This work is devoted to the development of a method for detecting mines (anti-tank, anti-personnel, tripwires). The feasibility of using an ultra-wideband synthetic aperture radar as a measuring tool is substantiated. The radar is located on an unmanned aerial vehicle (multicopter). An inertial (IMU) and satellite navigation system is used as a navigation unit. The resulting image is processed via a two-stage segmentation and classification framework. Segmentation is done over slices of b-scans of the image, resulting in a 3D image processed by a 3D U-net to detect the presence of features of the mine. The classification step uses segmentation map produced on the previous step and combines it with the slices of b-scans in a multi-view framework to identify the type of the landmine. Semi-supervised MixMatch algorithm with several custom 3D augmentations is utilized during the classifier training to improve the accuracy of classification by utilizing labeled and unlabeled samples. In the segmentation stage, two classes are defined: object and background. The object is mines, some of which are installed on the surface of the earth, and some of which are buried in the soil. During the classification stage detected objects are further classified into anti-personnel and anti-tank mines. Among the mines used: anti-personnel mines – MON-50, MON-90, MON-100, MON-200; anti-tank – TM-62M, TM-62P. The background is sparse and depressed vegetation, under which there are no mines. The measurement results show a high potential for detecting mines with minimal metal content.