Abstract: Unmanned aerial vehicle (UAV) image stitching technology provides crucial support for the development of the low-altitude economy through efficient integration of aerial data. To address issues such as low registration accuracy, misalignment, and ghosting caused by insufficient feature point extraction in low-texture UAV images, an \ell_1 -norm and gradient-constrained UAV image stitching method was proposed. First, feature points and feature lines of the target image and the reference image were jointly extracted to construct a multi-feature descriptor, enhancing matching robustness and effectively improving image misalignment. Second, the \ell_1 -norm was used for color difference measurement, and an energy function was constructed with gradient constraints to guide the seam to preferentially pass through highly similar continuous regions. Finally, the graph-cut algorithm was applied to search for the optimal stitching path in the overlapping area, and Poisson blending was employed to achieve a natural transition at the stitching boundary. Two sets of typical drone image datasets were selected, and comparative tests with three mainstream methods (SPW, LPC, and MSF) were conducted to verify the superiority of the proposed method in terms of stitching accuracy and visual effects.The results show that compared with SPW, LPC, and MSF, the SSIM values of the proposed method are improved by 2.97%, 5.87%, and 3.07% respectively, while the PSNR values are increased by 0.595, 0.848, 0.841 dB respectively. In terms of visual effects, misalignment and ghosting during the stitching process are significantly improved by the proposed method, with structural integrity and texture details of objects being better preserved, resulting in enhanced overall quality of UAV image stitching. Both quantitative and qualitative analyses fully demonstrate the superior performance of the proposed method in low-texture scenarios.