The mass trend that has developed in recent years to use deep machine learning, artificial neural networks and artificial intelligence methods in solving a variety of tasks, both scientific research and production problems, leads to significant methodological inaccuracies, namely, the use of complex, time-consuming and very voluminous algorithmic procedures while the problem can be solved by incomparably uncomplicated and more compact methods without loss in efficiency. To do this, it is necessary to use widely used and well-known methods, including classical, parametric, nonparametric, customized to the task. Machine learning is certainly universal in nature, which is its value, but it does not take into account peculiarities of a specific problem, replacing with learning the stage that traditional methods are designed to fill with skillful adjustment to a specific problem. The paper considers two methods, one of which is based on the classical maximum likelihood method with optimal estimation of unknown parameters of the distributions of individual samples of the fluctuating image (input data array), and the other on deep machine learning, classification of radar images of a specific task of recognizing three classes of spatially distributed targets that differ in their sizes.