Essentially, we have a bunch of satellite images that contain trees (or not), and we want to teach a computer how to identify which ones do and don’t have trees using Deep Learning techniques.

So, let’s say you have this image:

This is an example of what we might see in a satellite image. The goal here would be to teach the computer how to identify which parts of the image contain trees (the green areas), and which don’t (the brown or gray areas).

To do this, we use something called Convolutional Neural Networks (CNNs) a type of Deep Learning algorithm that can learn patterns in images. Essentially, what happens is the computer looks at all these satellite images and tries to identify which parts have similar features as the ones it’s been trained on.

For example, let’s say we train our CNN using this image:

The computer will look at all these different parts of the image and try to identify which ones have similar features as the green areas in the training set (the trees). It might notice that there are certain patterns or shapes that tend to appear more often in images with trees, like long lines or clusters of small dots.

Once it’s learned these patterns, we can use our trained CNN to identify which parts of new satellite images contain trees and which don’t. This is done by feeding the image through the CNN, which will output a score for each part based on how closely it matches what the computer has been taught to look for in tree-containing areas.

So, if we feed this new image:

Through our trained CNN, it might output something like this:

# This script is used to output a score for each part of an image based on how closely it matches what the computer has been taught to look for in tree-containing areas.

# First, we define a list containing scores for each part of the image.
scores = [0.23, 0.89, 0.45]

# Next, we use square brackets to create a nested list, with each inner list representing a part of the image.
# The numbers within the inner lists are the scores for each part.
scores = [[0.23], [0.89], [0.45]]

# Finally, we use the print() function to output the scores in a readable format.
print(scores)

Where the numbers represent how likely each part of the image is to contain trees (the higher the number, the more likely). By looking at these scores and identifying which parts have the highest values, we can create a map that shows where the trees are located in our satellite images.

Deep Learning for Tree Detection in Satellite Imagery simplified.