Personal Web Site. - Landscape Segmentation

Direct and indirect effects of hurricane disturbance in a mangrove ecosystem: An evaluation using remote sensing and landscape segmentation

Project is developing by:

Juan Pablo Serrano-Rubio, Ulises Vidal-Espitia and Mateo D.M. Ruiz

Sincere thanks are extended to Luis Oscar Rios for providing all photography equipment and piloting the drone.

We use drone images of a post-hurricane mangrove in south-west Mexico to apply landscape segmentation and evaluate direct disturbance effects on forest cover. The images were obtained in August 2015 at ten sites comprising two mangrove types. We deﬁned thirteen classes of landscape micro-patch to be segmented. Table 1 presents the kinds of landscape micro-patches found in aerial images and a assigned color in order to identify the class in the resulting segmented images. The image to the right presents the estimation of superpixels and the human labelling of superpixels in order to build the Ground-Truth.

Published Paper:

Serrano-Rubio, Juan Pablo, Ruiz, M.D.M. & Vidal-Espitia, Ulises. Integrating remote sensing and image processing to test for disturbance effects in a post-hurricane mangrove ecosystem. SIViP 15, 351–359 (2021). doi.org/10.1007/s11760-020-01754-9

Sincere thanks are extended to Luis Oscar Rios for providing all photography equipment and piloting the drone

Landscape images dataset. The landscape segmentation can be considered a classification problem. We use Random Forest Classifier to distinguish the corresponding class to each superpixel.

Landscape image number 1

$DCIM\100GOPRO\G0010884.$

Original landscape image

(Number 1)

Segmentation of the image using superpixels including its Ground-Truth

Segmentation of the landscape image

Class: 1. White Mangrove, 3. Coconut Palm, 4. Royal Palm, 6. Beach Hibiscus, 7. Dead Wood, 8. Substrate and 9. Water.

Landscape image number 2

$DCIM\100GOPRO\G0020970.$

Original landscape image

(Number 2)

Segmentation of the image using superpixels including its Ground-Truth

Segmentation of the landscape image

Class: 1. White Mangrove, 7. Dead Wood, 8. Substrate and 11. Equipment (high visibility jacket / boat).

Landscape image number 3

$DCIM\100GOPRO\G0041041.$

Original landscape image

(Number 3)

Segmentation of the image using superpixels including its Ground-Truth

Segmentation of the landscape image

Class: 1. White Mangrove, 7. Dead Wood, 8. Substrate and 11. Equipment (high visibility jacket / boat).

Landscape image number 4

$DCIM\100GOPRO\G0010522.$

Original landscape image

(Number 4)

Segmentation of the image using superpixels including its Ground-Truth

Segmentation of the landscape image

Class: 2. Red Mangrove, 7. Dead Wood, 9. Water, 10. Water with sediments and 13. Rushes.

Landscape image number 5

$DCIM\100GOPRO\G0031143.$

Original landscape image

(Number 5)

Segmentation of the image using superpixels including its Ground-Truth

Segmentation of the landscape image

Class: 2. Red Mangrove, 7. Dead Wood, 9. Water and 10. Water with sediments.

Landscape image number 6

$DCIM\100GOPRO\G0020568.$

Original landscape image

(Number 6)

Segmentation of the image using superpixels including its Ground-Truth

Segmentation of the landscape image

Class: 2. Red Mangrove, 5. Lily, 6. Beach Hibiscus, 7. Dead Wood, 9. Water, 11. Equipment (high visibility jacket / boat) and 12. Water with Reflections.