基于迁移集成学习的无人机图像识别算法

针对当前小型无人机目标图像识别方法准确率较低的问题，提出了一种基于迁移集成学习的无人机图像识别算法。首先，基于AlexNet、VGGNet-19、Inception-V3以及ResNet-50四种结构具有差异的卷积神经网络对源数据集进行预训练，获取图像的深层次特征；然后，对目标数据集进行迁移学习，得到目标的分类特征，构建分类模型；之后，采用相对多数投票法和加权平均法的集成学习方法，对分类模型进行集成得到迁移集成模型。构建了一个包含小型无人机图像、飞鸟图像以及直升机图像的图像数据集UavNet,在对数据集进行数据增强的基础上开展了图像识别算法性能实验，结果表明，算法对多类目标的识别准确率为99.42%,无人机类目标识别的F1-score指标为99.12%,优于主流的卷积神经网络方法和传统的支持向量机方法，具有一定的理论意义和应用价值。

A UAV Image Recognition Algorithm Based on Transfer Ensemble Learning

For the problem of low accuracy of current small unmanned aerial vehicle(UAV) target recognition methods,a UAV recognition algorithm based on transfer learning and ensemble learning is proposed. The algorithm first uses AlexNet,VGGNet-19,Inception-V3 and ResNet-50 convolutional neural networks(CNNs) for pre-training in the source dataset to obtain the deep features of the image. Then transfer learning is used to obtain classification features and generate four classification models on target dataset. A transfer ensemble model is finally established by adopting the relative majority voting method and the weighted average method. In addition,an image dataset called UavNet is built,which includes small UAV images,bird images and helicopter images. The image recognition algorithm experiment is carried out on the UavNet and the result shows that the accuracy of the new algorithm is 99.42%,and the F1-score of UAV target recognition is 99.12%,which is better than those of the mainstream CNNs method and the traditional support vector machines(SVM) methods.