Space-Time Image Velocimetry Based on Improved MobileNetV2

Space-time image velocimetry (STIV) technology has achieved good performance in river surface-flow velocity measurement, but the application a field environment is affected by bad weather or lighting conditions, which causes large measurement errors. To improve accuracy and robustness of STIV, we combined STIV with deep learning. Additionally, considering light weight neural network model, adopted MobileNetV2 improved its classification accuracy. We name this method MobileNet-STIV. also constructed sample-enhanced mixed dataset for first time, 180 classes images 100 per class to train our resulted performance. Compared current meter results, absolute error mean was 0.02, flow discharge 1.71, relative 1.27%, 1.15% comparative experiment. In generalization experiment, 0.03, 0.27, 6.38%, 5.92%. The results both experiments demonstrate that more accurate than conventional large-scale particle (LSPIV).