Abstract: How the brain processes the flow of sensory data over time and space to perform complex cognitive computations and adaptive behaviors has long been a difficult problem in the field of brain science and brain-inspired technologies. This problem is called the expression and computation problem of sensory encoding, and the key is to find a general method to realize the similarity or invariance expression learning in sensory data flow, and effectively apply it to cognitive computing and behavior control, as well as perception and consciousness formation. However, existing experiments, theories, and methods are difficult to explain the relationship between brain structure and function with biologically complex mechanisms, the relationship between the expression and computation of brain-inspired intelligence in information theory, and the relationship between the two. On the basis of illustrating the physiological mechanism of neural coding in the passageway from retina to lateral geniculate nucleus and to visual cortex, we propose a brain-inspired intelligence unit architecture for the expression and computation of the similarity or invariance of the spatiotemporal activity pattern of sensory data flow. The architecture is similar to in the human brain’s the latent representation of peripheral nervous systems, the systematic separation and concatenation of thalamus, the computation of the independent sparse coding and self-organizing mapping of cerebral cortex, and the oscillations and their synchronizations of the re-entrant connections between cortical functional regions. This brain-inspired intelligence unit realizes a large-scale integrated neural network model, which helps to better understand the representation and computation of the brain, and contributes to the global brain topographic map coding of brain consciousness, as well as be beneficial to brain-inspired intelligence realized by large-scale integrated neural networks.