IEEE Consumer Electronics Magazine - March/April 2020 - 55

Figure 3. Red dots are UAVs, green dots denote
EU requests, and blue lines are connections
between UAVs. Such an array can self-adapt to
various EU requests.

morphing rule of SOMs. In the visual map space,
all the neurons are replaced by UAVs. The
perceptron layer of the SOM becomes a
telecommunication layer that is responsible for
broadcasting any EU request to all the UAVs.
Whichever UAV receives an EU request, the
request is transmitted to all the other UAVs
through the telecommunication layer. At the initial deployment stage, the topology of the entire
array is like a grid structure in a uniform formation. The topology is the same as the top layer
shown in Figure 2. When more EU requests are
generated and when the density of requests
becomes higher, proximal UAVs at the location of
interest gather together. For example, the bottom-right part of Figure 3 indicates a highly populated area. Densely generated requests are
observed in contrast to the other areas. With the
autonomous deployment system powered by
SOMs, the UAV array can adapt itself to variable
environments. More importantly, when EUs relocate, the UAV array can reflect such changes by
modifying its topology. Such a system has several
merits for a UAV array. First, a dispatch center
does not have to be fully aware of environments
and to decide the initial formation of a UAV array.
The array can learn its best topology based on
environments. Second, SOMs provide a framework for collaborative computation or data
transmission. In complex environments, various
surveillance missions are assigned to the same
UAVs. A drone undertakes heterogeneous sensing
tasks simultaneously.13 To timely accomplish a

March/April 2020

mission, e.g., capturing aerial photographs, performing data compression, and communicating
with clouds, it is appropriate to implement collaborative modes in the UAV array. SOMs can provide solutions to allocation problems or site
selection problems when collaborative modes
are enabled. Third, if a UAV carries limited
resources, e.g., computational capability and
communication channels, these constrained
resources can be utilized in a balanced way by
using collaborative modes. To handle a huge
amount of data, big-data cloud centers are usually
required for collaborative modes.14 Finally, the
size of an array can be dynamically adjusted
when SOMs evolve into growing SOMs (GSOMs)
or growing hierarchical SOMs (GHSOMs).15
GSOMs allow a UAV array to expand when more
drones are required for surveillance missions.
New drones can join the existing structure, and
the topology will accordingly change. GHSOMs
advance GSOMs by providing hierarchical topologies for creating subarrays, and hierarchy is useful
for heterogeneous networks.

CONCLUSION
This article discusses how the topology of a
UAV array can be self-organized to fit various
emergency situations. To enable self-organization for UAV arrays, swarm intelligence based on
SOMs is applied. SOMs allow autonomous UAVs
to dynamically change the formation of the array
when different EU requests or geospatial surveillance missions are assigned. An operations center simply dispatches a number of autonomous
UAVs to a target zone without bothering with
array formation during deployment. The size of
the UAV array can even be expanded when more
EU requests or missions are required. The topology of the UAV array is fluid, and it can respond
to the change of terrains or relocation of EU
requests. For complex remote-sensing situations, where heterogeneous sensor networks are
blended with communications networks, the
UAV array is capable of deriving a hierarchical
structure. Such a structured array is composed
of logical layers, where each one contains variable numbers of SOMs. Moreover, UAVs can be
reconfigured into either terminals or hubs, so
that hubs can coordinate parts of the array.

IEEE Consumer Electronics Magazine - March/April 2020

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - March/April 2020