Algorithm for Classification of Fire Hazardous Situations Based on Kolmogorov-arnold Network
Year: 2024
Type: article
Abstract: Scanning electron microscopy (SEM) is one of the most common methods for analyzing the characteristicsof materials obtained through chemical synthesis. The use of this method makes it possible toobtain images with high resolution and magnification. The article examines algorithms for image analysisof materials with specific properties, such as porosity – bioneorganic matrices. Scaffolds are a broadclass of materials with a wide range of applications, including agriculture, medicine, catalysis, and manyothers. One of the important applications of such structures is tissue engineering, where such frameworksare necessary to ensure the regenerative processes of body tissues. And for each organism matrices mustbe personalized, which requires a laborious process of selecting the characteristics of the framework applicablein a particular case. This task is currently partially solved by the application of artificial intelligencetechnologies to improve accuracy or support decision making during matrix fabrication or analysis.However, some of the work in this process is still manual and represents a labor-intensive chore for thetechnician. In particular, the process of analyzing SEM images and characterizing the resulting materialstill involves many time-consuming steps using various tools. At the same time, such characteristics asporosity, tortuosity, and diffusivity are very important factors for an expert in the process of making adecision on the applicability of the fabricated bioinorganic matrix in each specific case. Accordingly, thepurpose of this research is to develop a set of algorithms for processing SEM-images. Also based on theset goal within the framework of the research we can distinguish a number of issues: development of algorithmsfor detection of objects in the image, development of a neural network model for refining the detectionresults, implementation of algorithms for calculating the characteristics of porous material, as well asdesign and execution of a number of verification tests to confirm the quality of the performed calculations.As a result of our research, we drew some conclusions. In particular, we found that an approach usingsynthetic data generation significantly speeds up and simplifies the learning process of neural networks,as well as improves the quality of output models. We also found that the algorithms we developed can fullyautomate the analysis of SEM images with porous structures, and their quality was confirmed through anumber of verification tests. These algorithms can be applied to other similar problems related to imageanalysis and identification of features and characteristics.
Authors: Sanni Singh, A.V. Pribylskiy
Institution: Southern Federal University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4406764548