Neuromorphic vision system

A novel neuromorphic vision system,

• comprising a retinomorphic array and a neural network, wherein the retinomorphic array and the neural network structurally use a three-dimensional (3D) vertical integration, the retinomorphic array is used to convert visual information into electrical signals, and the neural network performs information processing on input electrical signals to obtain a result of visual cognition, and the neural network is configured by a nonvolatile crossbar array with adjustable resistance;
• wherein the visual information is dynamic visual information, and observed dynamic target coordinate information is converted into two voltage signals by the retinomorphic array and input into a pre-trained nonvolatile crossbar array, and during an inference process of the neural network, each nonvolatile memory of the nonvolatile crossbar array, receives an input of a neuron from a current moment from the neural network and an output from a hidden layer neuron at a previous moment, performs a dot product operation and conducts an output of the neuron at this moment.