ABSTRACT(#br)Ground‐penetrating radar is one of the most effective methods of detecting shallow buried objects. Ground‐penetrating radar radargram is a vertical map of the radar pulse reflections that returns from subsurface objects, and in the case of cylindrical objects, it would be a hyperbola. In order to get clear and accurate information on the presence, location, and geometry of the buried objects, the radargrams need to be interpreted. Interpretation of the results is a time‐consuming task and needs an expert with vast knowledge. Development of an automatic interpretation method of B‐scan ground‐penetrating radar images would be an effective and efficient solution to this problem. A novel automatic interpretation method of ground‐penetrating radar images, based on simultaneous... perturbation artificial bee colony algorithm using tournament selection strategy, simultaneous perturbation stochastic approximation method, and new search equations, is introduced in this paper. The proposed algorithm is used to extract geometrical parameters, i.e. depth, location, and radius, of buried cylindrical objects in order to assess its accuracy. Synthetic data, simulated using GprMax2D forward modelling program, and real data, surveyed in the campus of Isfahan University of Technology, are used in the assessment. The performance of the proposed method in detecting synthetic hyperbolas is compared with that of the original artificial bee colony algorithm, genetic algorithm, and modified Hough transform. The results show superiority of the proposed algorithm, in detecting synthetic hyperbolas. Furthermore, the performance of the proposed method in estimating depth and radius of pipes in real ground‐penetrating radar images is compared with that of the modified Hough transform. The results indicate higher accuracy of the proposed method in estimating geometrical parameters of the buried cylindrical objects.