基于多视图卷积神经网络的船体分段性能研究

Research on Hull Segmentation Performance Based on Multi-view Convolutional Neural Network

在船舶调度过程中,用准确的船体分段识别号来识别船体分段的位置非常重要。为解决由于某些船体分段的位置和识别号的错误信息导致确切的船体分段所在位置查找困难的问题,需要配备系统来跟踪分段的位置,并自动识别分段的识别号。本文比较了5种卷积神经网络(CNN)模型与船体分段分类上的多视图图像集的性能,采用四个分段模型对船体分段进行图像采集并利用原始训练数据和其增强数据对CNN模型进行了迁移学习。

In the process of ship scheduling,it is highly important to use accurate hull segment identification numbers to identify the location of the hull segments.In order to solve the problem of difficulty in finding the exact position of the hull section due to the incorrect information of the position and identification number,it is necessary to equip a system to track the position of the section and automatically identify the number of the section.This paper compares the performances of five convolutional neural network(CNN)models and multi-view image sets on the classification of hull segments.Four segment models are used to collect images of the hull segments and applies the original training data and its enhanced data to transfer learning the CNN model.