面向水面无人艇的船舶舷号识别方法

Detection and identification of ship's hull number for unmanned surface vehicle

[目的]针对水面船舶舷号检测问题,提出一种面向水面无人艇的实时船舶舷号检测方法。[方法]基于原始的单阶段目标检测模型(YOLO),引入注意力机制,利用空间信息交互模块和分割注意力融合方法,提升神经网络对重要目标区域的敏感度。考虑先验知识对模型精度的影响,结合自适应锚框算法和正样本增强策略提高回归精度。针对深度神经网络(DNN)收敛困难的问题,改进损失函数,在保证网络收敛速度的同时提高神经网络训练的稳定性。最后,将改进的目标检查模型部署在无人艇上进行有效性验证。[结果]结果表明,所提算法在3级海情下能够准确识别船舶目标及其标志舷号,相比于原模型,改进后的YOLO算法在全类平均精度(mAP)方面提高了14%,识别速度满足实时要求。[结论]研究证明了所提舷号检测方法满足无人艇实时识别舷号任务的要求,并在复杂海洋环境中仍然具备识别能力。

[Objective]Aiming at the problem of ship hull number recognition,this paper proposes a real-time ship's hull number recognition method for unmanned surface vehicles(USVs).[Methods]Based on a one-stage object detection model(e.g.YOLO),the attention mechanism is introduced to make the network more sensitive to the target area by the spatial information interaction module and divided attention method.Considering the effect of prior knowledge on accuracy,the adaptive anchor method and positive sample assignment strategy are utilized to improve the accuracy of regression.Aiming to resolve the problem of slow convergence at the beginning,the loss function is redesigned to speed up the convergence and enhance the stability of the network in the training phase.Finally,the proposed method is deployed in a USV to validate the availability of the recognition performance.[Results]The results shows that the proposed method can achieve the recognition of ships and hull numbers simultaneously under Sea State 3 conditions,and has a 14%improvement in mean average precision(mAP)compared with the original model,with the ability to perform recognition in real time.[Conclusion]The results of this study indicate that the proposed method can be applied to USVs to perform hull number recognition,even under complex ocean conditions.