期刊名称：International Journal of Computer Science and Network Security
印刷版ISSN：1738-7906
出版年度：2020
卷号：20
期号：3
页码：29-36
出版社：International Journal of Computer Science and Network Security
摘要：In recent decades, the development of technical vision systems (TVS) has been actively conducted to automate production. To increase the effectiveness of the functioning of TVS, it is necessary to constantly replenish the arsenal of methods and means of preliminary image processing and the construction of classifiers that combine the required indicators for speed and reliability of identification. The main difficulty in developing neural networks (NN) with desired properties is the lack of theoretical design of their topologies and the prediction of recognition accuracy. In this regard, the study of the possible achievable accuracy for the NN recognition of mass objects with strong intra class visual variability and the development of appropriate methods to guarantee a sufficient specified accuracy is an urgent.
关键词：technical vision systems; neural networks; theoretical design; recognition accuracy; reliability of identification; preliminary image processing; construction of classifier; pattern recognition; system analysis; computer vision.
