基于YOLOv2的船舶目标检测分类算法

为克服传统船舶检测方法提取的特征在复杂多变的实际海域场景中泛化能力差而导致船舶检出率和识别率较低这一问题,提出一种基于YOLOv2和支持向量机(support vector machine,SVM)的船舶检测分类算法。基于YOLOv2网络检测船舶目标,通过卷积神经网络提取船舶区域的深度特征,特征全局池化后利用SVM分类器实现分类。实验结果表明,该算法在自建的船舶数据集上船舶检测的平均精确率达80.5%,船舶分类的准确率达90.87%,有效实现复杂海况下船舶目标的检测以及舰艇、货船、渔船的识别。

Detection and classification of ship target based on YOLOv2

To solve the problem that the features extracted using the traditional ship detection method are poor in generalization ability,which results in bad performance on ship detection and recognition in complex actual sea scenes,a ship detection-classification algorithm based on YOLOv2 and support vector machine(SVM)was proposed.The ship target was detected based on the YOLOv2 network.The deep features of detected ship regions were extracted from convolution layers.The deep features were fed to SVM classifier for classification after global average pooling.Experimental results show that the average precision of ship detection reaches 80.5%and ship classification accuracy reaches 90.87%on the self-built ship dataset,which demonstrate that the proposed method can effectively achieve the detection of ship targets and recognition of warships,cargo ships and fishing boats in complex sea conditions.