Current tasks for today are to search and improve the methods of automatic detailed decoding of objects on aerial photographs obtained from unmanned aviation complexes, which would provide sufficient accuracy of detection and recognition of fine-grained objects in the complex topographical conditions of the terrain above which aerial images are obtained. In order to solve this problem in the article an analysis of methods for automatic image processing and models of neural networks built on their basis. From the analysis, a multi-stage conveyor for aerial photographs has been selected, combining detection approaches, elemental segmentation and semantic segmentation for contextualization. Improved models of cascade of segmentation take into account geometric sizes of objects and their correlation, change in scale, conditions of removal. Using the segmentation cascade model for automatic decoding of objects on aerial photos will increase the accuracy of detection and recognition of such objects.