Identifying an Effective Semantic Segmentation Algorithm for Deforestation Analysis

Public project summary

The purpose of this project is to implement a method for analyzing changes in land cover due to deforestation in the United States. Previously, the USDA used decision-tree classification in semantic segmentation (i.e., labelling of each pixel in an image with a category label) to assess land cover change. However, deep learning has recently become popular, and it has yet to be applied successfully to deforestation analysis. The proposed algorithm incorporates convolutional neural networks, but with an additional filter, called atrous convolutions. These filters are dilated in order to achieve a broader view of the scene or image.