Spatiotemporal data fusion is a cost-effective way to produce remote sensing images with high spatial and temporal resolutions using multi-source images. Using spectral unmixing analysis and spatial interpolation, the Flexible Spatiotemporal DAta Fusion (FSDAF) algorithm is suitable for heterogeneous landscapes and capable of capturing abrupt land-cover changes. However, the extensive computational complexity of FSDAF prevents its use in large-scale applications and mass production. Besides, the domain decomposition strategy of FSDAF causes accuracy loss at the edges of sub-domains due to the insufficient consideration of edge effects. In this study, an enhanced FSDAF (cuFSDAF) is proposed to address these problems and includes three main improvements: (1) The TPS interpolator is replaced by an accelerated inverse distance weighted interpolator to reduce computational complexity. (2) The algorithm is parallelized based on the Compute Unified Device Architecture (CUDA), a widely used parallel computing framework for graphics processing units (GPUs). (3) An adaptive domain decomposition method is proposed to improve the fusion accuracy at the edges of sub-domains and to enable GPUs with varying computing capacities to deal with datasets of any size. Experiments showed while obtaining similar accuracies to FSDAF and an up-to-date deep-learning-based method, cuFSDAF reduced the computing time significantly and achieved speed-ups of 140.3–182.2 over the original FSDAF program. cuFSDAF is capable of efficiently producing fused images with both high spatial and temporal resolutions to support applications for large-scale and long-term land surface dynamics.
PS: If face compilation issues with ALGLIB, check the closed issues on the GitHub project for possible help.
cuFSDAF_PARAMETER_START
# The input fine image at t1
IN_F1_NAME = D:\Datasets\Daxing\Landsat\L-2018-12-4.tif
# The input coarse image at t1
IN_C1_NAME = D:\Datasets\Daxing\MODIS\M-2018-12-4.tif
# The input coarse image at t2
IN_C2_NAME = D:\Datasets\Daxing\MODIS\M-2018-10-1.tif
# The classified image for the fine image at t1
IN_F1_CLASS_NAME = D:\Datasets\Daxing\class_2018-12-4
# Window's size for searching similar pixels
W = 20
# Number of similar pixels for mitigating errors using neighborhood
NUM_SIMILAR_PIXEL = 20
# Number of purest coarse pixels in each class for unmixing analysis
NUM_PURE = 100
# Minimum of Digital Number (DN) value, used for correcting extreme values
DN_MIN = 0.0
# Maximum of Digital Number (DN) value, used for correcting extreme values
DN_MAX = 10000.0
# The scale factor, it is integer=coarse resolution/fine resolution, e.g., 480/30=16
SCALE_FACTOR = 16
# The value of background pixels
BACKGROUND = 0.0
# Which image including background area (0: IN_F1, 1: IN_C1, 2: IN_C2, default: 0)
BACKGROUND_IMG = 0
# Which band with value = BACKGROUND indicating background pixels
BACKGROUND_BAND = 1
# Search radius for IDW interpolator, recommend 2*SCALE_FACTOR
IDW_SEARCH_RADIUS = 32
# Power to calculate the weight of known points in IDW, if 2, Weight = 1/Distance^2
IDW_POWER = 2
cuFSDAF_PARAMETER_END