Robust Algorithm for Dress Recognition of Substation Operators Based on Sensitivity Radial Basis Function Neural Network

At the substation job site, the use of intelligent video surveillance technology can greatly reduce the supervision burden on safety inspectors for irregular operations of operators. However, in the outdoor complex working environment, the identification accuracy of recognition algorithm based on the traditional radial basis function neural Network (RBFNN) is not high enough and the missing alarm rate is high. In order to solve this problem, this paper proposes the RBFNN robust algorithm for dress recognition based on classifier output sensitivity. The algorithm firstly extracts the shape and color feature vector of the helmet, the top and the bottom of the operator image. Then, the Monte Carlo method is used to randomly sample the points in the neighborhood of training samples to expand the number of samples and reduce the volatility of the classifier output. And then, the loss function that considers the sensitivity of the sample neighborhood is established. Finally, the weights from the hidden layer to the output layer are solved by Gauss-Newton method. The RBFNN classifier based on Gaussian function is established. The simulation results show that the recognition algorithm based on sensitivity RBFNN (S-RBFNN) can effectively reduce the missing alarm rate, which is more robust in practical applications.