Deep Learning-based Auto-focus Method for Liquid Lens

The liquid lens is characterized by its small size and rapid response, making it highly suitable for diverse applications in micro-nano operations and medical care. However, for liquid lenses with unique focusing principles, current research on auto-focusing methods still relies mainly on traditional auto-focusing techniques. Thus, there is a pressing need for more efficient and precise auto-focusing methods that can fully leverage the advantages of liquid lenses. This study presents a novel auto-focusing method for liquid lenses based on deep learning. We introduce the principles of liquid lenses and utilize Gaussian functions to process the dataset to simulate the blurred images captured by liquid lenses. Subsequently, we trained a fine-tuned VGG-16 network using the preprocessed dataset and evaluated its performance on images captured by the liquid lens. Experimental results demonstrate that our proposed approach achieves a classification accuracy of 95.87% for blurred images, and the mean squared error for the predicted voltage values on the captured images is only 0.0154.