THE METHOD FOR APPROXIMATING THE EDGE DETECTION CONVOLUTIONAL OPERATOR USING A GENETIC ALGORITHM FOR SEGMENTATION OF COMPLEX-STRUCTURED IMAGES

The subject matter of the study in the article is the method for approximating the convolutional operator for edge detection using a genetic algorithm for segmentation of complex-structured images. The goal is to develop a method for approximating the convolutional operator for edge detection using a genetic algorithm for the segmentation of complex-structured images. The tasks are: analysis of known methods of segmentation of optoelectronic images, development of a method for approximating the edge detection convolutional operator using a genetic algorithm for segmenting complex-structured images, practical validation of the method for approximating the edge detection convolutional operator using a genetic algorithm for segmenting complex-structured images. The methods used are: digital image processing methods, data clustering techniques, matrix theory mathematics, swarm intelligence methods, the genetic algorithm, mathematical modelling techniques, optimization theory methods, as well as analytical and empirical methods for image comparison. The following results are obtained. The advantages and disadvantages of the main known methods for segmenting optoelectronic images have been identified. It has been established that the most effective segmentation methods for images from space-based optoelectronic observation systems (complex-structured images) are those based on swarm intelligence and genetic algorithms. An important case of segmentation – binarization (segmentation into two classes), has been considered. The task of binarization has been formalized, and the concepts of structural and amplitude predicates have been introduced. The method for segmenting complex-structured images has been improved, in which, unlike existing methods, a genetic algorithm is used for approximating the edge detection convolutional operator, facilitating segmentation of images at various scales with later integration of the results. A visual assessment of the quality of the segmented image has been conducted using the improved method. Conclusions. The method for segmenting complex-structured images has been improved, in which, unlike existing methods, a genetic algorithm is employed to approximate the edge detection convolutional operator, easing segmentation of images at various scales with later integration of the results. A visual assessment of the quality of the segmented image using the improved method shows a significant reduction in the number of noise objects present in the segmented image.