Designing a Neural Network Cascade for Object Detection in Drawing and Graphical Documentation Processing

The article is devoted to the problem of recognizing images of different types of rolled metal products during the analysis of the drawing and graphic documentation, which is the basis for the production enterprise to manufacture metal structures. Determining the rolled product type and calculating its consumption solves the problem of automation of keeping stock records and product costing, when it comes to the use of non-unit items, which consumption is measured in meters. Two processing methods were chosen to solve the problem - automatic parsing and neural network cascade, the assembly of which made it possible to obtain a high quality of object search.