Can deep learning algorithms be trained to automate the classification of information held in historical and repeat photographs?
Repeat photography is a valuable tool for evaluating long-term ecological change. Historical photographs used for repeat photography often predate conventional remote sensing data by decades, and the oblique perspective of the photographs capture details of the landscape absent in nadir imagery. To date, most approaches to quantifying landscape change using repeat photography have involved manual classification techniques (i.e. demarcating land cover categories by hand), which are time-consuming, expensive and difficult to reproduce consistently. The objective of this proposed research project is to develop a new approach that automates the classification process using deep learning models. Deep learning is a subset of machine learning in artificial intelligence, and has previously been used to dramatically improve the performance of tasks such as speech recognition, visual object recognition, and object detection. This project will configure and train a deep learning model to classify photographs, and assess the accuracy of these results using existing land cover data.