In this paper, we consider data capture of a cognitive radio-based red network in tactical wireless networks. In the blue (friendly) network, an eavesdropping sensor node is deployed to capture data of the red (adversary) network, where the red network adopts cognitive radio for opportunistic data transmission. Since most wireless networks suffer energy constraints, we investigate the data capture (eavesdropping) performance of the blue network in two scenarios of energy management: no energy harvesting and energy harvesting. In both of the energy scenarios, we formulate the problem of maximizing data-capturing performance of blue network in the framework of a partially observable Markov decision process (POMDP) by considering feedback observations of actions. Furthermore, a recursive method is adopted to solve the problem by which the optimal action policy for the eavesdropping sensor nodes of blue network is obtained. Finally, we compare the performances of the proposed scheme with those of different reference methods under the two mentioned energy scenarios. Numerical results show that the proposed POMDP-based scheme can improve the data-capturing performance of the eavesdropping sensor node.