Agricultural machinery operation trajectory identification and operation area estimation for cloud-platform...

Precise identification of agricultural machinery operation trajectories and efficient estimation of operation area are essential for cloud-platform-based machinery supervision and service settlement. However, raw GNSS trajectories collected from practical operation platforms are often affected by positioning drift, missing points, road-transfer trajectories, headland turns, and repeated or pseudo-missed operations, making it difficult for either purely rule-based trajectory screening or direct buffer-based area recovery to simultaneously achieve reliable recognition accuracy and engineering efficiency. To address this problem, this study proposes a staged trajectory identification and area estimation method for cloud-platform deployment. The method first standardizes raw trajectory sequences through attribute integrity checking, motion rationality filtering, speed cleaning, and temporal interpolation. Candidate operation trajectories are then organized using spatiotemporal neighborhood constraints to remove evidently non-operational long-distance transfer trajectories before image construction. A multi-channel trajectory image representation, in which speed, acceleration, and heading variation are encoded as feature channels, is further used as the input of an improved CBGAM U-Net semantic segmentation model for pixel-level refinement of field-operation trajectories. Finally, a cubic-spline-smoothed vector-buffer algorithm with width compensation and inward boundary correction is used to reconstruct operation coverage areas. Experimental results on the IAEMP platform dataset showed that the proposed method achieved an average trajectory recognition accuracy of 96.32%. In parcel-level area validation, the absolute area estimation errors of the tested field parcels were controlled within 3.00%, and the cloud-platform deployment test showed stable processing efficiency for practical operation records. In independent Real-Time Kinematic (RTK)-based field validation, the area accuracies reached 99.64% for wheat harvesting and 99.91% for rotary tillage. These results demonstrate that the proposed staged framework can improve trajectory identification reliability and area estimation consistency while maintaining practical deployability for agricultural machinery management platforms.