The purpose of the project I worked was to monitor mangroves by using drones, advanced imaging sensors, and machine learning. I worked specifically with deep learning methods (convolutional neural networks) to perform Mangrove species classification.
Mangroves are a coastal ecosystem that provides extremely valuable services: they sequester 2-4 times more carbon than tropical rainforests, protect coastlines from storm systems, and provide excellent fisheries. Unfortunately, mangroves are at risk due to human encroachment and the government lacks adequate information to provide effective enforcement to protect these sanctuaries.
Gulf of California Marine Program, Octavio Aburto Lab at Scripps Institute of Oceanography, and Engineers for Exploration traveled to La Paz, Mexico California Baja Sur to collect on-the-ground data with unmanned aerial vehicles. In addition to this, we performed field work to compare our data concerning the biomass of the three different Mangrove Species.
The biggest advantages of utilizing machine learning methods versus traditional classification methods is time and cost. Convolutional Neural Networks (CNN's) are most commonly applied to analyzing visual imagery. The CNN that I primarily worked with were YOLOv3 and OpenCV.
Using Agisoft, I created an Orthomosaic of our 5,000+ drone image tiles
We created and online labeling tool in Python to work with over a million image samples. I used this data to train my ML algorithms in OpenCV an YoloV3 frameworks.