AUTHOR=Yuan Xinru , Yu Haiyang , Geng Tingting , Ma Ruopu , Li Pengao TITLE=Enhancing sustainable Chinese cabbage production: a comparative analysis of multispectral image instance segmentation techniques JOURNAL=Frontiers in Sustainable Food Systems VOLUME=Volume 8 - 2024 YEAR=2024 URL=https://www.frontiersin.org/journals/sustainable-food-systems/articles/10.3389/fsufs.2024.1433701 DOI=10.3389/fsufs.2024.1433701 ISSN=2571-581X ABSTRACT=Accurate instance segmentation of individual crop is of great importance for field management and crop monitoring in smart agriculture. Traditional remote sensing methods primarily focus on overall population analysis, lacking accurate estimates for individual crop. To address this issue, this study proposes an instance segmentation scheme based on UAVs and the YOLOv8-Seg model for efficiently extracting features of individual Chinese cabbage plants.The YOLOv8-Seg model supports the generation of independent segmentation masks and detection results at different scales. It employs Path Aggregation Feature Pyramid Networks (PAFPN) for multi-scale feature integration and optimizes sample matching through the Task-Aligned Assigner, enhancing the stability and generalization of the training process.We collected multispectral data of Chinese cabbage using UAVs and built a high-quality dataset through semi-automatic annotation with the Segment Anything Model (SAM). The study uses mAP (mean Average Precision) under different thresholds as the evaluation metric, comparing the YOLO series algorithms with other mainstream instance segmentation algorithms. Additionally, we analyzed the performance of the model under different spectral band combinations and spatial resolutions.The results demonstrate that YOLOv8-Seg achieved 86.3% mAP50-95 under the RGB band. Even at lower spatial resolutions (1.33 ~ 1.14 cm/pixel), the model maintained high segmentation accuracy and successfully extracted the number and average leaf area of Chinese cabbage. This study highlights the superior performance of combining UAVs with YOLOv8-Seg for individual crop instance segmentation, providing strong technical support and promising applications for precision agriculture.