Synthesis of morphological and neural network approaches for solving the problem of counting round timber on a digital image

Due to the rapid growth in the development of various automation systems of production processes, an increasing number of areas are experiencing the process of introducing automated systems that significantly increase labor productivity and reduce the burden on staff. In this regard, the task of creating and developing various methods of software and hardware for the designed automated systems arises. One of the industries where the growing demand for the introduction of automation systems is manifested is the area associated with measuring the geometric characteristics of various objects. There are a large number of approaches aimed at organizing the process of measuring objects, among which we can single out the contactless approach, characterized by its versatility. As a task, we can consider the task of determining the geometric parameters of round timber in stacks intended for transportation. To solve this problem, it is proposed to consider an approach based on the synthesis of a method based on a morphological approach implemented using the Canny detector and the Haaf algorithm and a neural network approach based on the use of the YOLOv5 convolutional neural network architecture. As a result of the work, these methods were developed and their synthesis was carried out, as a result of which an approach was developed that ensures high accuracy of the measurements. During the experimental studies, images of round timber obtained from the Internet were used, and to verify this approach, models of logs made using a 3d printer with predefined characteristics were used. As a result of the experiments, the accuracy of the proposed synthesized approach was estimated, exceeding the indicator of 85-90%.