共查询到20条相似文献,搜索用时 859 毫秒
1.
运用神经网络图像特征提取联合SSA-SVM分类算法,对通航区域图像中的典型船舶目标进行识别以实现船舶目标的自动分类。首先通过摄像机获得通航区域的高分辨率图像,以AlexNet深度学习网络为基础经迁移学习后提取典型船舶目标特征,获得4种船舶类型、共5 505 024个特征数的典型船舶目标特征矩阵。以特征矩阵为训练依据训练SSA-SVM算法,在种群寻优下获得最佳识别参数,经训练得出在小数据集下具有较强辨识能力的SSA-SVM船舶目标识别模型。实验表明,相比于深度学习的大数据集驱动识别算法,使用AlexNet特征提取的SSASVM算法能够在数据量较少的情况下对散货船、集装箱船等典型船舶目标进行有效识别,识别准确率为88.87%、训练时长为1 856 s,满足实用需求,为水上监管提供了可靠的技术支持。 相似文献
2.
船舶图像细粒度检测是高分辨遥感图像分析的难题,受船舶尺寸、陆地背景、光照、风浪等因素影响,易降低图像检测的准确性.为克服船舶目标识别的影响因素,针对不同类型和型号的船舶目标检测建起特征提取算法模型,提升最终的识别精度.本文提出一种基于深度学习的船舶图像细粒度检测方法,将深度学习算法应用到高分辨率遥感图像中,借助算法训练... 相似文献
3.
本文提出一种基于卷积神经网络的船舶红外图像边缘检测方法。首先,介绍船舶红外探测技术的基本原理,针对船舶红外图像的预处理进行研究,包括灰度的均衡化、红外图像的背景抑制、图像分割等。设计了一个基于卷积神经网络的红外图像边缘检测模型,该模型采用多层卷积和池化操作,以及非线性激活函数,能够有效地捕捉图像中的边缘信息。最后,通过对模型进行训练和优化,得到了准确度较高的船舶红外图像探测算法,为后续船舶的目标识别和跟踪提供了有效的基础。 相似文献
4.
对船舶图像进行快速准确识别在军民领域都有广泛应用,随着船舶种类的增多、图像质量的提高,传统的卷积神经网络进行船舶图像识别需耗费大量时间。本文对深度神经网络的原理进行分析,并在此基础上研究基于深度神经网络的船舶图像识别流程,对船舶图像预处理技术进行研究,建立船舶图像训练集和测试集,对YOLOV2、卷积神经网络和本文算法的平均识别时间和识别准确率进行分析,最后研究3种算法的训练次数对识别准确率的影响。本文研究的深度神经网络船舶图像识别算法,在平均识别时间以及识别准确率上具有一定优势。 相似文献
5.
6.
7.
针对当前监控视频中船舶识别成功率低、无法进行在线识别的难题,为了对监控视频中船舶进行准确识别,提出基于深度学习的监控视频中船舶识别方法。首先对监控视频中船舶识别原理进行分析,采集船舶识别的监控视频,将船舶识别从背景中分割,然后提取船舶识别的不变矩特征,将不变矩特征输入深度学习算法中进行训练,建立监控视频中船舶识别模型,最后进行了多个监控视频中船舶识别验证性实验。实验结果表明深度学习算法可以准确对监控视频中的船舶进行识别,提高了监控视频中船舶识别成功率,误识率急剧下降,远低于当前其它监控视频中船舶识别方法,实时性要也要高于其它识别方法,是一种速度快、结果可信的监控视频中船舶识别方法。 相似文献
8.
9.
10.
11.
基于船舶自动驾驶存在的问题以及需求,提出了一个基于卷积神经网络的船舶自动识别系统.系统设计使用的生成对抗神经网络算法,基于互信息理论,能够无监督式地学习船舶图像特征.通过实验论证,在分类准确度上取得了显著地提升,表明本系统方法合理有效,具有较高地运用前景. 相似文献
12.
13.
14.
15.
16.
17.
18.
A ship is operated under an extremely complex environment, and waves and winds are assumed to be the stochastic excitations. Moreover, the propeller, host and mechanical equipment can also induce the harmonic responses. In order to reduce structural vibration, it is important to obtain the modal parameters information of a ship. However, the traditional modal parameter identification methods are not suitable since the excitation information is difficult to obtain. Natural excitation technique-eigensystem realization algorithm (NExT-ERA) is an operational modal identification method which abstracts modal parameters only from the response signals, and it is based on the assumption that the input to the structure is pure white noise. Hence, it is necessary to study the influence of harmonic excitations while applying the NExT-ERA method to a ship structure. The results of this research paper indicate the practical experiences under ambient excitation, ship model experiments were successfully done in the modal parameters identification only when the harmonic frequencies were not too close to the modal frequencies. 相似文献
19.
Institutes imparting training in marine engineering require replication of shipboard ambience for strengthening the competencies. For building an engineering centre for training, five options at different physical levels were considered based on a model of a new liquefied natural gas tanker ship build. A mock-up facility, simulator, full-scale engine room, scaled down version and a combination arrangement with live and dummy equipment were the options. Analytic hierarchy process was applied for selecting a suitable option based on criteria of cost, effectiveness in attaining competencies, practicability and environmental conduciveness. Analyses were carried out on the eigenvalues based on eight subselection criteria. The combination of live equipment (boilers and turbo alternator) and non-live shipboard equipment (auxiliaries) was chosen based on the global weightages obtained from the pairwise comparison matrix computations. The reliability was ascertained from the consistency index which was less than 0.1. For selection of the learning modes and the equipment, a constructivist approach of learners reflecting and choosing the learning mode was adopted. Industry practitioners were made into learner groups composed of trainers, shipboard personnel and company personnel. Established competencies were assigned as scores to the different learning modes. ANOVA application and statistical methods were used to analyse the scores to verify if there were too much variation in the choices. The calculated F ratio values were low (0.14 to 0.40) compared to the reference values indicating that the choices were even. The chi-squared test indicated that the group composition did have an influence on choosing the learning mode and equipment for training. The objective of identifying the learning mode and equipment for the training centre was achieved. 相似文献
20.
在建立船舶能耗评价模型基础上,针对能耗影响因素众多,提出了基于知识融合的船舶能耗决策支持系统的研究。以船舶能耗为对象,应用系统辨识理论,建立重点设备能耗模型库;以方法、经验和案例为基础,建立船舶能耗知识库;融合模型库、知识库、关系数据库等多源知识,建立船舶能耗评价与决策支持系统。以实现船舶航运管理、船舶操纵控制、航线设计以及船舶设计等各个过程的决策支持。 相似文献