NERAL NETWORK CLASSIFIER OF AUTOMATION DOCUMENT PROCESSING SYSTEMS
10.33815/2313-4763.2021.2.25.083-091
Abstract
The work is aimed at solving the applied problem of developing an automatic system for processing electronic documents, namely one of its parts of the classifier. To solve this problem, it is proposed to use approaches to machine learning and artificial intelligence. Solving this problem under normal conditions is not difficult, but this paper considers the case of limited training sample, which is a common case in the development of systems based on the proposed approaches. The study of initial data on the basis of which the model will be taught and the number of classes to be recognized, the number of representatives in each class and the peculiarities of their presentation. The paper presents approaches, the application of which allows to increase the accuracy of systems of this type, in the conditions of limited initial initial sampling. Among the proposed approaches, the principle of minimizing the parameters in the formation of the architecture of the artificial neural network, data augmentation, pre-training of the artificial neural network by using an autoencoder. The obtained accuracy of 94–95 %, after the application of the proposed approaches in contrast to 70 % of the original, confirms the possibility of rapid development of similar classifiers of this type, with limited sampling and time minimization, achieving high accuracy.
References
2. Sokolenko D. H., Kornaha Ya. I. «Systema rozpiznavannia pysemnykh symvoliv za dopomohoiu neironnoi merezhi», Vcheni zapysky TNU imeni V.I. Vernadskoho. Seriia Tekhnichni nauky Tom 29 (68) Ch. 2 # 5 2018 s. 56-58.
3. Karpovych Artem Valeriiovych «Vykorystannia zghortkovykh neironnykh merezh dlia zadachi klasyfikatsii tekstiv», International scientific journal «Internauka» # 14(54), 2018 // Technical sciences s. 69–73.
4. Orhan G. Yalçın Image Classification in 10 Minutes with MNIST Dataset, URL: https://towardsdatascience.com/image-classification-in-10-minutes-with-mnist-dataset-54c35b 77a38d.
5. Feiyang Chen, Nan Chen, Hanyang Mao, Hanlin Hu Assessing Four Neural Networks on Handwritten Digit Recognition Dataset (MNIST) / Сhuangxinban journal of computing, june 2018, URL: https://arxiv.org/pdf/1811.08278. pdf.
6. Yifan Wang, Fenghou Li, Hai Sun, Wenbo Li, Cheng Zhong, Xuelian Wu, Hailei Wang, Ping Wang Improvement of MNIST Image Recognition Based on CNN, 7th Annual International Conference on Geo-Spatial Knowledge and Intelligence IOP Conf. Series: Earth and Environmental Science 428 (2020), URL: https://iopscience.iop.org/article/10.1088/1755-1315/428/1/012097/pdf.
7. Wan Zhu Classification of MNIST Handwritten Digit Database using Neural Network, URL : http://users.cecs.anu.edu.au/~Tom.Gedeon/conf/ABCs2018/paper /ABCs 2018_paper_117.pdf.
8. Korotynskyi, A., Zhuchenko, O. Development of a classifier for the system of automatic document processing with limited sampling, ATIT 2020 – Proceedings: 2020 2nd IEEE International Conference on Advanced Trends in Information Theory, 2020, р. 349–352.
9. A. Korotynskyi, O. Zhuchenko A system of automated control for the baking process that minimizes the probability of defects, Eastern-European Journal of Enterprise Technologies, 2020, (2-103), р. 58–67.
