This paper proposes a cooperative online target detection methodology by multiple autonomous underwater vehicles (Multi-AUV) equipped with the side-scan sonar (SSS) sensor for real-time, accurate, and efficient underwater target detection and positioning in unknown environments. Due to the existence of unfavorable factors such as severe noises and geometric deformation of SSS images, this study incorporates the prior-based threshold segmentation with multi-scale cascaded networks (MSCNet) to reduce the high false alarm rate significantly. Specifically, to the real-time requirements of the AUVs computational platform, this study proposes the sequentially dual-branch lightweight block (LWBlock) as a baseline to obtain dense feature maps, which provide a good trade-off between accuracy and speed. Meanwhile, this study establishes the comprehensive correction model, which obtains the accurate target positioning information fusing with the predicted results. Furthermore, according to the target information provided by the automatic target recognition (ATR) system, the data-driven behavior-based (DDBB) path re-planning algorithm is performed that endows each AUV to scan above the interest target autonomously and in detail by designed maneuver behavior. Simulation and actual sea trial experimental results show that the proposed method outperforms other state-of-the-art algorithms, and achieves the recognition accuracy of 92.16%, inference speed of 2.45 s, and obtained the best FPR indicator in three SSS targets of 2.54% (metal ball),1.96% (seabed rock) and 1.03% (metal rod), respectively. At the same time, the proposed algorithm can improve detection efficiency by at least 40% compared with a single AUV, which can be widely used in marine mission exploration and resource deployment.