卷积神经网络在飞机类型识别中的应用

Application of Convolution Neural Network in Aircraft Type Recognition

为提高现代战争过程中对敌军飞机的识别能力,针对军用飞机样本量少、不同视角条件下形变明显的特点,提出一种融入center loss的卷积神经网络与ANN分类器结合的飞机类型识别方法。首先利用3Dmax软件制作的6 000张5类飞机图片对构建的多层CNN模型进行训练,并利用这些图片的CNN特征训练ANN分类器,然后用训练好的网络模型和分类器对真实飞机样本进行预测分类。实验结果表明:在样本量少且目标形状复杂的情况下,该方法对5类军事飞机的识别精度可达到97.17%,是一种切实可行的飞机类型识别算法。

In order to enhance recognition ability of enemy aircrafts in modern warfare process, with the characteristics of small amount of military aircraft samples and variety of shapes under different viewing conditions, the aircraft recognition algorithm combining center loss convolution neural network and ANN classifier is proposed. Firstly, five kinds of 6 000 aircraft images, which are produced by using 3 Dmax software, is used to train the designed multi-layer CNN network model and the CNN features of these images are used to train ANN classifier. Next, the trained network model and classifier are used to forecast and classify the real aircraft samples, respectively. Finally, experimental results show that the recognition accuracy of the proposed algorithm to the five kinds of military aircraft can reach 97.17% in the case of small amount of aircraft samples and complex target shapes, which can prove that the algorithm is feasible to recognition of aircrafts type.