Automatic land cover validator

Overview

Final degree project for University of Malaga at Khaos Research. Author: Jesús Aldana Martín

Description

This end-of-degree project aims to implement a software tool that allows the validation of land cover classification models in the Iberian Peninsula. With the intention of monitoring the cultivated fields and their distribution, for this we are going to make use of the data of the Geographic Information System of Agricultural Plots (SIGPAC) provided by the Ministry of Agriculture, Fisheries and Food and of free access. With the help of these data and processed satellite images, it will be possible to execute a Python script that obtains an output raster with the new input data and metrics that allow us to measure and validate the performance of the classification model. To test the tool, the entire community of Andalusia will be analyzed, obtaining different analysis rasters and statistical metrics for validation.

Shapefiles data source

All the shapefiles used in this project have been downloaded from the 'Junta de Andalucía' geographic informaton data.

Contents

Code functions:

• Reproject rasters
• Merge tiff images
• Create mask from SHP data
• Point in polygon
• Raster comparison
• Validation metrics

Setup

To run locally the script you just need to install all libraries specified in the requirements.txt. The code below showw how can you do it.

python3 -m venv <venv_name>

source <venv_name>/bin/activate

python3 pip install -r requirements.txt

Usage

Use all the functions as you wish or run the whole workflow with the launch.sh app. In order to run the script please, replace the <"parameters"> with your own paths.

bash launch.sh -r <raster path> -s <shp path> -o <output path> -t <delete tmp>

Testing

The framework used for the unit tests is pytests. In order to run the tests:

pytest tests/

Output example

SIGPAC raster

Image raster with the SIGPAC data as the new band values.

Validation True/False

This raster shows wich pixels are common in both rasters.

Validation Confusion Matrix

For this image each colour represents the

For extra information check out the showcase.ipynb notebook.

Results

The following table summarizes the results of our analysis for various classes or categories. It provides insights into the number of pixels, correct classifications (aciertos), misclassifications (fallos), and the percentage of accuracy for each class.

Clase	Num Pixeles	Aciertos	Fallos	Porcentaje de Acierto
Citricos Frutal	1069300	89403	979897	8.36
Citricos	8041756	5856071	2185685	72.82
Citricos-Frutal de cascara	8699	1545	7154	17.76
Citricos-Viñedo	2952034	2950975	1059	99.96
Frutal de Cascara-Frutal	838815	7312	831503	0.87
Frutal de Cascara-Olivar	279257	82857	196400	29.67
Frutal de Cascara	20177469	9080721	11096748	45.0
Frutal de Cascara-Viñedo	12436	3806	8630	30.6
Frutal	14711965	4774161	9937804	32.45
Imvernadero y cultivos bajo plastico	5015050	771053	4243997	15.37
Olivar-Citricos	110703	36820	73883	33.26
Olivar-Frutal	93675	30066	63609	32.1
Olivar	162620685	72449454	90171231	44.55
Tierra Arable	161368050	105376719	55991331	65.3
Huerta	1238266	570850	667416	46.1
Frutal-Viñedo	98892	35013	63879	35.41
Viñedo	2373903	1122568	1251335	47.29
Olivar-Viñedo	147654	55533	92121	37.61

This data provides a valuable overview of the performance of our classification system for each class. It can be used for further analysis and decision-making in our project.

---

The MIT License (MIT)

This project is licensed under the MIT license. See the LICENSE file for more info.