Sub-pixel mapping in farming area remote sensing image based on improved spatial gravity model

Due to the limitation of the sensor spatial resolution and the complexity and diversity of objects, mixed image pixels generally exist in remote sensing images. Pixel unmixing can only get the composition ratio of each endmember in the pixel, rather than the spatial distribution of each endmember. Sub-pixel mapping was proposed to solve above-mentioned problem. Spatial gravity model is an iterative solution of sub-pixel mapping which is based on the sub-pixel scale, the spatial correlation is expressed by gravitational relationship between sub-pixels and neighboring mixed pixels. This model does not require complicated parameters and its calculation is relatively simple, so it has the advantages of iterative solution and has the potential to improve mapping accuracy and speed. From the discussion above, this paper proposes a sub-pixel mapping method based on improved spatial gravity model for farming area remote sensing image. Firstly, this paper analyses the initialization algorithm and optimization algorithm of original spatial gravity model. The original initialization algorithm uses random assignment, which affects the calculation accuracy of neighboring mixed pixel gravity values, decreases the mapping accuracy, increases the number of iterations of the whole model, and decreases the overall speed of the model; Based on original initialization algorithm, the original optimization algorithm also affect the model accuracy and speed to a certain extent. Secondly, this paper improves the initialization algorithm and optimization algorithm of spatial gravity model. Improved initialization algorithm enables the model to combine the advantages of direct solution and iterative solution, after initialization data have more spatial correlation, the initialization accuracy and speed are improved compared with random assignment; Improved optimization algorithm optimizes data on the basis of initialization, greatly reduces the number of iterations and improves the speed. Lastly, this model was used to analyses the farming area remote sensing image in Zhenlai county, Jilin province, and a remote sensing image sub-pixel mapping experiment was conducted with the original image spatial resolution degraded by four times. Every 4×4 pixel value in original SPOT-5 remote sensing image was averaged once according to weight to make the spatial resolution degrade from 10 to 40 m, and the original spatial gravity model and improved model was used to map the degraded image sub-pixel. The results indicate that compared with the original one, the improved model can improve the precision of sub-pixel mapping by 6.67% and increase the operation speed by 10.69 times. Therefore, the improved model can break through the limits of spatial resolution in remote sensing image of farming area with relatively complex objects, and effectively bolster the precision of the crop planting area extraction and remote sensing-based regional yield estimation.