep learning models. In Proceedings of the IEEE ICDM Workshop on Spatial and Spatiotemporal Data Mining (SSTDM).
```
