Abstract: High-precision leaf segmentation and extraction is a crucial prerequisite for establishing flue-cured tobacco agricultural monitoring algorithm models based on image data. To accurately extract tobacco leaf image information in farmland environments, this study developed the AO-YOLOv8 algorithm, which can precisely segment and extract tobacco leaves from multispectral images in farmland environments. The algorithm first enhanced the C2f module in the backbone network of YOLOv8. By adding the Aggregated Attention module, global image information can be captured completely, thereby ensuring the integrity of leaf segmentation. Secondly, the C2f module in the neck of YOLOv8 network was replaced with a GELAN module, which was further refined using the Reparametrized Convolution module (OREPA). This dual modification successfully reduced the number of model parameters while simultaneously improving the leaf segmentation accuracy. The test results showed that the AO-YOLOv8 model achieved 87.3% mAP50(Mask), 89.2% mAP50(Box), and 83.4% mIoU on the multispectral tobacco leaf instance segmentation test set, which represented an improvement of 6.99%, 8.38% and 8.88% over the baseline YOLOv8 model across these three metrics, respectively. Furthermore, compared with the classical instance segmentation models, AO-YOLOv8 showed improvements of 8.72%, 10.26%, and 7.61% over Mask R-CNN, and 11.92%, 11.64%, and 17.13% over YOLACT on the same metrics. AO-YOLOv8 can accurately segment tobacco leaves in multispectral images under farmland environments, making it well-suited for intelligent monitoring of flue-cured tobacco growth and development using spectral image data.