Hybrid Activation Function in Deep Learning for Gait Analysis

Various values are carried out in Neural Network Connections, to check whether the neurons are activated equally or not and the activation layer is called to conduct the process statistically balanced. Mathematical functions like ReLU, softmax and sparsemax are used in Activation Layer. In this paper the investigator combined softmax and sparsemax in the last activation layer of the Deep Learning Convolutional Neural Network for gait analysis using silhouettes. Human identification process is done without the conscious of the person. The researcher used gait features to predict the person. Python Language that runs over tensorflow is used to implement the functions in activation layer. Benchmarking dataset CASIA Band C is used for better performance. The combined hybrid formula is implemented in the last activation layer along with probability calculation. The results reveals that the hybrid usage of both softmax and sparsemax function in the activation layer helps in better performance and provides high accuracy while comparing the individual usage of functions separately.