Detection for Mixed-Characters Based on Machine Learning

We propose a new method based on a machine learning technique to detect the mixed-characters on the surface of all kinds of cables applied in industry. Firstly, the images of these characters are captured by a high precision CCD, and then, the captured images need to be preprocessed, normalized, and divided into multiple images. Each image includes a single character. Finally, the training set image is sorted and optimized. We establish a convolutional neural network for the characters recognition, and its parameters are improved based on character features. The average recognition rate of mixed-character is 92.6%. Experimental results show a recognition rate based on machine learning could be higher than the one by using other algorithms.