基于视觉的船舶智能识别技术应用

Application of Vision-based Ship Intelligent Recognition Technology

目的:为实现视频图像船舶目标智能识别的工程应用,需要解决样本分布不均衡、分类体系差异大和训练成本高的问题,并提高技术方案的复用性。方法:提出了卷积神经网络结合支持向量机(SVM)的技术。首先基于卷积神经网络训练了具有多尺度、可泛化特点的船舶目标检测器,实现对视频图像中的船舶目标进行检测,然后将可解释的船舶形状、颜色、纹理特征组成特征向量,用SVM算法进行分类。结果:最后通过多组实验对识别效果进行了验证。结论:实验结果表明,本文提出的技术方案切实可行,经济成本低,型号识别率达到90%以上。

Objectives:In order to achieve the engineering application of intelligent recognition of ship targets in video images,it is necessary to address the issues of imbalanced sample distribution,significant classification system differences,and high training costs.Additionally,it is important to enhance the reusability of the technical solution.Methods:A technical solution of combining convolutional neural networks with support vector machines(SVM)is proposed.Firstly,a ship target detector with multi-scale and generalization characteristics is trained based on convolutional neural networks to achieve detection of ship targets in video images.Secondly,interpretable features such as ship shape,color,and texture are combined to form a feature vector,which is used for classification with the SVM algorithm.Results:The recognition effect was finally validated through multiple sets of experiments.Conclusions:The experimental results indicate that the proposed technical solution in this paper is practical and feasible,with low economic costs.The model recognition rate has achieved over 90%.