Enhanced Field–Road Classification of Trajectories Using Noise-Robust Multi-Scale Feature Enhancement and Spatiotemporal Fusion

Abstract

Accurate point-wise recognition of field-operation and road/non-field states from Global Navigation Satellite System (GNSS) trajectories is essential for large-scale agricultural machinery monitoring and precision farmland management. In practice, however, GNSS trajectories are affected by measurement noise, irregular sampling intervals, and frequent state transitions. We propose DTE-SRM, a spatiotemporal classification framework that combines a U-Net-inspired, noise-robust multi-scale feature extraction network (DST-Net), a Transformer encoder, and a bidirectional recurrent module. Parking-like and scattered points are retained as valid observations and assigned to the road/non-field class, preserving complete movement sequences within a binary classification setting. The framework was evaluated on large-scale maize, wheat, and paddy datasets using a fixed, leakage-resistant trajectory-level split, together with visual analyses across ten representative Chinese provinces. Experimental results demonstrate that DTE-SRM achieves an F1-score of 92.02%, consistently outperforming all competitive baselines, and remains highly robust under synthetic GNSS perturbations, which confirms strong resilience to noise. Both quantitative evaluation and visual case analyses further verify its reliable identification of state transitions across heterogeneous agricultural environments, including fragmented mountainous fields and uniform large-scale farmlands. These results indicate that DTE-SRM provides an accurate, noise-tolerant, and generalizable solution for trajectory-based agricultural machinery state recognition, with significant potential for practical deployment in large-scale intelligent farming systems.

Keywords

GNSS trajectory classification / field-road recognition / noise-robust feature extraction / spatiotemporal feature fusion / operational scenario recognition

Rights & permissions

Higher Education Press 2026

Just Accepted

This article has successfully passed peer review and final editorial review, and will soon enter typesetting, proofreading and other publishing processes. The currently displayed version is the accepted final manuscript. The officially published version will be updated with format, DOI and citation information upon launch. We recommend that you pay attention to subsequent journal notifications and preferentially cite the officially published version. Thank you for your support and cooperation.