In fluid mechanics, particle image velocimetry (PIV), a technique for determining velocity fields by following marker particles, is a routine operation. Acquired images quickly reduce accuracy with noise, non-uniform lighting, and background contamination. Traditional preprocessing techniques, such as background subtraction and spatial filtering, need manual parameter adjustment and cannot be adapted to different experimental conditions. This paper proposes a U-Net convolutional neural network deep learning approach for PIV image preprocessing automation and enhancement. For flexibility in a wide variety of situations, the network is trained on a dataset of synthetic images. The method greatly enhances velocity field estimation through improved image quality, reduced noise, and preserved important particle features. Quantitative comparison with measures such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) highlights the superiority of the model compared to traditional methods. The paper also verifies the generalisability of the system to experimental settings and its computational efficiency with promising real-world applications. This research enhances automatic PIV analysis by offering a flexible, accurate, and adaptive solution for fluid dynamics investigations in challenging experimental conditions.     «

In fluid mechanics, particle image velocimetry (PIV), a technique for determining velocity fields by following marker particles, is a routine operation. Acquired images quickly reduce accuracy with noise, non-uniform lighting, and background contamination. Traditional preprocessing techniques, such as background subtraction and spatial filtering, need manual parameter adjustment and cannot be adapted to different experimental conditions. This paper proposes a U-Net convolutional neural network d...     »