The present paper develops and analyzes an encoder-decoder neural network model based on convolutional networks. This model simulates the transduction of retinal photoreceptors for the extraction of essential image features (encoder) and their subsequent reconstruction in the primary visual cortex (decoder), replicating the biological process that humans do for image formation. The model was trained using various data configurations and deepened the analysis of synaptic weights to understand how neural networks adjust and optimize their learning. In addition, algorithms were designed to visualize and evaluate the importance of each layer and convolutional block in the reconstruction, highlighting similarities with human visual processing, where certain regions of the retina and cortex have distinct roles in the perception of colors, shapes and details.