This is a computer vision model that detects if forest areas experienced deforestation. Neural netwroks is a method to help predict the probability rather the image we inputted belongs to which possible category. We can input a satellite image of a forest region into the model, which then categorizes the image classified as experienced deforestation or hasn't experienced deforestation. By neural networks and Python techniques, the model does image classifications and calculates the probability which category the satellite image belongs to. The model will process and analyze the image data, which can be used to understand if the forest region experienced deforestation. This project was created within the AI4ALL Ignite Program SP24 with a supportive instructor, instructor assistant and mentor.

Predicting deforestation using satellite imagery is crucial because it is a step forward in climate change mitigation efforts. Deforestation contributes significantly to greenhouse gas emissions, which exacerbate climate change. By accurately predicting areas at risk of deforestation, we can take real-time measures to mitigate its impact on the climate. Secondly, preserving biodiversity and ecosystems relies heavily on monitoring deforestation patterns. Forests are home to countless endangered species of plants and animals, many of which risk losing their habitats and extinction. By predicting deforestation, we can prioritize areas for protection and restoration, helping to safeguard biodiversity. Moreover, human populations often depend on specific areas for their livelihoods. Many communities residing in or near forests rely on them for resources such as food, water, and medicine. Predicting deforestation allows for better planning and management of resources, ensuring the sustainability of these communities.

The ability to predict deforestation using satellite imagery is not only crucial for combating climate change but also for preserving biodiversity, ecosystems, and the livelihoods of human populations dependent on forested areas. The significance of deforestation extends globally for identifying deforestation patterns in different regions and contributing to initiatives of preserving natural ecosystems. It is important to identify if an ecosystem is experiencing deforestation early, so advocates can stop deforestation before a forest suffers from negative environmental effects

• Computer vision model has an accuracy rate of 80% from training sets
• Creates a bar chart indicates the proportion of 0 and 1 labels based on the training set
• Enable users to save and download a submission file of results

To accomplish this, we used the Python libraries for neural networks to help us detect deforestation in satellite images. The image data is processed and analyzed which enable the image classification. We create a graph to show our training sets that shows the red bar graph represents the satellite images that didn't experienced deforestation and the blue graph represents the satellite images that had experienced deforestation. In the model, we have thershold of the output probabilities which help in understanding the deforestation predictions.

Kaggle Dataset: Link

• Python
• pandas
• numpy
• Google Colaboratory
• Neural networks
• keras models and layers
• sklearn
• matplot

This project was completed in collaboration with:

• Betty Cheng ([email protected] and GitHub profile)
• Emily Rosenfeld ([email protected] and GitHub profile)
• Brianna Stan ([email protected] and GitHub profile)
• Mridul Pahwa ([email protected] and GitHub profile)