We adapt matched‐filter searching to a region of sparse stations and seismicity, and demonstrate that earthquake detection can be significantly enhanced. The Earthscope Transportable Array (TA) increased the density of broadband seismometers in New England from 2013 until 2015, allowing for a higher resolution characterization of the regional seismicity. During this time, there were two transient increases in seismicity rates in Connecticut: one in August 2014 near the town of Deep River and one shortly after in January 2015 near Plainfield. Using the TA stations along with regional permanent stations, we implement a network‐based matched‐filter search to find events that have escaped previous detection during these bursts of seismicity. Applying a matched‐filter approach in a region of sparse seismicity with a distributed regional network presents two challenges. First, a standard matched‐filter search imposes a fixed template event duration across the seismic network; this constraint is not appropriate for the waveforms of shallow crustal seismicity that vary significantly with epicentral distance. We adapt the matched‐filter algorithm to produce a network correlation coefficient sum that is properly weighted by the duration of the seismic wavetrain that varies across the network. Second, the low seismicity means there are few events to use as templates. We develop a recursive method to expand the catalog using newly identified earthquakes as template events in the following iteration of the matched‐filter search. With these improvements in place, we are able to implement a matched‐filter search that identifies three times more events than the regional catalog. Thanks to our denser catalog and precise relative locations, we observe an aftershock sequence following an 2.6 mainshock near Deep River and a swarmlike sequence without any clear mainshock in Plainfield.