Extremely low count detection for EELS spectrum imaging by reducing CCD read-out noise

Extremely low count detection for EELS spectrum imaging is required to overcome problems with electron irradiation and widen the range of available applications. We have made a systematic statistical study of the reduction of CCD noise for EELS. We propose a calculation method to estimate the properties of noise and a procedure to reduce it. Since the dominant noise is a practically random component, it can be reduced by subtracting the population mean of the dark reference and a summation over an appropriate number of spectra, depending on the standard deviation of the noise. A gain-averaging method can further improve the signal-to-noise (SN) ratio. It is thereby demonstrated that a high-SN spectrum can be obtained even for a single-count core-loss signal. The present method would be useful for measuring low signal spectrum such as monochromated spectra and for radiation sensitive materials.