The article deals with the problem of image segmentation. Most segmentation methods are not universal to work in different conditions. It is also difficult to find a suitable method to work in a specific environment. In this research, a method is proposed for determining large particles of a mixture in an image. The method is distinguished by its versatility and the ability to adapt to specific conditions. The method is based on the use of neural networks and a post-processing algorithm for a binary mask of image boundaries. The approaches described in the research can be applied in various industries, for example, to determine the granulometric composition of rocks.