Human Action Recognition by Deep Learning Technique with Multiple-Searching Genetic Algorithm

The purpose of this paper was to review the hybrid method in the human action recognition approach utilizing deep learning and multiple-searching genetic algorithm. This study reported that the deep learning classifier stated the weights of the convolutional neural network in light of the arrangements created by reducing the classification error method from the multiple-searching genetic algorithm. Gradient descent was utilized to find a local minimum and train the convolutional neural network classifiers amid the fitness assessments of the chromosomes from the multiple-searching genetic algorithm. The local search capacities of the gradient descent and global search capacities of the multiple-searching genetic algorithm used could discover a solution that was nearer to the global optima. Furthermore, this paper demonstrated the confirmations of the classifiers produced by utilizing the multiple-searching genetic algorithm to enhance the execution. Moreover, the adequacy of the proposed arrangement framework of the human action recognition system was shown by using the action bank features from the UCF50 dataset.