Research on Distinguishing Different State of Levitated water droplet Based on CNN algorithm

Ultrasonic Levitator can levitate small object due to the node created by several waves’ interference. [1] Sometime, due to some external variations, the small object will become unstable and start to vibrate. This article first demonstrates how to construct a standardized database for transparent water droplet. The evaluation of the database includes background selection, instrumentation parameter setting, and comparative experiments. Then, the database is studied by a Convolution Neural Network for observing the stability of water droplet. The CNN model is revised due to the observation outcome, and various changes, including adding layer, changing parameter, adding inception, etc, have been made to further improve the accuracy of the model. Finally, some risk mitigations of the potential issues are discussed to ensure no wrong recognition or judgement made by the CNN mode. As this levitator technology can be used for pharmacy, new material, the control system will enable better observation of materials. [2] Moreover, the database can be used for analyzing transparent object classification problems.