Community Projects

The Data and Decision Sciences Lab is constantly working with local community partners on several data related projects that have a local and global impact.

Pixelwise Classification of Agricultural Crops in Aerial Imagery Using Deep Learning Methods

Community Partner: Texas A&M University
Year: 2018
Team Members Involved

Information extracted from multispectral imagery obtained from satellite or aerial devices has found applications in a wide range of areas including urban planning, crop and forest management, disaster relief, and climate modeling. At present, much of the extraction is still performed by human experts, making the process slow, costly, and error prone. The goal of this project was to develop methods for automatically extracting the locations of crops from aerial images. We investigated the use of deep learning neural networks for the task of semantic image segmentation. Semantic segmentation refers to labeling each single pixel in an image with the class it belongs to. Devising it as a supervised machine learning problem, a deep neural network was designed, implemented and experimentally evaluated. A single Fully Convolutional Network (FCN) was implemented to solve a seven- class segmentation problem, distinguishing corn, sugarcane, sorghum, soybeans, cotton, grass and non-crop areas. The network was trained with both spectral and spatial information. The standard U-Net model was trained using many small patches taken from the original image. The project investigated the effect of patch size and the number of crops per patch on the segmentation results. The most promising model achieves a training accuracy of 82.73% and was able to successfully segment all seven classes. This project has also been published as a Master’s thesis.

Input Image
Ground Truth or Labels Image
Segmentation Output Image