A banknote serial number recognition method for banknote detectors

Abstract

The invention discloses a banknote serial number recognition method for a banknote detector. The banknote detector includes a processor, a memory, an upper image sensor and a lower image sensor, and the processor is electrically connected to the memory, the upper image sensor and the lower image sensor respectively. , the banknote detector is electrically connected with the host machine; comprises the following steps: the present invention includes a rough positioning of the serial number; extracts each pixel point coordinate (fx, fy) of the area where the serial number is located, and utilizes 4 pixel points around (fx, fy) Perform bilinear interpolation fitting to obtain the sub-graph A' of the serial number of banknotes; perform binarization on the sub-graph A' of the serial number, and use the method of 8-neighborhood contour extraction to extract the features of each target in A' Vector F; identification and output of serial numbers The present invention has the characteristics of fast extraction of serial numbers and high recognition accuracy.

Description

technical field [0001] The invention relates to the technical field of banknote detectors, in particular to a banknote serial number recognition method for banknote detectors with fast serial number extraction speed and high recognition accuracy. Background technique [0002] For banknote serial number recognition, regardless of the classification and recognition method, extracting stable features is the prerequisite for recognition. [0003] Usually, the neural network method is used to identify the crown number. The neural network method has the following advantages and disadvantages: [0004] A large number of samples are required for training. When a character is misjudged, it can only be attributed to the fact that the network is not well trained, and then re-collect batch samples for training; or the network structure is unreasonable, adjust the number of network layers and nodes, and the activation function, etc. . If some characters are wrongly judged, in order to ...

AI Technical Summary

Problems solved by technology

If some characters are wrongly judged, in order to change this state, a lot of work needs to be invested in retraining. However, there is no guarantee that the newly trained network will have good performance, which will lead to a vicious circle.

[0005] If the entire character image is input into the neural network, the overall features will be similar, and the recognition rate of characters with obvious local features will be low

Images