CNN Multi-Position Wearable Sensor Human Activity Recognition Used in Basketball Training

With the development trend of artificial intelligence technology and the popularization of wearable sensors, human activity recognition based on sensor data information has received widespread attention and has great application value. In order to better optimize the network structure and reduce the total number of main training parameters in the convolutional layer, a convolutional network entity model based on shared resources of main parameters is clearly proposed. We analyzed the CNN multi-position wearable sensor human activity recognition used in basketball training. According to the entity model of the main parameters of shared resources, the effectiveness of the entity model is verified from both the total number of sensors and the accuracy of single-class recognition. In addition to maintaining the actual effect of recognition, the main training parameters are also reduced. The simulation results verify the actual effect of the SVM algorithm and motion simulation of the convolutional network entity model. On this basis, scientific research physical exercise methods are selected to reasonably ensure the smooth progress of appropriate physical exercise at a certain level, improve the quality of training and the actual effect.