Formation of an ensemble of models for recognizing objects in radiographic images based on information theory

The paper considers the main functional approaches to the construction of image processing schemes for various spectral ranges, energies and scales. Their effectiveness is estimated from the point of view of the information value determined by the degree of achieving the goal of the nondestructive testing system - detection of all possible objects of interest with a minimum number of images. It is established that in the absence of the possibility of optical (visual) control and the technical possibility of obtaining radiographic images at different emitter energies, it is possible to improve the quality of object recognition by using an ensemble of recognition models. This conclusion is confirmed by a specific example demonstrating an improvement in the quality of recognition of objects of interest by a two-level ensemble of Yolo8 models by 12-18% compared to their recognition using a single neural network model.