Improved MobileNet-SSD Recognition Method for Electric Vehicles in Elevators
Since the charging of electric vehicles in the home is likely to cause a fire risk,the traditional video detection method re-quires data upload,which is likely to cause data loss and high latency.In response to the above problems,this paper proposes an improved MobileNet-SSD method that can be run on embedded devices to identify electric vehicle targets in elevators.In the data preprocessing stage,the CycleGAN method is used for data enhancement to improve the generalization ability of the model.Aim-ing at the problem of excessive calculation of the MobileNet network,this paper proposes to introduce the width multiplier and the resolution multiplier as hyperparameters to reduce the amount of model calculations.After optimizing the hyperparameters by the BOHB(Hyperband-Bayesian Optimization)method,the optimal hyperparameter combination is obtained.At the same time,in order to solve the problem of the weight cannot be updated and the information loss,LReLU is used instead of ReLU as the activa-tion function of the model.Experimental results show that the improved MobileNet-SSD algorithm in this paper can quickly and accurately identify electric vehicle targets on embedded devices.The improved MobileNet-SSD model can increase the map of the original SSD model by 6% and reduce the response delay by 20%.
万方数据
维普
