Real-time extend depth of field imaging based on liquid lens and deblur-unet in automated optical inspection

Abstract

The depth of field is of significant importance in a multitude of imaging systems. It is difficult to balance imaging efficiency and quality. Here, we propose a real-time extended depth of field imaging system based on a liquid lens and Deblur-Unet for automated optical inspection. This method includes calibration, training and real time imaging. In the calibration setup, the liquid lens obtains both focus stack images and a focus scanning image in the same work distance range, and a fused extended depth of field image can be obtained by calculating the sharpness of the focus stack images. In the training setup, the fused extended depth of field image and the focus scanning image will be used to train a deconvolution and denoise Deblur-Unet. In the real time imaging setup, each frame of the focus scanning image obtained by the liquid lens imaging system is processed by the trained Deblur-Unet, and the corresponding extended depth of field image can be obtained. By using the proposed system, we extend the depth of field by >10 times for imaging in online Li-ion battery pole piece flip inspection. The system achieves a frame-rate of up to 50 Hz for 5MP or 20 Hz for 12MP mono images with an RTX 4060 GPU. Furthermore, the concise architecture of the proposed system has the potential to be used in 100MP high resolution and extend depth of field imaging with only one frame sampling, such as in screen inspection or microscopic biological imaging.