|
In the infrared detection of small targets, the blind pixels greatly interfere with the detection accuracy of the target. Therefore, the way how to correctly detect and eliminate blind pixels is of great importance. In this paper, an in-orbit real-time blind pixel detection method that combines the time-domain noise of imaging sensor and the characteristics of non-uniform correction coefficients is proposed. Firstly, according to the in-orbit working condition of the remote sensing camera, the imaging data of the reference source that under high and low temperature circumstances is acquired, and the pixels of which the noise exceeds the threshold are marked out through the real-time analysis of time-domain noise of the imaging data. Secondly, a time domain filter is applied to the imaging data of both high temperature reference source and low temperature reference source to reduce noise interference. The two-point correction is then implemented on the filtered imaging data to obtain the gain correction coefficient and the offset correction coefficient of each pixel, and a limitation process is given on the pixel coefficients which are beyond range of bit width or cannot be corrected by hardware. After that the statistical distribution chart of coefficients is acquired through the statistical analysis of the gain coefficient and offset coefficient of all pixels, and according to the statistical characteristic value “μ” and “σ”, the pixels of which the distribution of its coefficients exceeds “±3σ” boundary are marked out. Finally, seeking the union of the two marking position sets, and the marking position of the “union” indicates the coordinate position of the blind pixel detected in real time in orbit. The method proposed in this paper takes the use of the in-orbit calibration mode of the remote sensing camera to realize the in-orbit real-time detection of blind pixels, which made a superiority of this method in identification accuracy comparing to the laboratory reference-source based blind pixel detection method, as the imaging mode and the environmental circumstance of the camera are based on actual working condition. Moreover, comparing to the scene-based detection methods, this method proposed proceeds before ground imaging, which means it does not lead to the misjudgment of infrared small targets to blind pixels. According to the experiment data comparison, the method proposed in this paper provides a matching rate of blind pixel identification of above 90% to the result obtained through the way of the Military Standard of China under the same condition, which demonstrates it has the capability to be widely applied to the infrared remote sensing cameras that have reference-source attached.
|
