迁移学习在舰船识别中的应用

Transfer learning in ship recognition

目前舰船识别技术相对落后,大部分识别算法还是基于传统的机器学习理论。舰船识别受舰船背景、光照、遮挡等因素的影响较大,识别正确率较低,无法满足现实需求;随着深度卷积神经网络识别率正确率的不断提高,一些复杂的分类任务都得到了较好的解决。本文将深度卷积神经网络AlexNet迁移到舰船识别中,对原网络顶层进行改进,微调底层特征,采用数据扩充技术构建的舰船数据集训练调试模型,得到了有效的舰船识别模型,识别正确率达到91.08%。

The technology of ship recognition has relatively dropped behind. Most of the recognition algorithms are based on traditional machine learning theory. However, due to the influence of environment, illumination and occlusion, accuracy of recognition ships is low and can＇t meet the actual needs. With the improvement of the accuracy of recognition rate by convolutional neural networks in depth learning, some complicated tasks of recognition have been solved well. In this paper, the convolutional neural network is transferred to the task of recognition ship. The top layer of the original model is improved and the characteristics of the bottom layers are finely tuned. Besides, the model trained with a limited data set of ships can obtain a recognition accuracy of 91.08%.