.png){kind=icon}
.png){kind=icon}
Return
YOLO-based region of interest segmentation method for cockpit external environment
.png){kind=icon}
PRE
AI
DOI：10.1016/j.aei.2026.104573.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Accurate segmentation of cockpit Regions of Interest (ROIs) is fundamental for pilot visual assistance systems, yet current methods struggle with environmental variability and the high complexity of cockpit information. To address these challenges, this study proposes YOLO-EES, an enhanced semantic segmentation network designed for robust perception in dynamic flight environments. Specifically, we introduce wConv and EMA into the backbone to mitigate feature degradation caused by visual noise. Furthermore, the C2f module is reconstructed using Multi-order Gated Aggregation (MOGA), and combined with CPCA in the neck architecture to optimize multi-scale feature fusion, thereby improving the detection of small or obscured instruments. To validate the method, a specialized dataset was constructed based on pilot eye-tracking data collected from simulated flight tasks under varying conditions (daytime, foggy, and nighttime). Experimental results demonstrate that YOLO-EES achieves significant performance gains—improving precision by 2.75% and mAP50-95 by 3.75%—while maintaining a manageable increase in model complexity. These findings confirm the efficacy of the proposed computational framework for reliable cockpit monitoring in adverse engineering scenarios.
Keywords:
YOLO-EES
cockpit ROI segmentation
semantic segmentation
multi-scale feature fusion
pilot visual assistance
Journal
IF:
9.9
Papers:
4.0K
Citations:
1.7W