Instruments and Systems: Monitoring, Control, and Diagnostics Annotation << Back Accurate Multi-Class Recognition of Defects in Rivet Joints in Aircraft Products by Their Video Images Using Deep Neural Networks O.S. AMOSOV, S.G. AMOSOVA, I.O. IOCHKOV To solve the problem of accurate multiclass recognition of defects in the diagnosis of rivet joints in aviation products using their video images, the use of deep neural network learning is proposed. For machine learning, when solving the problem of detecting and classifying defects, two novel datasets were created based on a real physical model of rivet joints. The mathematical formulation and algorithms for solving the problem of recognizing defects in aviation rivet joints are given. Modifi ed deep neural networks YOLO-V5 for detection and MobileNet V3 Large for classifi cation of the state of rivet joints are used. The accuracy of the result obtained in the simulation was 100 % both in the case of binary and six-class classification. Keywords: diagnostics, defect, rivet connection, aviation technology, detection, classification, recognition, deep neural network. DOI: 10.25791/pribor.5.2022.1339 Pp. 30-41.