Satellite image enhancement & crop classification

Our research project focuses on monitoring crops using satellite images. To address the problem, we leverage a fast-growing field of graph signal processing (GSP), which expands upon traditional signal processing techniques such as Fourier transform and wavelets to accommodate the graph domain. Specifically, we propose to unroll a designed graph-based algorithm into an interpretable feed-forward neural net for end-to-end parameter tuning, in order to enhance satellite images and classify field crops. The primary benefit for the public is the development of a robust and efficient model for crop monitoring. The model is resilient against the known covariate shift problem, which occurs when the distribution of input data differs significantly from the training and test datasets. Unlike conventional “black box” deep learning models, our model requires much less data for training to train significantly fewer network parameters, enabling an operator to save resources and time. With this technology, an operator can swiftly and accurately assess crop health and productivity, leading to informed decision-making and optimized agricultural practices.

Faculty Supervisor:

Gene Cheung

Student:

Partner:

Zenith Analytica

Discipline:

Computer science

Sector:

Professional, scientific and technical services

University:

York University

Program: