Fusion of Deep Convolutional Neural Networks and Brain Visual Cognition for Enhanced Image Classification

Abstract

The brain visual system is one of the core centers for human perception of external information. How to establish the brain visual cognitive system to classify and process image information is a key matter in the area of human-computer connection. In order to improve the accuracy of computer vision image classification, a fusion intelligent computing model based on deep convolutional neural network and brain visual cognition is proposed. This model simulates the visual processing mechanism of the human brain and uses brain computer interface technology to extract electroencephalogram signals, thereby achieving efficient classification and processing of image information. When designing an image classification model based on DCNN, a long short-term memory network structure is introduced to extract time series features of electroencephalogram signals. In order to enhance the classification accuracy of the model, attention mechanism and occlusion independent neural response methods are also applied to improve the accuracy of capturing the correlation information between brain response and image features. The results show that the prediction accuracy of the research model reaches 93.54% and 94.03% in the V4 visual region and L0 visual region, respectively. The highest accuracy on facial visual images reaches 95.46%, while the lowest accuracy on animal visual images is 91.57%. By introducing the long short-term memory module, the loss value of the model decreases from 0.26 to 0.21, with a reduction of 19.23%. In addition, ablation experiments show that by introducing attention mechanisms and occlusion independent neural responses, the final classification accuracy is improved to 93.94%. In summary, the research on the fusion intelligent computing model grounded on deep convolutional neural networks and brain visual cognition effectively improves the accuracy of image classification and demonstrated its potential in the field of intelligent computing.