CNN Algorithm for Roof Detection and Material Classification in Satellite Images

Abstract

This paper suggests an algorithm for extracting the location of a building from satellite imagery and using that information to modify the roof content. The materials are determined by measuring the conditions where the building is located and detecting the position of a building in broad satellite images. Depending on the incomplete roof or material, there is a greater possibility of great damage caused by disaster situations or external shocks. To address these problems, we propose an algorithm to detect roofs and classify materials in satellite images. Satellite imaging locates areas where buildings are likely to exist based on roads. Using images of the detected buildings, we classify the material of the roof using a proposed convolutional neural network (CNN) model algorithm consisting of 43 layers. In this paper, we propose a CNN structure to detect areas with buildings in large images and classify roof materials in the detected areas.

An Object-Oriented Approach to the Classification of Roofing Materials Using Very High-Resolution Satellite Stereo-Pairs

“…The burdensome requirement of a large training set limits the adoption of deep learning algorithms for roofing materials classification, because in many practical cases ground-truth datasets are relatively small and they can be exploited more efficiently with SVM [51]. Here, the achieved accuracies (overall 91%, user and producer reported in Table 3) are comparable (or outperform in some cases) with the reported experiments using similar imagery and OBIA approach [12][13][14], and also with few works exploiting convolutional neural networks but with a more limited number of material classes [9,23].…”

“…A recent work by Kim et al [23] implemented deep learning algorithms using aerial images with four bands only, but it was limited to a few materials (concrete and metal) and required an extensive training dataset (more than 11,600 images). Usually, when working with multispectral sensors (with a more limited number of bands compared to hyperspectral ones), images are to be coupled with elevation information and an objectoriented approach is to be preferred, in order to achieve comparable accuracy levels [12,15].…”

A Comparative Study of PP-LiteSeg, Dual Attention Network, DeeplabV3p and Asymmetric Neural Network for Rooftop Detection in UAV Images

“…The proposed methodology and experimental results validate the effectiveness of CNNs for accurate and efficient building extraction, contributing to the advancement of remote sensing analysis techniques. [16] proposed a rooftop extraction method using a CNN-based classification model trained on a labeled dataset of UAV images. Their results demonstrated the effectiveness of CNNs in unique rooftops from other image components.…”

Towards a circular building industry through digitalisation

“…Such methods for automated retrieval of material information are becoming increasingly popular due to advancements in both software and hardware sensors. For instance, Raghu et al (2022b) built a model to detect external façade materials such as brick, stone, wood and stucco, while Kim et al (2021) explored the generation of algorithms to identify concrete and metal roofs. The algorithms can also be leveraged for condition assessment of buildings, providing insights into the current state of the building and identifying potential maintenance issues, thus, supporting maintenance operations.…”