Neural Networks and Photogrammetry for Analysis of Marine Remote Sensing Data

Project Summary

KC and Patrick led two hands-on data workshops for ENVIRON 335: Drones in Marine Biology, Ecology, and Conservation. These labs were intended to introduce students to examples of how drones are currently being used as a remote sensing tool to monitor marine megafauna and their environments, and how machine learning can be used to efficiently analyze remote sensing datasets. The first lab specifically focused on how drones are being used to collect aerial images of whales to measure changes in body condition to help monitor populations. Students were introduced to the methods for making accurate measurements and then received an opportunity to measure whales themselves. The second lab then introduced analysis methods using computer vision and deep neural networks to detect, count, and measure objects of interest in remote sensing data. This work provided students in the environmental sciences an introduction to new techniques in machine learning and remote sensing that can be powerful multipliers of effort when analyzing large environmental datasets.

Themes and Categories
Year
2019

Graduate Students: KC Bierlich and Patrick Gray

Faculty: Dr. David Johnston

Undergraduate Course: "Drones in Marine Biology, Ecology, and Conservation" (ENVIRON 335)

Course Summary

Learning Objectives

Lab 1:

  • Understand how to accurately measure objects from aerial images depending on different camera/sensors and the altitude of the photo.
  • Understand why it is important to measure body condition and health of top predators, such as whales, in an ecosystem.
  • Hands-on activity to measure the total length of blue, humpback, and minke whales from California and Antarctica using drone imagery.
  • Compare results to classmates/class and discuss reasons for variation in measurements.

Lab 2:

  • Understand basic remote sensing technology, temporal, spectral, spatial, and radiometric characteristics of different sensors.
  • Learn some intuitive ideas about machine learning and applications in marine science
  • Build up a library of examples of how remote sensing and machine learning can be used for marine science.
  • Learn more specifically how deep learning functions and gain an understanding of the theory behind convolutional neural networks. Use this knowledge and the general remote sensing knowledge to develop a neural network model to classify animals and everyday objects.

The first lab consisted a short lecture on the theory of photogrammetry and its conservation importance and then students worked directly with the drone imagery using ImageJ for hands-on photogrammetry. The second lab session began with a short crash course in machine learning and a summary of applications in marine remote sensing. Students then had time to familiarize themselves with the datasets and the python tools we used in a guided format using Jupyter notebooks stored on Github and run on the Google Colaboratory environment. We then trained our neural network models as a group but with each student running it independently on Colab.

Artificial intelligence-based techniques are bringing incredible advances to difficult research problems, but these methods are often inaccessible to non-technical ecologists. Developing an intuitive understanding of neural networks, without the complicated setup and coding challenges, was an essential goal of this, with the hope that students will be empowered enough to bring these new techniques into their own research.

Datasets

We used three datasets during this project. Lab 1 used images of blue, humpback, and Antarctic minke whales collected by our research group in Antarctica and California. Lab 2’s initial Neural Network Development was done on the publicly available MINST and CIFAR datasets. Students used these to experiment with model design and then as a final component of the project we demonstrated how a neural network could be used to identify and measure cetaceans using the same dataset as Lab 1.

Course Materials

Drone images of whales that students measured can be found here.

Students also downloaded ImageJ, a photogrammetry software, that they used for measuring. All code from Lab 2 can be found here. An overview of both deep learning and remote sensing intended to be an hour long lecture can be found here.

Graphics

Equation for measuring whales
Figure 1) Equation and example of measuring whales using aerial photogrammetry.
Drone imagery examples
Figure 2) Example of drone imagery in raw form and after being analyzed by a convolutional neural network to identify the species and measure each animal.
CIFAR dataset images
Figure 3) CIFAR Dataset for neural network model development

 

Related Projects

The goal of this Data Expedition was to introduce students to the exploration of social networks data using R. Students learned to load and plot a social network in R and then perform some basic analyses on two different networks: Hockey Fights in the National Hockey League in 2018-2019 and characters in Game of Thrones Season 3. Students used social network analysis to better understand who is connected to whom, how frequently they interact, and how they are interacting.

The data that students see in their statistics courses are often constrained to numeric and tabular data. However, there is an exciting field of data science and statistics known as text analysis. This expedition introduces students to the concept of treating text as data frames of words, and demonstrates how to perform basic analyses on bodies of text using R. Tweets of four Democratic candidates for the 2020 Primary are used as data, and demonstrated text analysis techniques in the expedition include comparisons of word frequencies, log-odds ratios for word usage, and pairwise word correlations.

Fluid mechanics is the study of how fluids (e.g., air, water) move and the forces on them. Scientists and engineers have developed mathematical equations to model the motions of fluid and inertial particles. However, these equations are often computationally expensive, meaning they take a long time for the computer to solve. 

To reduce the computation time, we can use machine learning techniques to develop statistical models of fluid behavior. Statistical models do not actually represent the physics of fluids; rather, they learn trends and relationships from the results of previous simulations. Statistical models allow us to leverage the findings of long, expensive simulations to obtain results in a fraction of the time.

In this project, we provide students with the results of direct numerical simulations (DNS), which took many weeks for the computer to solve. We ask students to use machine learning techniques to develop statistical models of the results of the DNS.