Tree-Identification-Algorithm

Competed against 90 teams and placed 1st 🥇 in the Agriculture and Agri-food Canada (AAFC) challenge and 3rd 🏆 overall in the CanDev Data Challenge hackathon!

Task

We need your help to develop an algorithm to identify and extract the tree crowns and/or a row of trees in and around Canadian croplands. Trees in croplands serve as sinks of carbon and play a significant role in mitigating climate change. We are interested in monitoring the changes in tree distribution to quantify carbon losses or gains in croplands.

Installation Steps

1. Download Anaconda: https://www.anaconda.com/products/individual

2. In the Anacona Prompt, enter the following commands:

   Create a python environment

   conda create -n pyEnv python=3.9
   conda activate pyEnv
   

   Install libraries

   conda install -c conda-forge opencv
   conda install -c conda-forge matplotlib
   conda install -c conda-forge spyder
   

   Launch Spyder:

   spyder
   

3. Download or clone this repository to run the source code on spyder.

Input Image Requirements

1. Use a PNG image for more accurate results.
2. Image must be of a 64 hectare (800m by 800m) plot of land.
3. Hue will need to be adjusted according to the image (Example images and hues can be found in the test_image directory).

How to Run the code

In TreeDetectionAlgorithm.py:

• update img = cv2.imread("image.PNG") with the image filename/filepath. (For example img = cv2.imread("test-images\OnePlot.PNG"))
• adjust lower_color = np.array([40, 20, 0]) with the corresponding Hue (the first number)

Run the program.

Output

The program outputs:

• an image with tree areas surrounded by either green or red
• the final area (in both pixels and hectares)
• the percentage of the final tree area

The areas in green are less than 1 hectare, which and are included in the final area.

The areas in red are larger than 1 hectare, which are not included in the final area.

Example Output

Size of image in pixels: 1062 x 1039
Area of trees in pixels: 53451.5
Area of trees in hectares: 3.100272063714748
Percentage of area covered by trees: 4.84 %

Contributors

• Shrimei Chock
• Maisha Abdullah
• Thanuja Sivaananthan

Sources

• Detect Objects of similar colour: https://techvidvan.com/tutorials/detect-objects-of-similar-color-using-opencv-in-python/
• Calculate area of non-contiguous shape: https://stackoverflow.com/questions/55544388/how-can-i-calculate-the-area-inside-non-contiguous-shapes-in-an-image
• Find area of a contour: https://stackoverflow.com/questions/46491643/find-area-of-cv2-findcontours-python-opencv?fbclid=IwAR33h-jPe5MAaAGe53r7rbYjJ_aAHSTmwvkWlVz0g6va7_xqN9Dwe5YtYWY