Application of an Incremental Method of Neural Network Training for Recognition of Symbols on Railway Wheels

The paper discusses the use of incremental neural network training for recognition of images of symbols on a metal surface. Incremental learning allows you to adapt to changing shooting conditions, increase recognition accuracy and eliminate re-training on the entire data set. Experimental results demonstrate the effectiveness of incremental neural network training. A set of images of marking symbols obtained on a production line to produce railway wheels was used as initial data.