Flaw detection of the inner surfaces of especially critical pipes is based on visualization, which requires complex optical systems equipped with artificial intelligence functions. Training of such systems is very complicated due to the limited volume of defective products. The paper describes training and testing of machine learning algorithms in poor data by the example of detecting a defect at the inner surface of a pipe. The authors propose a method for developing a set of synthetic training images obtained using 3D models of technical objects with parameterized defects applied to them. Images can be generated by parametric description of the artificially defected inner surface of a pipe 3D model in Autodesk Inventor environment. Windows AutoIt OS automation environment is applicable to generate synthetic images and masks. The method allows obtaining a set of synthetic pipe images for training neural networks with U-Net and LinkNet architectures. The trained neural networks testing has shown the defect recognition at a high level both on a synthetic sample of images and on real images of inner surface of rejected pipes.
This article is devoted to the development of an algorithm for segmentation of visual signs of DR and DMO. The paper references the global statistics of patients with diabetes mellitus and their need for regular fundus screening. We propose the use of telemedicine applications to the problem of regular ophthalmological screening of patients with diabetis mellitus. The main features of DR and DME are identified with the help of artificial intelligence algorithms. A list of scientific and technical problems that needed to be solved is presented: the collection of training data, their markup and the choice of artificial neural network architectures for the tasks of feature segmentation. The process of validation of the algorithm is described and the current results are presented.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.