To address the issue of sharply increased detection errors in RGB images caused by vegetation occlusion shadows and sandy soil spectral interference during unmanned aerial vehicle(UAV) remote sensing identification of planting pits in the Three-North Shelter Forest Program, this study proposes a YOLOv11-SimAM-NAM model that combines multispectral vegetation indices(MS-VIs) and spatial features. Three innovations were implemented based on the YOLOv11:1)Construction of a fused normalized VIs space (NDVI+OSAVI+MSR), which enhances the spectral separability between planting pits and vegetation-covered areas through three-channel feature stacking, enabling morphological feature extraction under vegetation occlusion;2)Design of a spectral-spatial collaborative enhancement module that leverages the high-resolution geometric details of RGB images to compensate for spatial information loss in MS data, establishing a cross-modal feature transfer channel;3) Development of a dynamic weight allocation mechanism to auto-adjust MS feature weights in vegetation coverage regions, reducing the false detection rate in bare soil areas. Experiments indicated that MS-VIs fusion improved AP@0.5 to 0.977, denoting a 12.3 % increase over single-source RGB detection. The feature fusion strategy enhanced recall by 18.7 % in scenarios with 90 % vegetation coverage, while the F1-score remained at 0.93 +/- 0.02 under shadow interference. Studies revealed that using MS data alone led to a 19.4 % lower AP@0.5:0.95 compared to the fused model due to limited spatial resolution and loss of fine-grained features, confirming the necessity of multisource data synergy. This study validates the critical role of MS-VIs in overcoming complex vegetation interference. The proposed fusion framework provides a novel solution for UAVbased intelligent monitoring in arid region ecological engineering, offering practical significance for improving desertification control efficiency.