基于深度学习海上船舶图像识别与表征研究

Research on Image Recognition and Characterization of Maritime Ships Based on Deep Learning

提出了一种基于深度神经网络的目标检测算法,主要通过分析海上船艇拍摄的图像和视频,对其中的目标船只进行特征提取及表征检测。首先,利用海事领域公共数据集对深度学习模型进行训练,通过实验证明该模型能够有效检测海上出现的各类船只、漂浮物等,并可基于外观特征将它们分类为具体的类型,如渡轮、快艇和浮标等。其次将提出的深度学习模型与通用训练模型进行比较,结果表明提出的模型在海面物体的检测方面表现优于其他模型。最后比较了不同的深度神经网络结构,以获得最佳的检测性能,结果表明提出的模型具有约30帧每秒的实时检测速度。

This article proposes a target detection algorithm based on deep neural networks,the feature extraction and characterization detection of target ships are mainly carried out by analyzing images and videos taken by ships at sea.Firstly,the deep learning model is trained using the public dataset of maritime areas.Through experiments,it has been proven that the model can effectively detect various types of ships,floating objects,etc.that appear at sea,and can be classified into specific types based on appearance features,such as ferries,speedboats,and buoys.Secondly,the proposed deep learning model was compared with the general training model,and the results showed that the proposed model outperformed other models in detecting sea surface objects.Finally,different deep neural network structures were compared to achieve the best detection performance,and the results showed that the proposed model displayed a real-time detection speed of approximately 30 frames per second.