Zhang Zhen, Chen Lin, Yuan Zhang, Gao Ling
College of Information Science and Engineering, Hohai University, Changzhou 213200, China.
Key Laboratory of Hydrologic-Cycle and Hydrodynamic-System of Ministry of Water Resources, Nanjing 210024, China.
Sensors (Basel). 2025 Jan 5;25(1):257. doi: 10.3390/s25010257.
Fast Fourier Transform-based Space-Time Image Velocimetry (FFT-STIV) has gained considerable attention due to its accuracy and efficiency. However, issues such as false detection of MOT and blind areas lead to significant errors in complex environments. This paper analyzes the causes of FFT-STIV gross errors and then proposes a method for validity identification and rectification of FFT-STIV results. Three evaluation indicators-symmetry, SNR, and spectral width-are introduced to filter out invalid results. Thresholds for these indicators are established based on diverse and complex datasets, enabling the elimination of all erroneous velocities while retaining 99.83% of valid velocities. The valid velocities are then combined with the distribution law of section velocity to fit the velocity curve, rectifying invalid results and velocities in blind areas. The proposed method was tested under various water levels, weather conditions, and lighting scenarios at the Panzhihua Hydrological Station. Results demonstrate that the method effectively identifies FFT-STIV results and rectifies velocities in diverse environments, outperforming FFT-STIV and achieving a mean relative error (MRE) of less than 8.832% within 150 m. Notably, at night with numerous invalid STIs at a distance, the proposed method yields an MRE of 4.383% after rectification, outperforming manual labeling.
基于快速傅里叶变换的时空图像测速法（FFT-STIV）因其准确性和效率而备受关注。然而，诸如运动目标误检测和盲区等问题会在复杂环境中导致显著误差。本文分析了FFT-STIV严重误差产生的原因，进而提出了一种FFT-STIV结果有效性识别与校正方法。引入了对称性、信噪比和谱宽这三个评估指标来滤除无效结果。基于多样且复杂的数据集确定了这些指标的阈值，能够在保留99.83%有效速度的同时消除所有错误速度。然后将有效速度与断面速度分布规律相结合来拟合速度曲线，校正盲区中的无效结果和速度。所提方法在攀枝花水文站的各种水位、天气条件和光照场景下进行了测试。结果表明，该方法能有效识别FFT-STIV结果并在不同环境中校正速度，性能优于FFT-STIV，在150米范围内平均相对误差（MRE）小于8.832%。值得注意的是，在夜间远处存在大量无效时空图像的情况下，所提方法校正后的MRE为4.383%，优于人工标注。
