Advancing Land Classification with AI: Exploring Kolmogorov-Arnold Networks

Land classification is crucial for environmental monitoring, resource management, and urban planning. This research explores using Kolmogorov-Arnold Networks (KAN), a novel machine learning model, for multispectral land classification. Unlike traditional neural networks, KAN utilizes univariate functions as activation mechanisms, enhancing its ability to capture complex spatial and spectral patterns in satellite imagery. The study focuses on classifying land in the Edmonton-Calgary corridor using hyperspectral data, which contains rich spectral information but presents computational challenges. The Finnish Centre for Artificial Intelligence (FCAI) will provide expertise in optimizing KAN for high-dimensional datasets, reducing computational demands, and improving model interpretability. A key goal is to compare KAN’s performance with conventional models regarding accuracy, robustness, and efficiency. The study will also assess whether KAN’s high accuracy results from superior learning abilities or potential overfitting. This project contributes to Canada’s leadership in AI and environmental monitoring, offering a more efficient, data-driven approach to land classification. The findings will aid policymakers in making informed urban development, agriculture, and conservation decisions. Additionally, the collaboration between the University of Alberta and Aalto University strengthens global research ties in AI-driven remote sensing.

Faculty Supervisor:

Arturo Sanchez-Azofeifa

Student:

Partner:

Aalto University

Discipline:

Earth science

Sector:

Artificial Intelligence

University:

University of Alberta

Program: