ABSTRACT
The river discharge is the most important information in the water resource management. Large-Scale Particle Image Velocimetry (LSPIV), one of the non-contact techniques, is widely used to estimate the discharge. However, the environmental conditions are usually complex and not controllable in the field. These unfavorable factors make the difficulty to use the proper setting for IA and directly raise the errors in the surface velocity measurement, while conducting direct cross-correlation algorithm (DCC). To improve aforementioned errors, the Convolutional Neural Network (CNN) powerful on the object detection was conducted to take more surface characteristics on open channel flow into account. Several scenarios were designed to simulate the most common noises happened in fields. The velocity measurements done by CNN were compared with the ADV measurement. The effect of IA size and CNN layers toward velocity estimation was evaluated to show the applicability of CNN for velocimetry in the open channel flow.
