Abstract: ‍ ‍In recent years， unmanned aerial vehicle （UAV） networks have been progressively and extensively employed in various industries， which places higher demands on the network capacity that drone networks can provide. Directional antennas combined with drone networks form directional drone networks to address the problem of low network capacity caused by the competition for limited communication resources among nodes in UAV networks. Directional UAV networks can improve slot utilization through the spatial reuse capability of directional antennas. In response to the problem of low slot utilization and limited network capacity within directional networks employing the TDMA protocol， this study proposes a protocol for slot reuse and power control in directional UAV networks based on deep Q-networks （DQNs）. To enhance slot utilization， we consider multiple links communicating in a single slot to achieve slot reuse. However， the simultaneous communication of multiple links in a single slot introduces inter-link interference. Managing power control to increase network capacity while considering this interference among links is a key focus in slot reuse research. To address this problem， we first consider imposing constraints based on power requirements and the minimum channel capacity for each link. Considering a more complex and practical link interference model compared with other studies， the problem is formulated as a maximum overall network capacity problem. Subsequently， for a more intricate inter-link interference environment， the instantaneous channel information and directional gain states of multiple links are incorporated into the state of the DQN framework. The reward for the DQN is defined as the sum of the channel capacities of links that exceed the minimum channel capacity. Finally， by extending the optimization problem of each time slot to that of each frame， multiple DQNs are utilized. Simulation results demonstrate that， while ensuring the minimum channel capacity of each allocated slot， the proposed method significantly increases network capacity compared with the benchmark methods.