A distributed energy detection scheme that exploits the spatial-temporal correlation is proposed for heterogeneous cognitive radio (CR) sensor network. We first use the weighted centroid algorithm to localize the position of the PU. Then according to the localization result, selective measurement combining for weighted average consensus is adopted to fuse the energy detection outputs of only homogeneous sensor nodes. From the simulation results, the better receiver operating characteristics of the whole network can be obtained than the conventional cooperative sensing and unconstrained weighted average consensus algorithms. To accomplish a low-complexity implementation, folding and hardware sharing techniques are employed. To support wide signal dynamic range, block floating-point representation is used and thus the design has tiny implementation loss.