| abstract |
The invention relates to a vehicle identification method based on CNN and domain self-adaptive learning. By adding a rotation invariant layer, distinguishing a discriminant layer and designing a new objective function in the Alexnet network, an initial model based on the CNN network is established; The feature map of the convolutional layer of samples in different fields, calculate the cosine similarity between sample feature maps, determine the shared convolution kernel or non-shared convolution kernel of the CNN network, retain the weight and bias of the shared convolution kernel, and update the non-shared convolution kernel The weight and bias of the convolution kernel; based on the training samples in the target field, calculate the cosine similarity between the feature maps of each layer and the average similarity of the entire target field, and cluster each similar feature map according to the average similarity; The source domain samples with similar distribution characteristics in the medium samples are expanded to new samples in the target domain, and the new samples in the target domain are used to fine-tune the entire CNN network model, and then the test samples in the target domain are classified by softmax classifier. |