A Novel CCA-NMF Whitening Method for Practical Machine Learning Based Underwater Direction of Arrival Estimation

Underwater direction of arrival(DOA)estimation has always been a very challenging theoretical and practical problem.Due to the serious non-stationary,non-linear,and non-Gaussian characteristics,machine learning based DOA estimation methods trained on simulated Gaussian noised array data cannot be directly applied to actual underwater DOA estimation tasks.In order to deal with this problem,environmental data with no target echoes can be employed to analyze the non-Gaussian components.Then,the obtained information about non-Gaussian components can be used to whiten the array data.Based on these considerations,a novel practical sonar array whitening method was proposed.Specifically,based on a weak assumption that the non-Gaussian components in adjacent patches with and without target echoes are almost the same,canonical cor-relation analysis(CCA)and non-negative matrix factorization(NMF)techniques are employed for whitening the array data.With the whitened array data,machine learning based DOA estimation models trained on simulated Gaussian noised datasets can be used to perform underwater DOA estimation tasks.Experimental results illustrated that,using actual underwater datasets for testing with known machine learning based DOA estimation models,accurate and robust DOA estimation performance can be achieved by using the proposed whitening method in different underwater con-ditions.

Design Research on Drive System of Underwater Tapping Machine

The underwater tapping machine is composed of a center bit, a tapping cutter, a seal box, a main drive box, a boring bar assembly, a envelop, a gear case, a counter and so on. The drive system in underwater tapping machine consists of a worm drive, a gear drive system and a screw drive. The worm drive is in the main drive box. The worm is connected with a hydraulic motor and driven by the hydraulic motor. The gear drive system is a combined gear train which is the combinations of the fixed axes and differential gear train in the gear case. On the one hand, by means of the fixed axes gear trains the turn and power of transmission shaft are transferred to the boring bar and the screw rod, causing differential turn between the boring bar and the screw rod. On the other hand, the turns of the boring bar and the screw rod are transferred to the differential gear train. The differential gear train is used to drive a special counter to count axial travel of the boring bar. The screw drive is composed of a feed screw and a nut on the boring bar. There is the differential turn between the boring bar and the feed screw. By means of the nut, the boring bar can feed automatically. With the movement of the sliding gear 7 in the gear case, the designed drive system can also be provided with the ability of fast forward and fast backward movement of the boring bar in its idle motion, resulting in the increase of the tapping efficiency.

Modeling of underwater wet welding seam forming based on support vector machines

Underwater welding is developing fast because of the exploration of marine resources, and underwater wet welding automation is urgently needed because of the rigorous environment. To control the welding process automatically, the model of the process should first be built to predict the current welding process status. In this paper, arc and visual sensors were used simultaneously to obtain the electrical and visual information of underwater wet welding, and support vector machines （SVM） were used to model the process, experiment results showed that the method could effectively use the information obtained and give precise prediction results.

ANALYSIS AND SIMULATION ON THE MECHANISM OF A NOVEL DUAL-WAVE SHOCK TEST MACHINE

For qualifying the anti-shock performance of shipboard equipments and simulating actual underwater explosion environments, a novel dual-wave shock test machine is proposed to increase testing capability of shock test machines as well as to meet certain shock testing specification. The machine can generate a double-pulse acceleration shock for test articles according to specification defined in BV043/85. On the basis of the impact theory, a nonlinear dynamic model of the hydraulically-actuated test machine is established with thorough analysis on its mechanism which involves conversion of gas potential energy and dissipation of kinetic energy. Simulation results have demonstrated that the machine can produce a double-pulse acceleration shock in the time domain or a desired shock response spectrum in the frequency domain, which sets a theoretical base for the construction of the proposed machine.

Classification of underwater still objects based on multi-field features and SVM

A Support Vector Machine is used as a classifier to the automatic detection and recognition of underwater still objects. Discrimination between the objects can be transferred into different projection spaces by the process of multi-field feature extraction. The multi-field feature vector includes time-domain, spectral, time-frequency distribution and bi-spectral features. Underwater target recognition can be considered as a problem of small sample recognition. SVM algorithm is appropriate to this kind of problems because of its outstanding generalizability. The SVM is contrasted with a Gaussian classifier and a k-nearest classifier in some experiments using real data of lake or sea trial. The experimental results indicate that SVM is better than the others two.

Application of RVM for prediction of bead shape in underwater rotating arc welding

Bead sttape in underwater rotating arc welding was affected by several welding parameters. RVM （ relevance vector machine） was used to build a model to predict weld bead shape. The training data set of RVM eortsists of the welding parameters which are rotational frequency, rotational radius, height of torch and welding current and the features of the bead shape. The maximum error and mean error for prediction of width are 0. 10 mm and 0. 09 mm, respectively, and the maximum error and mean error for prediction of penetration are 0. 31 mm and 0. 12mm, respectively, which are showed that the prediction model can achieve higher prediction precision at reasonably small size of training data set.

水下人-机协同搜救作业人因可靠性分析研究

为分析水下人-机协同搜救作业人因可靠性并识别其关键影响因素,根据水下搜救训练实际和相关研究结论提取人因可靠性影响因素;建立人因可靠性影响因素体系,利用解释结构模型定性分析各影响因素间相关关系,将建立的各影响因素的解释结构模型转换为贝叶斯网络拓扑结构,分别构建潜水员作业、无人遥控潜水器(ROV)作业、人-机协同作业人因可靠性分析贝叶斯网络;通过对3类贝叶斯网络进行推理,揭示影响人因可靠性的关键因素及其依赖关系,并定量分析3种作业模式下的人因可靠性变化。结果表明:人-机协同搜救作业中“误解意图”“通信不畅”“信息不对称”“作业精度不足”“丧失救助能力”等影响因素起到关键作用;当上述因素得到较好控制时,人-机协同搜救作业的人因可靠性将会表现出优于单纯的ROV或潜水员独立施救的效能。研究结果可为水下搜救作业人因可靠性分析与提升提供参考。

相位感知的水下平台背景宽带噪声干扰抑制

针对水下航行器等平台存在宽带背景噪声干扰导致目标检测能力下降问题,提出一种适用于多通道水听器阵列的深度学习宽带背景噪声干扰抑制方法。该方法通过含干扰信号的多通道频域特征级联保留了相位信息,使用深度复数神经网络建立了一个估计纯净目标信号频谱特征的学习模型实现干扰抑制,再利用常规波束形成方法实现目标检测与跟踪。仿真结果表明,在信干比为-15 dB,-20 dB,-25 dB的双目标信号条件下,所提干扰抑制方法可以有效减少近场干扰影响,提高了常规波束形成的检测能力。湖试结果表明,该方法能够自适应地实现平台宽带背景噪声干扰抑制,有效提升目标检测和跟踪性能。

声光联合引导UUV回收系统光学子系统研究

当今水下无人航行器(UUV)在军事与民用领域都有着广泛的应用前景,布放与回收技术是影响UUV实际应用的一项关键技术。目前主要回收引导方式有声学导引、光学导引、电磁导引这3种,而各自都存在明显的优缺点。本文针对声光联合导引机械臂回收系统光学子系统进行研究,设计一组可在水下200 m、0.05~10 m范围工作的机器视觉镜头及其配合使用的靶标,并完成了系统的仿真。该系统可在环境要求下准确地提取出靶标特征点。

阴极等离子体营造水下射流电解加工环境研究

电解加工是水下原位加工的关键技术,常用于海底工作站、潜艇结构件的加工与修复,而射流电解加工是一种特殊的电解加工类型,具有加工装置便捷、工艺灵活等优点,但在加工中会引起电场分散,使加工定域性降低。为解决此问题,提出一种阴极等离子体营造水下射流电解加工环境的方法,通过管电极端面剧烈析出的气体及诱导产生的等离子体薄膜隔离电解液,从而实现电流密度在射流内呈高斯分布。结果表明,水下射流电解加工能获得与空气中基本相同的微凹坑加工结果。通过调控脉冲电压幅值能有效调节水下射流电解加工过程,提高气膜/等离子体膜的致密程度,提升管电极侧壁的绝缘效果,压缩工件表面的液膜厚度,从而提高加工定域性。

基于MobileNetV2-DeepLabv3+的混凝土坝水下裂缝语义分割模型

为解决深度学习算法难以有效检测混凝土坝水下裂缝的问题,构建了基于MobileNetV2-DeepLabv3+的混凝土坝水下裂缝语义分割模型。该模型引入轻量化网络MobileNetV2,同时将深层特征下采样倍数降为8,以提高小数据集工况下的识别准确率和推理速度;将交叉熵损失函数与Dice损失函数的组合作为模型的损失函数,以缓解类别不平衡问题。工程实例验证结果表明:该模型在测试集上的平均像素准确率和平均交并比分别高达90.87%和86.33%,满足水下裂缝语义分割精度要求;典型工况下的混凝土坝水下裂缝的分割效果优于其他对比模型,泛化能力强;模型具有内存占比小、推理速度快的特点,可用于混凝土坝水下裂缝的检测。

基于机器视觉的UUV水下沉底目标自主识别

水下无人航行器对目标的自动识别研究是未来水下目标探测领域的发展趋势。针对现阶段水下沉底目标声呐图像探测人工识别工作量大、判别困难等问题,研究了机器视觉中基于Canny算子图像边缘检测结合霍夫线变换去噪的识别方法在水下无人航行器自主探测的应用,为此在湖底布放多个水下沉底模拟目标,尝试了水下无人航行器的目标自主探测方案,通过实际湖上试验对比分析三种水下沉底目标声呐图像的算法识别效果,验证了论文所研究方法的有效性,为后续无人平台自主识别算法研究提供参考,以期提高无人航行器对水下沉底目标的自主探测识别能力。

基于对抗注意力机制的水下遮挡目标检测算法

水下环境复杂,遮挡目标信息缺失严重而难以提取到足够的特征信息,导致水下遮挡目标易被漏检。为解决该问题,提出一种基于对抗注意力机制的水下目标检测算法。以Faster R-CNN算法为框架,提出基于空间注意力机制的对抗生成遮挡样本网络(AOGN)。AOGN与Faster R-CNN网络相互竞争,通过三阶段训练过程,在不增加推理负担的情况下学习生成检测网络难以正确区分的样本,提高Faster R-CNN网络对水下遮挡目标的检测精度。使用Focal loss增加困难样本的损失比重,解决水下数据集难易样本不平衡的问题。在此基础上,为获得更丰富的水下目标特征信息,使用SE-ResNet50代替VGG16作为骨干网络,通过残差网络和SE模块的结合获得更有效、更丰富的水下目标信息,提高对检测目标的特征提取能力,同时加入多条ROIpooling支路实现多尺度特征融合,增加特征的丰富性。实验结果表明,该算法在URPC数据集和水下垃圾数据集上分别取得了73.76%和86.85%的平均精度均值(mAP),遮挡目标漏检率分别达到2%和7%,相较于其他检测算法能够有效提升检测性能。

ROV水下障碍物检测和避障技术的应用综述

全面回顾了远程操作车(ROV)在水下障碍物检测和避障技术方面的技术进展。研究集中于声呐系统、光学系统及其与机器学习和人工智能算法的结合,分析了这些技术如何提高水下作业的自主性、效率和安全性。尽管声纳和光学系统在环境适应性和障碍物检测精度方面已取得显著成果,但动态障碍物实时识别和复杂环境适应性的挑战仍待克服。此外,探讨了机器学习和人工智能技术在增强ROV自主避障能力方面的潜力和挑战,指出了这些技术在未来ROV操作中的重要性。该研究为深海探索和海洋科学提供了新的理论视角和应用实践。

海参捕捞机器人动态定位技术研究

为提高海参的水下定位精度,提出一种基于双目视觉系统的水下海参动态定位方法。首先,建立双目立体视觉模型并利用Bouguet算法对图像进行立体校正,从而使左相机与右相机的成像平面共面、铅垂方向高度一致;然后,构建优化函数并利用改进布谷鸟算法优化双目相机的内参,削弱了多介质折射对定位精度的影响;最后,对距相机防水玻璃(300~1000)mm任意位置的水下海参进行定位测试,结果表明,参数优化后的模型在对海参进行定位时,在x、y、z方向的平均误差分别为3.09mm、2.96mm、8.21mm,可为水下机器人捕捞海参提供准确的信号输入。

复杂附体结构仿生机器蟹水动力特性研究

借鉴变形机器人的变形技术,在传统仿生机器蟹设计上引入可变形水下滑翔机的概念设计,仿生机器蟹可根据不同工作任务,自主切换工作模式。基于粘性流体理论,将滑移网格法与多计算域法高效结合,对计及复杂附体结构的仿蟹机器蟹水动力性能进行研究,获得在不同航速、不同攻角下的升阻特性;对其压力场、速度场和涡量场的分布进行分析,得到仿生机器蟹在滑翔模式下的最优攻角,并为仿生机器蟹优化设计打下基础。

Numerical Modeling of Dual-Pulse Shock Test Machine for Simulating Underwater Explosion Shock Loads on Warship Equipments

In order to qualify shock resistance performance of shipboard equipments and simulate real underwater explosion environment,a novel dual-pulse shock test machine is proposed.The new machine will increase testing capability and meet special shock testing requirement.Two key parts of the machine,the velocity generator and the shock pulse regulator,play an important role in producing the positive acceleration pulse and the succeeding negative acceleration pulse,respectively.The generated dual-pulse shock for test articles is in conformity with an anti-shock test specification.Based on the impact theory,a nonlinear dynamic model of the hydraulically-actuated test machine is established with thorough analysis on its mechanism that involves conversion of gas potential energy and dissipation of kinetic energy.Simulation results have demonstrated that the proposed machine is able to produce a double-pulse acceleration shock in the time domain or a desired shock response spectrum in the frequency domain,which sets up a base for the construction of the machine.

利用跟踪轨迹特征和SVDD-SVM联合分类器的水下慢速小目标分类方法

针对蛙人、无人水下航行器(UUV)等慢速小目标分类识别所面临的小样本、类不平衡问题,提出了利用轨迹特征、支持向量数据描述(SVDD)与支持向量机(SVM)的联合分类方法。该方法将水下慢速小目标类型简化为蛙人、UUV、其他3类,利用跟踪轨迹特征设计多维特征量,构建SVDD-SVM联合分类器获得分类结果。具体为,针对小样本问题,采用参数维度小、训练数据量要求低的SVDD、SVM作为分类器的基本单元。针对类不平衡问题,使用2个并联的单分类SVDD和1个与两者串联的二分类SVM设计联合分类器,同时为联合分类器的输出设计投票机制保证分类结果的稳健性。实测数据处理结果表明,所提SVDD-SVM联合分类器对蛙人目标的平均召回率可达86%,平均精确率可达87%;对UUV目标的平均召回率可达85%,平均精确率可达86%。所提方法在小样本、类不平衡条件下具有优于传统方法的分类准确性和稳健性。

DRHA-UIE:基于双重残差混合注意力模块的水下图像增强方法

在水下环境中,悬浮的颗粒会对光造成散射和波长相关的衰减,使水下图像呈现出颜色失真、对比度低等问题.针对上述问题,本文提出一种基于双重残差混合注意力模块的水下图像增强方法(Dual Residual Hybrid Attention Underwater Image Enhancement method,DRHA-UIE).该方法采用改进的双重残差块完成特征学习,并通过在双重残差块中引入混合注意力模块,沿通道和空间两个维度对特征进行注意力权重推断,以捕获显著特征.本文设计结合像素级损失、结构相似性损失和内容感知损失的联合特征损失函数,以获得具有更精细纹理的增强图像.此外,本文应用水下图像形成模型(Image Formation Model,IFM)对水下图像基准数据集(Underwater Image Enhancement Benchmark dataset,UIEB)进行优化,获得了具有更高视觉质量的N-UIEB(New-UIEB)数据集,实验表明,相较于UIEB数据集,基于N-UIEB训练得到的增强图像具有更自然的颜色和更清晰的细节.为验证本文方法的有效性,将本文方法与主流的10种方法进行测试和比较,结果表明,DRHA-UIE方法有效提高了图像对比度并恢复了图像颜色,在与水下图像增强算法的定量比较中获得了最优的性能.

基于深平均流预测模型的水下滑翔机路径优化

随着水下滑翔机在海洋调查及声学探测领域的广泛运用,精准、高效控制其路径对精细化海洋观测至关重要。针对水下滑翔机受海流影响产生较大偏航差问题,采用最小二乘支持向量机法(LSSVM)预测深平均流,以单剖面偏航差最小为目标函数,以实际航向与计划航向夹角不超过一定值为约束条件,构建非线性约束极值模型,确定预设剖面最优目标航向及出水点坐标,从而实现路径优化目的。采用“海燕-II”型水下滑翔机历史数据进行验证,结果表明:1) LSSVM法预测深平均流准确性较高,但当局部流向有明显变化时预测效果不佳,取前3个剖面数据作为训练样本时预测效果更好;2)采用文中方法优化后,水下滑翔机路径更稳定,各剖面偏航差平均为281.1 m。