现有基于稳定目标场的交叉定标方法受限于卫星同步过境人工选取定的伪不变目标且靶场光谱特性较为单一，难以进行多频次、高精度的交叉定标，而传感器的精确在轨校准对其数据产品的有效性至关重要。本文提出了一种基于迭代重加权多变量变化检测（IR-MAD）的交叉定标方法，以风云三号B星（FY-3B）光学成像仪的中分辨率成像光谱仪（MERSI）为参考基准，展开对同平台的可见光红外扫描辐射计（VIRR）的交叉定标研究。该方法通过IR-MAD算法自动识别两探测器同时相场景中的不变像元，基于不变像元的反射率信息，经过光谱匹配和正交回归获得交叉定标系数。本文以FY-3B/VIRR、MERSI中国西北地区和北非地区数据展开研究，获取长时间序列的定标斜率系数。结果表明，IR-MAD方法与业务定标、敦煌沙漠场地定标结果有很好的一致性，VIRR经定标校正后与MERSI的辐射信息相对偏差优于2%，中国西北地区和北非地区获取的IR-MAD交叉定标结果吻合（多数通道优于2%）。此方法可在卫星观测场景内自动高效的选取伪不变目标，并基于所选取的伪不变目标实现交叉定标。

Objective: During the operation of the satellite in orbit, the radiation detection performance and stability of the sensor will change, and the accuracy of the acquired radiation observation signal will decrease. Accurate on-orbit calibration of sensors is crucial to ensure the validity of their data products. The stable target site based cross calibration method is limited by the pseudo-invariant target which is manually selected by the satellite synchronous transit and has simple spectral characteristics. So it’s difficult to carry out multi-frequency and high-precision cross calibration. Method: This paper proposes a cross calibration method based on iterative reweighted multivariate alteration detection (IR-MAD). Taking the Fengyun-3B (FY-3B) Medium Resolution Spectral Imager (MERSI) as the reference, the cross-calibration of Visible and Infra-Red Radiometer (VIRR) on the same platform is studied. Firstly, the IR-MAD algorithm linearly combines the simultaneous multi-channel images of the two sensors to construct canonical correlation variables, and performs canonical correlation analysis to eliminate the correlation between different channels of a single sensor and unmatched channels of different sensors. Then, through multiple probability reweighting iterations, the IR-MAD algorithm can automatically identify the no-change pixels(NCPs) in the same phase scene of the two sensors. The radiance of the NCPs have the largest linear correlation between VIRR and MERSI matching channels. By correcting the spectral differences between matched channels of two sensors, the cross calibration coefficients can be obtained by a linear regression model based on the apparent reflectance information of the NCPs. This paper implements this procedure with data from Northwest China region and North Africa region, and compares it with the calibration results obtained from other related studies. Result: Experimental results show that the proposed method is well consistent with the operational calibration and the Dunhuang desert site calibration. The relative deviation of the radiometric information between VIRR and MERSI after calibration exceeds 2%. The cross calibration results of IR-MAD obtained in northwest China and north Africa are consistent (most channels are better than 2%)，which verifies the generalization and stability of this method. The long-time sequence results show that there is a correlation between the channel response decay trend and wavelength. On the whole, the shorter the wavelength, the more serious the decay. In addition, there are some seasonal fluctuations in the relative calibration slope time series of near-infrared and shortwave infrared channels. Conclusion: This method can automatically and efficiently select pseudo-invariant targets in satellite observation scenes, and realize high precision cross calibration based on the selected pseudo-invariant targets. It is applicable to the re-calibration of historical satellite data and the relative response monitoring of sensors on the same platform. The application of algorithms in different platforms and nonlinear problems are the direction of our future work.