UAV low-altitude remote sensing platform has become an important means to obtain high-throughput phenotypic information in agricultural field. The haze removal and quality restoration of agricultural field images are the premise and basis for analyzing remote sensing information. Aiming at the disadvantages of traditional dark channel prior algorithm, such as large amount of computation, slow operation speed and poor applicability in remote sensing image, a method based on super-pixel level dark channel prior and adaptive tolerance mechanism to improve the guided filtering algorithm was proposed. Firstly, super-pixel segmentation method was utilized to obtain super-pixel blocks with consistent color and luminance properties, and transmittance of each irregular super-pixels block was estimated.The guided filtering algorithm was introduced and improved by using adaptive smoothing parameters to refine the transmittance for the detailed edge information. Then the adaptive tolerance mechanism was added to enable the algorithm to make adaptive compensation and correction for the transmittance according to the change of the bright region of the image and the concentration of fog, after which, the optimal transmittance was acquired. Finally, the local atmospheric light estimation and adaptive adjustment mechanism were used to obtain higher quality images based on the atmospheric scattering model. The experimental results showed that the proposed algorithm can effectively recover images affected by different concentrations of fog. Six different fog concentrations of remote sensing images were taken as examples. Compared with the traditional haze removal algorithms based on dark channel prior using subjective and objective evaluation index evaluation, the proposed method had more real color and more abundant information details. Within the scope of a certain pixel, the method had high real-time performance, which can provide research foundation for the field of remote sensing image splicing and agricultural information parsing.