高光谱亚像元图像仿真与异常检测应用

为满足高光谱异常检测研究所需的大量地物高光谱图像需求,提出利用待观测地物的高光谱特性仿真数据及背景特性数据生成高光谱图像的方法,开展了典型飞机流动与传热模型、红外辐射特性模型、高光谱图像仿真模型研究；以实验测定的飞机反射率为输入开展目标特性计算,结合实际观测的背景起伏图像,在特定遥感器光谱响应特性、遥感器相对定标误差、随机加性噪声等条件下,生成了不同像元丰度、不同信噪比的高光谱图像,并应用经典的RX算法、CEM算法检测了仿真图像的异常像元。研究结果表明:建立的模型可以根据遥感器的性能指标参数、目标丰度要求生成亚像元高光谱仿真图像。图像可以反映目标飞机像元丰度、信噪比对检测结果的影响,通过调节输入参数可以高效建立针对亚像元异常检测的高光谱仿真图像；应用仿真图像进行RX算法检测高光谱仿真图像时,噪声会对检测结果产生较大影响,当信噪比低至10 dB时,RX算法难以检测出丰度0.4以下的异常像元,采用光谱匹配检测的CEM算法可以在较低像元丰度和信噪比下检测异常,提高检测概率。

Hyperspectral subpixel image simulation and application to anomaly detection

To meet the large demand of hyperspectral images for hyperspectral anomaly detection, a hyperspectral image generation method was proposed based on the hyperspectral characteristic simulation data and background characteristic data of the target. The flow and heat transfer model, infrared radiation characteristic model, and hyperspectral image simulation model were studied. The aircraft reflectivity measured by experiment was taken as the input for the calculation of target characteristics. Combined with the actual observed background undulating image, hyperspectral images with different pixel abundances and signal-to-noise ratios were generated under the conditions such as specific spectral response characteristics of remote sensor, relative calibration error of remote sensor, and random additive noise. The abnormal pixels of the simulation image were detected by RX algorithm and CEM algorithm. Results show that the model could generate hyperspectral subpixel simulation images based on the performance parameters and target abundance requirements of the remote sensor. The image could reflect the impact of target aircraft pixel abundance and signal-to-noise ratio on the detection results. By adjusting the input parameters, the hyperspectral simulation image for subpixel anomaly detection could be efficiently constructed. When the RX algorithm was used to detect hyperspectral simulation image, the detection results were largely affected by the noise. When the signal-to-noise ratio was as low as 10 dB, it was difficult to detect the abnormal pixels by RX algorithm with the abundance less than 0.4. While the CEM algorithm based on spectral matching detection could detect anomalies and improve the detection probability under low pixel abundances and signal-to-noise ratios.