Использование методов машинного обучения для решения задачи поиска дефектов на растровых изображениях

Abstract

The problem of defects detection on images of different classes is considered. Two variants of trained classifiers – direct propagation neural networks and convolutional networks are considered for solving this problem. A distinctive feature of the analyzed images is the presence of objects located on a textural background. Algorithms for defects detection are designed to identify artifacts such as applicative errors, distortions of object boundaries or their parts. Test results for various types of images are given