Brain-Inspired Perception Feature and Cognition Model Applied to Safety Patrol Robot

摘要

To satisfy the development trend of a few workers or unmanned production in industries, especially in dangerous mining industry, the safety patrol robot is required to replace the safety inspectors. The challenge is how to model the cognition mechanism of inspectors for the safety patrol. The specific problems involve the cognition modeling scheme, the aliasing of the commonly used empirical mode decomposition (EMD) of the electroencephalograph (EEG) filtering, the multisource EEG feature vector construction, and the brain-inspired modeling method. To this end, this article focuses on the perception feature and cognition model applied to safety patrol robot in mining industry. First, the inspector's cognition modeling scheme is designed by using brain-computer interface. Second, a filtering algorithm is developed by embedding the sample entropy and independent component into the EMD. Third, a multisource EEG feature vector is fused by using the power spectral density and the EEG map and the functional brain connectivity. Fourth, the cognition model is built by a convolutional neural network embedded the inception module. The experiments indicate that the modeling scheme is effective. The developed filtering algorithm increases the signal-to-noise ratio by 4.16%. The integrated model reaches the average accuracy of 88.17%.