The integration of multispectral imaging data with vegetation indices (VIs) offers potential improvements in weed detection for precision agriculture. Weed detection, particularly for species like black-grass (Alopecurus myosuroides), remains challenging due to its visual similarity to crops like wheat, especially during the limited spring period in which the weed is both observable and still relevant for control. By utilizing data captured from uncrewed aerial vehicles (UAVs) over winter wheat fields, we assess the impact of key VIs-including the Normalized Difference Vegetation Index (NDVI), Triangular Greenness Index (TGI), and Excess Green Index (ExG)-on the accuracy of black-grass detection. A You Only Look Once (YOLO)-based convolutional neural network instance-segmentation detector was adapted from red-green-blue (RGB) imagery to multispectral inputs by changing the input configuration to accept either five aligned spectral bands (blue, green, red, red edge, and near-infrared) or the same five bands together with one VI channel. The results suggest that input-channel configurations incorporating VIs, particularly TGI and ExG, can improve the precision-recall trade-off compared with the baseline configuration using only the five spectral bands. Specifically, the TGI configuration achieved an average precision (AP) of 0.89 between rows and 0.55 within rows in the evaluated field subset. These results indicate that VI-channel selection can influence detection behavior, while the benefit depends on the spatial context and operating point.