Oil palm has become a means of rapid agricultural growth in tropical areas where it thrives. This resource can be found in products ranging from shampoo to frozen foods. While increasing job availabilities, oil palm plantations have led to a dramatic rise in deforestation and loss of biodiversity. Species like the Orangutan and Sumatran Tiger are being displaced from their lands and pushed closer to extinction.

Objective: use high-resolution satellite imaging to automate mapping of oil palm plantations and track this rising and alarming trend.

Strategy and outcome: using PyTorch as the deep learning framework, I trained the VGG16 neural network on satellite images, some of which contained oil palm plantations. After fine-tuning parameters, I achieved an accuracy score as high as 99%, and an f1-score of 0.95.

Image visualizations

View code

Powerpoint presentation