AssessEdge 2.0

Accuracy assessment on edge

edge_assessment

An assessment tool for evaluating the quality of object edges within an input map.

Table of Contents

• Background
• Install
• Usage
• Contributors
• License

Background

An accurate delineation of edge is critical for object-based image analysis on land resources such as agricultural field objects and building objects, and change detection particularly for climate-induced land change such as snow melting and treeline shifts. The traditional F1-score has been widely used for assessing map quality, which, however, often fails to provide information on object boundary quality. To address this problem, we developed a new evaluation index to take account of both thematic and edge accuracy.

Install

Install AssessEdge using pip:

pip install AssessEdge

Use requirements.txt to install the required packages.

pip install -r requirements.txt

Usage

The two main functions are eval_edge and plot_results in edge_buffer.py

eval_edge(testing_field, reference_field, step, min_width, initial_max_width): input two variables(testing_field & reference_field), the testing_field represents the result of classification, the reference_field represents the real reference object field. The pixel values of target objects should be processed as 1, and pixel values of background should be 0 or no data. The function will return a dictionary include commission error, omission error, edge f1 score, middle point, etc. The step defaults as 3, the initial_max_width defaults as 0, and the min_width defaults as 3.

plot_results: show the testing_field and reference_field data fit curves and their data points.

Users can type the following codes to use these two functions

from AssessEdge import eval_edge, plot_results
from AssessEdge import rasterio_loaddata

def main(map_path, reference_path):
    map_path = "/images/testing_field.tif"
    reference_path = "/images/reference_field.tif"
    map_array = rasterio_loaddata(map_path)
    reference_array = rasterio_loaddata(reference_path)
    result = eval_edge(map_array, reference_array)
    plot_results(result)

if __name__ == "__main__":
    main()

Test

Testing images have been placed in the images folder. The test procedure is assess_image.py in test folder. To test it

python assess_image.py --map_path map_path --reference_path reference_path

It will show the test result as following

Contributors

Yingfan Zhang ([e-mail]), Su Ye

License

Apache License