| Résumé : |
(auteur) In this article, a new feature detection approach based on integration of LiDAR data and visible images in the form of a semi-automatic method has been proposed. In this approach, a two-step method for feature detection was developed using object-based analysis in order to increase the level of automation and level of accuracy in the detection process. The first step is providing a method for integration of two data sources for detection process by maintaining independency between image data and LiDAR altimetric data. In this step, the feature detection process is started based on image data and for detecting areas that detection properly is not done, LiDAR altimetric data is used. In the second step, a new method for detection of vegetation is implemented. Of the characteristics of this method is that there is no need to use the infrared band in the image data and also there is no need for LiDAR intensity data. The implemented method in the recent step is based on the new indices developed for detection of vegetation using three visible bands (red, green, and blue). The results of applying the method on two sample data sets show that the proposed approach and developed indices have the lowest dependency on the type and region of imaging and about each input image data includes visible bands (red, green, and blue) along with LiDAR data (that both data have a high spatial resolution), feature detection process is done with acceptable accuracy. Only thresholds depend on image data and change about different images. The changes are very small. Therefore, using the mean of these thresholds, despite may not be optimal for all image data, but generally is useful and for different images is efficient. In the case of many accessible images from Iran, the thresholds determined optimally by the trial-and-error method, the changes were very small. About the image data of Toronto and Iran which great changes were expected in the thresholds, the optimal thresholds showed very small changes. The results of this research demonstrated that the proposed method can successfully detect urban features (include vegetation, road, and building) with different shapes. Evaluation process showed that the overall accuracy, kappa coefficient, producer’s accuracy, and user’s accuracy of the proposed method about vegetation are 97%, 92%, 96%, and 94%, respectively. Also, the producer’s accuracy, user’s accuracy, and kappa coefficient about the building class are 94%, 95%, and 91%, respectively. About the road class these parameters are 95%, 89%, and 91%. |
Ajouter le résultat dans votre panier
Visionner les documents numériques
Affiner la recherche Interroger des sources externesA deep learning framework based on generative adversarial networks and vision transformer for complex wetland classification using limited training samples in International journal of applied Earth observation and geoinformation, vol 115 (December 2022)
Construction d’un plugin QGIS de détection d’îlots de chaleur urbains à partir d’images satellitaires de type optique (2022)
