Land Use Classification using Convolutional Neural Networks Applied to Ground-Level Images

TL;DR

This paper introduces a novel framework for land use classification using ground-level images from Flickr, leveraging deep learning and filtering techniques to overcome challenges like imprecise geolocation and uneven data distribution.

Contribution

It presents a comprehensive approach combining filtering, semi-supervised augmentation, and deep learning features for effective land use mapping from ground-level images.

Findings

Indoor/outdoor classifier achieves state-of-the-art accuracy.

Deep learning features yield over 76% accuracy in land use classification.

Framework effectively addresses geolocation noise and data imbalance.

Abstract

Land use mapping is a fundamental yet challenging task in geographic science. In contrast to land cover mapping, it is generally not possible using overhead imagery. The recent, explosive growth of online geo-referenced photo collections suggests an alternate approach to geographic knowledge discovery. In this work, we present a general framework that uses ground-level images from Flickr for land use mapping. Our approach benefits from several novel aspects. First, we address the nosiness of the online photo collections, such as imprecise geolocation and uneven spatial distribution, by performing location and indoor/outdoor filtering, and semi- supervised dataset augmentation. Our indoor/outdoor classifier achieves state-of-the-art performance on several bench- mark datasets and approaches human-level accuracy. Second, we utilize high-level semantic image features extracted using deep learning, specifically convolutional neural net- works, which allow us to achieve upwards of 76% accuracy on a challenging eight class land use mapping problem.