Commodity image recognition method and system based on distributed deep learning

The invention discloses a commodity image recognition method based on distributed deep learning, and the method comprises the steps: a main node obtains a plurality of commodity images {p0, p1,..., pn-1}, all the commodity images are averagely distributed to all slave nodes, the ith slave node recognizes the image sequence number set Assignnti corresponding to all the commodity images according to the need, and the image sequence number set Assignnti is set as the image sequence number set Assignnti; the length, the width and the size of each commodity picture corresponding to the picture sequence number set are reset, tensors T corresponding to the commodity pictures are obtained after standardization processing, and tensors corresponding to all the commodity pictures jointly form a tensor set {Ti}; the ith slave node sequentially inputs all tensors in the tensor set {Ti} corresponding to all the commodity pictures obtained in the step (2) into a trained deep neural network model to obtain an identification