A Noval Image Sensing Bionic Pre-encoder

Image preprocessing is important for deep learning image recognition tasks, and this paper introduces a new method called a bionic precoder, inspired by the human retinal visual system, to improve deep learning image recognition models. The precoder simulates a non-uniform distribution of retinal photoreceptors and iterates with trainable parameters. The trained precoder performs retinal-like preprocessing to improve the quality of the input data. Significant improvements in the accuracy/recall curve and receiver operating characteristic curve were observed experimentally. This study presents a promising technique for incorporating precoders into deep learning, leading to more accurate and efficient image recognition models.