基于激光回波时频图纹理特征的飞机目标分类方法

Aircraft Target Classification Method Based on Texture Feature of Laser Echo Time-Frequency Image

为实现直升机、螺旋桨飞机和喷气式飞机的激光遥感探测分类,研究了基于时频图的飞机目标微动纹理特征提取算法。根据旋翼微多普勒模型仿真三类飞机旋转部件回波信号,将平滑伪魏格纳-维利变换得到的时频分布生成灰度图像。采用大津(OTSU)法结合灰度拉伸对图像进行阈值去噪处理,提取目标灰度共生矩阵(GLCM)特征以及Tamura特征,并针对时频图差异进行特征优化,最后使用支持向量机(SVM)实现飞机目标分类。仿真数据分类结果表明:GLCM特征对噪声表现敏感,经所提方法对时频图去噪,信噪比(SNR)RSN=0dB时的分类正确率可达96.4%。Tamura特征在高信噪比条件下分类正确率较高,但当RSN<5dB时下降明显。因此提取时频图纹理特征可以达到较为理想的飞机分类效果,且利用改进GLCM特征能够实现低信噪比条件下的目标准确分类。

To achieve laser remote sensing classification of helicopter, propeller and turbojet aircraft, a texture feature extraction algorithm of aircraft target based on time-frequency image is studied. Three types of aircraft rotating parts echo signal are simulated according to the rotor micro-Doppler model, and the grayscale image is generated by the time-frequency distribution obtained by smoothed pseudo Wigner-Ville transform. OTSU method combined with grayscale stretching is used to perform threshold de-noising on the image, and the gray-level co- occurrence matrix （GLCM） feature and the Tamura feature are extracted. Feature optimization is carried out for the time-frequency difference, and finally the support vector machine （SVM） is used to classify the aircraft targets. Simulation data classification results show that the GLCM feature is sensitive to noise performance. When the time- frequency image is denoised by the proposed method and the signal-to-noise ratio （SNR） RSN=0 dB, the classification correct rate reaches to 96.4%. The Tamura feature has a higher classification accuracy under high SNR conditions, but decreases significantly when RSN 〈5 dB. Therefore, good classification performance can be obtained with the extraction of the texture feature of time-frequency image; and accurate Classification of targets can be achieved by the improved GLCM feature under low SNR conditions.