Airborne ultrasonic images are difficult to obtain, because of the high absorption and attenuation by air. The airborne ultrasonic images are generated by a biomimetic sonar head. For each ultrasonic frequency, a pair of two images is generated, from the left and right sides. A single image has poor information content when ordinary ultrasonic transducers are used. The objective of the paper is to use the ultrasonic images at various frequencies in a single paradigm of image fusion, to obtain an image closer to the explored objects. In order to meet real-time requirements and to cope with small computation resources, imposed by the navigation tasks based on ultrasonic images with mobile-grounded vehicles, the fusion rules should follow the fastest path, e.g. the fusion rules based on the weighted average of available images. The results obtained by computer simulation show that the optimum set of weights, from the imposed cost criterion based on similarity and dissimilarity of images, has a distribution which is inversely proportional to the ultrasonic frequency. The obtained fused images are more appropriate for detection and classification of the explored targets in sonar based applications.