Measuring the absolute calibration constant is crucial for the radiometric calibration of synthetic aperture radar (SAR) systems. However, it is expensive to monitor the calibration constant continuously using manmade calibrators, and it is regionally restricted using the rainforest as the calibration field. In this study, the stability of SAR backscattering for common objects on the earth surface was analyzed, expecting to find the stable backscattering feature that could be used for maintaining absolute radiometric calibration. A database was established using Sentinel-1 dataset, and a classification model based on neural networks was proposed to extract the image slices of proper objects. Based on these, a temporal stable backscattering feature with a standard deviation of 0.19 dB was obtained from urban areas, and it was proved to be even more stable than the rainforest. Finally, the calibration scheme was given using this stable feature as a reference, which provided a new means of monitoring the SAR radiometric calibration constant. illuminated in the intervals of the observation tasks. Thus, the frequency of the constant measurement cannot be guaranteed.Besides the manmade calibrators and rainforests, the researches about calibration reference mainly focus on deserts, oceans, and permanent scatters (PSs). The study about the Simpson Desert shows that its backscattering coefficients are consistently 12 dB with a root mean square error (RMSE) of 0.2 dB as time changes and the accuracy when it is used to cross-calibrate radar altimeters is about 1 dB [12]. However, the desert's backscattering stability depends on the surface topography and soil moisture. Oceans can also be used to derive the calibration constant, using the empirical models of the relationship between the oceans' backscatter coefficient and the wind speed. The accuracy is about 0.5 dB when it is used to calibrate ERS-2 SAR images [13]. Nevertheless, this method needs massive images to fit the model and then to obtain model parameters; so its accuracy is susceptible to the quantity and selection of dataset. Moreover, these two methods cannot resolve the regional-restricted problem and the accuracy is relatively low. The method based on PS has also been studied. PSs usually appear in urban areas, rocky areas, and some P-land forests [14]; they can be detected in SAR images by the coherent or noncoherent methods in D' Aria, D et al. and Iannini, L et al. [14,15]. Since the RCSs rarely change with time, they can relatively calibrate multi-temporal images. If their RCSs are calibrated by corner reflectors or transponders, they can be used in absolute calibration. It has been proved that the stability of this method is better than 0.1 dB and the calibration difference between this method and the transponder method is less than 0.2 dB [1,16]. However, this method requires repeated-pass images with highly similar imaging geometry, such as the images that were used for differential interferometric applications [14].Given the above, the existing methods...