Ekstraksi Data Bangunan Dari Data Citra Unmanned Aerial Vehicle Menggunakan Metode Convolutional Neural Networks (CNN) (Studi Kasus: Desa Campurejo, Kabupaten Gresik)

Abstract

: Along with increasing development due to population growth, a proper monitoring of land use is needed, one of which is through mapping the distribution of buildings. Mapping the distribution of buildings can be done by analyzing remote sensing images taken using various vehicles, one of which is the Unmanned Aerial Vehicle (UAV) which can provide very high-resolution images. However, up to now the classification process is often done by manual digitization which is considered less effective and efficient so that an automatic extraction method is needed. In this study, the Convolutional Neural Networks (CNN) method was used to overcome the challenges of building extraction using high resolution aerial photo image data in Campurejo Village, Gresik Regency using the R-CNN where this is expected to be able to help the classification process automatically by using the Mask R-CNN algorithm. input data (training data). the results of the classification are validated and tested for accuracy to produce a large-scale building distribution map, namely 1: 5000. The accuracy of the building classification results using the Mask R-CNN method was tested using a confusion matrix which resulted in a precision value of 94.78%, recall 82.63%, F1 Score 88.29% and accuracy 79.03% for region 1 and for region 2 resulted precision value 98.10%, recall 78.37%, F1 Score 87.13% and accuracy 77.20%. While the number of buildings that can be in area 1 is 2102 buildings and area 2 is 247 buildings. shows great potential to utilize the Convolutional Neural Networks (CNN) method in extracting buildings.