With the rapid advances in data transmission over various advanced underwater acoustic networks, it is important to develop the efficient protocol to obtain a high data-rate with QoS guarantee requirements for acoustic wireless communications. Because of the harsh underwater acoustic environments, such as time-varying network topology and channel conditions, limited bandwidth and constrained underwater signal propagation, we develop and implement the multi-hop adaptive RTTdriven transport layer flow and error control protocol (ARTFEC) for QoS guaranteed image transmission in underwater wireless networks. ARTFEC is based on the congestion window size control and the Q-learning optimal timeout selection with QoS provisioning. For the congestion window size control, we propose an RTT-based data flow control policy to adapt to the varying acoustic channel environments, aiming to guarantee the reliability and high data rate of data transmission. In addition, we develop a Q-learning based optimal timeout selection algorithm to improve the channel utilization efficiency, which can increase the endto-end throughput while decreasing the packet loss rate. We implement ARTFEC using our lab testbed, which consists of the Aqua-Net protocol stack and the acoustic OFDM modems. The simulation results obtained show that our developed ARTFEC transport layer protocol outperforms the other existing reliable data transmission schemes, in terms of end-to-end throughput and packet loss rate.Index Terms-Underwater wireless networks, multi-hop, orthogonal frequency division multiplexing (OFDM), round trip time (RTT), congestion window size control, Q-learning timeout selection, quality of service (QoS), packet loss rate, end-to-end throughput.