Evaluation of the accuracy of pattern recognition by a neural network with various filters in the receptor layer of the retinal simulation module

ICLR 2026 Conference Submission20830 Authors

Keywords: Pattern recognition, classification, neural networks, retina, fully connected neural networks, supervised learning

Abstract: The purpose of this work is to evaluate the effect of the location of receptors in the first layer of the retinal simulation module on the ability of a neural network to recognize images. The retinal simulation module serves as a means for preprocessing images. The retinal simulation module is described and compared with existing popular preprocessing methods. The module processes the image using three layers. The object of this study is the first layer of the module, which simulates the receptor layer of the real retina of the human eye. The experiments were conducted on a fully connected neural network. The retinal simulation module preprocessed a sample of fruit images photographed from different angles, which was then fed to the input of the neural network. In the process, ninety-four experiments were performed with different module settings. In each of the experiments, the settings of the fully connected neural network remained unchanged. The results of image recognition by a neural network are presented. Recommendations are given for configuring the receptor layer of the retinal simulation module to improve the accuracy of pattern recognition.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 20830