Research on vegetation coverage estimation based on panchromatic and multispectral remote sensing image fusion

Remote sensing technology has become an important tool in grassland science research, and its application in estimating grassland vegetation coverage has been strengthened. However, due to the limitations of remote sensing imaging sensors and the high cost of acquiring high spatial and spectral resolution images at the same time, high-precision estimation in the field of grassland observation is difficult to obtain from a single remote sensing data source. Therefore, it is necessary to integrate the key information of the source image and fuse remote sensing image data of different resolutions, so that the fused image has higher clarity, richer texture, and more detailed spectral information, thereby improving the accuracy of vegetation coverage extraction. In this study, the panchromatic and multispectral remote sensing images collected by GF-1 were used as the research objects. A series of preprocessing steps including radiometric calibration, atmospheric correction, orthorectification, image registration, and cropping were performed. Seven component substitution-based methods such as principal compontent analysis (PCA) and intensity-hue-saturation (IHS), five multi-resolution analysis-based methods such as Wavelet, and two deep learning-based methods, Pansharpening by convolutional neural networks (PNN) and PanNet, were used to fuse the panchromatic and multispectral remote sensing images, and comparative analysis was conducted. An improvement strategy was proposed and validated for the optimal fusion algorithm PanNet, and the results showed that all indicators of the improved PanNet algorithm were better than those of the original PanNet algorithm. Finally, the fused image was applied to estimate vegetation coverage, and the operational feasibility and superiority of the improved PanNet remote sensing image fusion algorithm in vegetation coverage estimation were demonstrated.