Abstract: In an actual land battle environment,people cannot carry large computing devices such as GPUs with them.Therefore,it is more difficult to calculate the large-scale neural network parameters,which further leads to the target recognition network not working in real time.To this end,a target recognition algorithm based on lightweight convolutional neural network (E-MobilNet) is proposed.In order to improve the network learning effect,based on the existing target learning framework MobileNet-V2,an ELU function is inserted as an activation function.Firstly,use the expansion convolution to increase the number of channels to get more features to activate and output through the ELU function,which can alleviate the disappearance of the gradient of the linear part,the nonlinear part is more robust to the noise of the input change.Then,the way of the residual connection Combine high-level features with low-level features and then output.Finally,output to Softmax using global pooling.The experimental data shows that compared with the current mainstream lightweight deep learning target recognition algorithm,E-MobileNet has improved the accuracy of recognition and the frame rate per second in the same test environment of the same test set.The experimental data fully demonstrates that the use of the ELU activation function and the global pooling layer reduces the number of parameters,enhances the generalization ability of the model,and improves the robustness of the algorithm.On the basis of ensuring the lightweight of the neural network model,the recognition accuracy of the target is effectively improved.

Key words: Activation function, Lightweight model, Residual network, Separable convolution, Target recognition