In order to improve the localization accuracy of SLAM task of UAV in weak texture scene, a visual VO system based on point-line feature fusion, PLK2-SLAM, is proposed based on ORB-SLAM3 system. In this system, according to the statistical analysis of feature similarity, the features with large differences are retained to achieve the de-noising of features and improve the quality of features. In the feature matching process, the two-thread point feature matching method is used, the optical flow method is used for tracking features, and the LK2LBF method designed in this paper is used for feature reextraction, so as to reduce the calculation amount. A vector-based reprojection error method for line features is proposed to improve the accuracy of pose estimation. Finally, the performance of the PLK2 -SLAM system is verified on the open data set TUM. In the weak texture scene, the accuracy is improved by 84% compared with the classic point-and-line system PL-SLAM, which improves the stability and robustness of the ORB-SLAM3 system.