利用深度森林进行船舶类型分类识别

Classification and Identification of Ship Types Based on Deep Forest

针对船舶轨迹特征单一、分类的船型有限和利用AIS数据对海上船舶类型分类识别的研究比较欠缺等问题,提出一种基于深度森林(Deep Forest)的船舶类型分类识别方法。首先,该方法提取船舶轨迹段的16个运动特征,提高船舶特征的多样性和辨识度;其次,利用随机森林的重采样技术和特征选取的随机性,构造随机森林作为个体学习器,增强类型分类的鲁棒性;最后,对4个森林集成并串联构成具有自适应迭代层数能力的深度森林,从深度上加强表征学习,提升泛化能力。实验表明,该方法适用于小规模数据集的船舶类型分类识别,对游艇和其他类型的船舶分类效果显著,能够有效提升货船和油轮的分类精度,分类时效性好,并且随着样本库的扩展和模型的优化存在提升空间。

Aiming at the problems of single trajectory features,limited ship types of classification and lack of research on classification of marine ship types based on AIS data,a classification and recognition method of ship types is proposed based on Deep Forest.Firstly,16 motion features of the trajectory subsections are extracted to improve the diversity and identification of ship features.Secondly,the random forest is constructed as an individual learner to enhance the robustness of types classification using the resampling and selecting features randomly techniques of random forest.Finally,the four forests are integrated and connected in series to form a deep forest with the ability of adaptive iterative layer number,so as to enhance representational learning and generalization ability from depth.The experimental results show that the method is suitable for the classification and identification of ship types in small-scale data sets,and has a significant effect on the classification of yachts and other types of ships,which can effectively improve the accuracy of cargo and oil tankers and also have a high efficiency of classification.Moreover,there is room for improvement with the expansion of sample base and the optimization of model.