D polygons obtained on tiles in the test set with various bands and also the corresponding reference. The polygons obtained working with the composite images are more aligned with the reference data and with fewer false positives than these obtained from RGB photos or the nDSM only. The overall performance acquire is specifically visible for big buildings with complex structures and buildings with holes. Fewer false positives are observed for modest buildings within the results obtained applying composite photos. Compared together with the polygons obtained from RGB photos, the polygons obtained in the nDSM have fewer false positives and are additional aligned with ground truth. Also, the polygons of big buildings are additional common than the small ones in dense urban regions. There are actually additional false positives for little buildings in dense urban regions than in sparse locations. By visual observation, we could conclude that some of them are storage sheds or garden homes, which are not included inside the reference footprints. Their QX-314 supplier similar spectral character and height make it difficult to differentiate them from residential buildings. In summary, the nDSM enhanced constructing outlines’ accuracy, resulting in better-aligned creating polygons and stopping false positives. The polygons obtained from various composite pictures are Remote Sens. 2021, 13, x FOR PEER Overview 14 of 23 extremely equivalent to each other.nDSM + PredictionRGB + PredictionnDSM + PredictionRGB + Prediction(a)(b)(c)(d)(e)Figure eight. Benefits obtained on two tiles from the test dataset for the urban location. The loss functions are cross-entropy plus the background could be the aerial image and the corresponding nDSM. The predicted polygons are produced with 1 pixel for Dice. the tolerance parameter on the polygonization technique. From left to proper: (a) The predicted polygons are(b) predictedwith The background would be the aerial image as well as the corresponding nDSM. reference creating footprints; produced 1 pixel for the on aerial images (RGB);in the polygonizationon nDSM; (d) predicted polygons on composite image 1 (RGB + polygons tolerance parameter (c) predicted polygons strategy. From left to suitable: (a) reference building footprints; nDSM); polygons on aerial on composite image two (RGB + polygons on (b) predicted(e) predicted polygonsimages (RGB); (c) predicted NIR + nDSM). nDSM; (d) predicted polygons on composite image 1 (RGB + nDSM); (e) predicted polygons on composite image 2 (RGB + NIR + nDSM). Figure 9 shows the predicted polygon on diverse datasets. Comparing the polygon obtained in the aerial image (RGB) with that on composite image 1 (RGB + nDSM) shows that the model can’t differentiate nearby buildings with only spectral facts. This benefits inside the predicted polygon in the aerial image (RGB) corresponding to a number of individual buildings. Additionally, part of the road on the left side of your building is regarded to be a constructing. Comparing the polygon obtained with the nDSM with that on compositeFigure 8. Outcomes obtained on two tiles of the test dataset for the urban region. The loss functions are cross-entropy and Dice.Remote Sens. 2021, 13,14 ofFigure 9 shows the predicted polygon on different datasets. Comparing the polygon obtained inside the aerial image (RGB) with that on composite image 1 (RGB + nDSM) shows that the model can’t differentiate nearby buildings with only spectral Brivanib References details. This results within the predicted polygon within the aerial image (RGB) corresponding to several individual buildings. Moreover, part of the road on.