Search

Deep learning has substantially pushed forward remote sensing image change detection through extracting discriminative hierarchical features. However, as the increasingly high-resolution remote sensing images have abundant spatial details but limited spectral information, the use of conventional backbone networks would give rise to blurry boundaries between different semantics among hierarchical features. This explains why most false alarms in the final predictions distribute around change boundaries. To alleviate the problem, we pay attention to feature refinement and propose deep learning networks that deliver improved separability (ISNet). Our ISNet reaps the advantages from two strategies applied to refining bitemporal feature hierarchies: 1) margin maximization that clarifies the gap between changed and unchanged semantics and 2) targeted arrangement of attention mechanisms that direct the use of channel attention (CA) and spatial attention (SA) for highlighting semantic and positional information, respectively. Specifically, we insert CA modules into share-weighted backbone networks to facilitate semantic-specific feature extraction. The semantic boundaries in the extracted bitemporal hierarchical features are then clarified by margin maximization modules, followed by SA modules to enhance positional change responses. A top–down fusion pathway makes the final refined features cover multiscale representations and have strong separability for remote sensing image change detection. Extensive experimental evaluations demonstrate that our ISNet achieves state-of-the-art performance on the LEVIR-CD, SYSU-CD, and Season-Varying datasets in terms of overall accuracy (OA), Intersection-of-Union (IoU), and F1 score. Code is available at https://github.com/xingronaldo/ISNet. [ABSTRACT FROM AUTHOR]

Accurately mapping forest carbon density by combining sample plots and remotely sensed images has become popular because this method provides spatially explicit estimates. However, mixed pixels often impede the improvement of the estimation. In this letter, regression modeling and spectral unmixing analysis were integrated to improve the estimation of forest carbon density for the You County of Hunan, China, using Landsat Thematic Mapper images. Linear spectral unmixing with and without a constraint (LSUWC and LSUWOC) and nonlinear spectral unmixing (NSU) were compared to derive the fractions of five endmembers, particularly forests. Stepwise regression, logistic regression, and polynomial regression (PR) with and without the forest fraction used as an independent variable and the product of the forest fraction image and the map from the best model without the forest fraction were compared. The models were developed using 56 sample plots, and their results were validated using 26 test plots. The decomposition of mixed pixels was assessed using higher spatial resolution SPOT images and a corresponding land cover map. The results showed that 1) LSUWC more accurately estimated the endmember fractions than LSUWOC and NSU, 2) PR had the greatest estimation accuracy of forest carbon, and 3) combining regression modeling and spectral unmixing increased the estimation accuracy by 31%–39%, and introducing the forest fraction into the regressions performed better than the product of forest fraction image and the results from PR without the fraction. This implied that the integrations provided great potential in reducing the impacts of mixed pixels in mapping forest carbon. [ABSTRACT FROM PUBLISHER]

Compared with airborne laser scanning, terrestrial laser scanning (TLS) offers ground-based point cloud data of trees and provides greater potential to accurately estimate tree and stand parameters. However, there is a lack of effective methods to accurately identify locations of individual trees from TLS point cloud data. It is also unknown whether the estimation accuracy of the parameters, including tree height (H), diameter at breast height (DBH), and so on, using TLS can meet the requirement of forest management and planning. In this letter, a novel method to effectively process point cloud data and further determine the locations of individual trees in a stand based on the central coordinates of point cloud data on a defined grid according to the largest DBH was developed. Moreover, a point-cloud-data-based convex hull algorithm and the cylinder method were, respectively, used to estimate DBH and H of individual trees. This study was conducted in a pure Chinese fir plantation of 45 trees located in Huang-Feng-Qiao forest farm, You County of Hunan, China. The comparison of the estimated and observed values showed that the obtained tree locations had errors of less than 20 cm, and the relative root mean square errors for the estimates of both DBH and H were less than 5%. This implies that TLS is very promising for the retrieval of tree and stand parameters in forest stands. For the applications of these methods to mixed forests with a structure of multilayer canopies, further examination is needed. [ABSTRACT FROM PUBLISHER]

Combining forest inventory plot and Landsat Thematic Mapper (TM) data has been widely used for mapping forest carbon. However, uncertainty analysis is a great challenge. This study investigated the uncertainties of mapping and scaling up aboveground forest carbon (AGFC) due to plot location errors in Wu-Yuan of East China. Plot location errors were simulated by randomly perturbing the location of each plot with eleven different distances that varied from 5 to 8000 m. Given a perturbed distance (PD) such as 100 m, a forest carbon map was created by combining and scaling up the plot and TM data from a spatial resolution of 28.5 m \times 28.5 m to 969 m \times 969 m using a sequential Gaussian block cosimulation algorithm. The maps obtained from the perturbed plot locations were compared with that from the true plot locations. The results showed that, as the plot location PD increased, the accuracy of predicted AGFC values decreased, but their spatial patterns (clustering of high and low values) remained until the PD of 800 m, slightly changed at the PD of 1600 m, looked more different at the PDs of 3000 and 5000 m, and became totally random at the PD of 8000 m. More importantly, it was found that scaling up the spatial data mitigated the impacts of plot location errors on the map accuracy compared to those without the up-scaling. [ABSTRACT FROM PUBLISHER]