Industrial meter identification method based on multi-frame vision

The invention discloses an industrial meter identification method based on multi-frame vision. The industrial meter identification method comprises the following steps: 1, continuously shooting N photos towards a preset target near a preset shooting point by utilizing a shooting device; 2, extracting depth features of the N photos through the same set of convolutional neural networks to obtain feature tensors of N HxWxCs, wherein H is the height, W is the width, and C is the length; 3, respectively normalizing the characteristic tensors of the N HxWxCs into an intensity matrix of HxW by utilizing a trained full connection layer; and 4, normalizing each intensity matrix through softmax, and calculating the weighted sum of the characteristic tensors of each characteristic tensor HxWxC through weight to obtain N new vectors with the length of C as target characteristic vectors x '. According to the invention, the problems of meter identification failure and errors caused by accidental shooting factors are greatl