The integration of unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs) in ground object search tasks utilizes the complementary strengths of these heterogeneous agents, thereby enhancing search efficiency. This approach holds significant potential applications in disaster relief and object rescue. However, existing research on UAV-UGV collaboration typically uses UGVs to play roles as mobile charging stations or ground executors, lacking an effective information exchange mechanism between UAVs and UGVs. Reinforcement learning, widely employed in environments where agents interact to maximize rewards or achieve specific goals, has demonstrated remarkable performance in solving object rescue and search problems. This paper introduces a novel bidirectional feedback Actor-Critic (BiF-AC) algorithm, which utilizes probabilistic map updates to facilitate bidirectional information feedback between UAVs and UGVs. The algorithm incorporates three evaluation metrics for assessing UAV search outcomes, which form the basis for UAV reward calculations. This method promotes deep collaboration between UAVs and UGVs, significantly improving the efficiency and accuracy of ground object searches. A comparative analysis with random search algorithms, the A* algorithm, genetic algorithms, deep Q-networks, double deep Q-networks, and the traditional Actor-Critic algorithm demonstrated that BiF-AC outperforms state-of-the-art algorithms in both search success rate and search cost.