Development of a Classifier for the System of Automatic Document Processing with Limited Sampling

the work is aimed at solving the current problem of automatic recognition and processing of scan/photo of documents. To solve this problem, the work uses approaches to artificial intelligence, namely artificial neural networks. The main difference between this work and the existing ones today is the solution of the described problem in conditions of limited sampling. The approaches of neural networks pre-learning using the well-known structure of the autoencoder, data augmentation and minimization of parameters were used to achieve an effective solution at minimal cost.