Detection of Srew Nut Images Based on Deep Transfer Learning Network

The rising of deep learning has great application value for the accurate recognition of industrial components. Based on Inception-V3 model, the network use suitably factorized convolutions and regularization to scale up networks, this method will make the added computation more efficient. When using transfer learning method on this modified neural network model, we locate the screw and nut with high recognition accuracy. This paper retrains the pre-trained weights fixed and tune some part of layers, we use a small learning rate to insure this model can learn the previously obtained knowledge, which can reduce the training time and solve the problem of insufficient datasets.