Cosmic Ray Signal Detection
Charged particles (cosmic rays) passing through the CMOS camera image sensor, a silicon semiconductor sensor, generate electrons and positive holes through Coulomb force. These electrons are attracted to and collected by the electrodes of the pixels. When visible light hits the sensor, electrons are also generated through the photoelectric effect and collected by the electrodes. Therefore, if a reaction similar to when visible light hits occurs in complete darkness with no light, it can be identified as a cosmic ray signal.
The number of collected electrons determines the strength of the voltage signal, which is then converted into pixel information in a digital image. A higher voltage results in a strong signal, and the pixel becomes brighter. In cases where few electrons are collected, the signal is weak, resulting in a darker pixel.
Since cosmic ray signals are extremely weak, it requires searching for faint light in dark images.
Noise Reduction
CMOS sensors may produce noise that closely resembles cosmic ray signals. It is difficult to distinguish them based on shape and brightness alone. However, noisy pixels often generate noise repeatedly in the same location. Therefore, it is crucial to identify and carefully remove pixels where noise is generated. As the number of pixels generating noise tends to increase over time with long exposure and repeated shooting, it is also necessary to periodically identify the locations of noisy pixels.
Left: Noise Right: Signal
Detection Algorithm
We use Python with OpenCV for signal detection. The detection steps are as follows. Since the amount of noise varies depending on the sensor or device, the detection conditions are adjusted accordingly while exploring suitable thresholds.
- Convert the image to grayscale with an appropriate threshold
- Detect reactive regions of pixels
- Obtain the width and length of the reactive regions through rectangle detection
- If there are regions that meet the conditions, crop and save the corresponding regions from the image