In this work a navigational concept for a heteroge- neous swarm of AUVs is treated. Inspired by search and rescue manoeuvres of emergency forces, the swarm consists of two types of agents that detect and inspect possible objects of interest like contaminations of munition, crash sites or pipeline leaks. Each agent type differs by its degrees of freedom, environmental sensors, navigational devices and software defined role. The main part of this work is the obligatory and optimized information exchange between the agent types to localize and inspect objects of interest. To evaluate the concept, a simulation environment for autonomous agents using the proposed algorithms for target localization is realized. Since this work is a general representation of a powerful concept, we use two common feature extraction methods for information encoding. First, we use Ri-HOG because it is deterministic in parametrization and results. On the other hand, we use artificial neural networks since they provide excellent results for image-related tasks such as segmentation, reconstruction, and object classification. Both methods are eval- uated for their ability to reference the same object in two images with short feature vectors. We then test the performance of the methods as well as the advantageous concept of heterogeneous swarm behavior in terms of mission time and power consumption using an underwater simulation that uses synthetic sonar image data.