PROJECT TITLE :

Automatic Land Cover Reconstruction From Historical Aerial Images An Evaluation of Features Extraction and Classification Algorithms

ABSTRACT:

As large-scale epidemiological studies including retrospective study of spatial patterns have proliferated, the scientific community has become increasingly interested in land cover reconstruction from monochromatic historical aerial photographs. Remote sensing applications are, however, the primary focus of the computer vision community's work in the interpretation of high-resolution multispectral data collected by advanced spatial programmes. As a result, this study proposes four new ideas. They want to serve as a benchmark for future computer vision algorithms used to automate the restoration of land cover from black-and-white historical aerial photographs. An annotated map of France from 1970 to 1990 has been developed with the cooperation of geography specialists, which includes 4.9 million non-overlapping annotated patches. For this dataset, we've dubbed it HistAerial. A thorough comparison analysis of the current state of the art texture Feature Extraction and classification algorithms, including deep convolutional Neural Networks (DCNNs), has also been conducted. In the form of a rating, it's given An orthogonal combination representation of the binary gradient contours filter, known as the rotated-corner local binary pattern (R-CRLBP), is offered as a simplification of the binary gradient contours filter. To wrap things up, a novel mixture of low-dimensional texture descriptors, including the R-CRLBP filter, is introduced as a light combination of local binary patterns (LCoLBPs). In comparison with the DCNN techniques, the LCoLBP filter obtained state-of-the-art results on the HistAerial dataset with a low-dimensional feature vector space (17 times shorter).