Tree Crown Delineation Algorithm Based on a Convolutional Neural Network

Abstract

Tropical forests concentrate the largest diversity of species on the planet and play a key role in maintaining environmental processes. Due to the importance of those forests, there is growing interest in mapping their components and getting information at an individual tree level to conduct reliable satellite-based forest inventory for biomass and species distribution qualification. Individual tree crown information could be manually gathered from high resolution satellite images; however, to achieve this task at large-scale, an algorithm to identify and delineate each tree crown individually, with high accuracy, is a prerequisite. In this study, we propose the application of a convolutional neural network—Mask R-CNN algorithm—to perform the tree crown detection and delineation. The algorithm uses very high-resolution satellite images from tropical forests. The results obtained are promising—the R e c a l l , P r e c i s i o n , and F 1 score values obtained were were 0.81 , 0.91 , and 0.86 , respectively. In the study site, the total of tree crowns delineated was 59,062 . These results suggest that this algorithm can be used to assist the planning and conduction of forest inventories. As the algorithm is based on a Deep Learning approach, it can be systematically trained and used for other regions.

U-Net-Id, an Instance Segmentation Model for Building Extraction from Satellite Images—Case Study in the Joanópolis City, Brazil

“…For example, a very high resolution satellite image of a city can easily contain thousands of houses and buildings. This characteristic complicates the use of iterative region proposal frameworks, such as Mask R-CNN, mainly because: (i) during processing, this model creates one activation layer per object, thus it can quickly surpass the computer capacity for images with high density of objects of interest [9,16]; and (ii) in remote sensing, images are not independent. For example, even if one image is clipped into several smaller tiles for prediction, the tiles will need to be merged, and deciding whether or not to merge the object bounding box overlapping the border is a subjective expert-based decision, which decreases the segmentation accuracy.…”

Multisensor Data Fusion for Improved Segmentation of Individual Tree Crowns in Dense Tropical Forests

“…Other deep learning methods have recently been developed to segment tree crowns, often using convolutional neural networks (CNNs). The Mask R-CNN for instance does not simply draw bounding boxes but delineates crowns exactly [59], while the Faster-CNN has been applied to ALS data [60], [61] to extract points corresponding to individual trees. These deep learning methods present additional promising opportunities for fusing data from different sensors.…”

A benchmark dataset for canopy crown detection and delineation in co-registered airborne RGB, LiDAR and hyperspectral imagery from the National Ecological Observation Network

“…While there are dozens of proposed algorithms, they are often designed and evaluated using a range of different data inputs [2][3][4], sensor resolutions, forest structures, evaluation protocols [5][6][7][8], and output formats [9]. For example, [10] proposed a pixel-based algorithm for 50 cm pan-sharpened satellite RGB data from a tropical forest in Brazil evaluated against field-collected tree stem locations, and [11] proposed a vector-based algorithm for 10 cm fixed-winged aircraft RGB data from oak forests in California evaluated against image-annotated crowns. Given these differences, a comparison among algorithms is difficult to make based on reported statistics to interpret the relative accuracy, generality and cost effectiveness.…”