Abstract:Recently, many artificial intelligence network models have achieved excellent performance on image classification tasks. However, these models do not take advantage of knowledge about how memory works in the human brain. Most deep learning methods are designed and run based on primary visual networks. With the further deepening of cognition in the field of brain science, the biological mechanism of the neocortex, such as the hippocampus and perirhinal cortex with long-term semantic memory, has been found to play an essential role in the human recognition of images. Therefore, we propose a semantic memory neural network model based on two-dimension coding method(SMNN-TDC), which is used to simulate the long-term semantic memory operation mechanism of the human brain to recognize visual images and obtain key target semantics, further improving the recognition of important image features. Specifically, the method improves the residual network and adds an attention module to simulate the operation process of the memory mechanism of the human brain. The improvement of the residual network is based on the mechanism of the interconnection of the various brain subsystems, and the attention mechanism is added because each visual cortex has the function of perfecting edge information. Final image classification experiments show that our framework can capture good target semantic features and perform well on benchmark datasets.

Semantic Memory Neural Network Model Based on Two-Dimension Coding Method