RECO flow

Flow RAW data to reconstructed quantities

The acquisition system provides images with 2304×2304 pixels captured by the Orca Fusion CMOS sensor. Each pixel provides a response, here called intensity, proportional to the number of collected photons added to a baseline, also known as pedestal, which can be defined as the intensity value corresponding to zero photons, however, it can vary from pixel to pixel. We can define some parameters:

• µi is the mean of the pixel i in the time axis (e.g. over 1000 images), defined here as pixel baseline;
• σi is the standard deviation of the pixel i in the time axis (e.g. over 1000 images), defined here as average noise;

Noise Suppression

On the algorithm side, it’s important to understand the pre-processing steps:

1. The upper limit allows to remove pixels with a too large intensity, produced mainly due to leakage currents that go into sensor wells - also known as hot-pixels;
2. Pedestal subtraction is carried out pixel by pixel by subtracting µi from their original intensity values, generating new intensity values defined as Ii.
3. The Noise Thresholding is the lower limit, and it is set as n times σ. The lower limit was optimized and set to be just above the noise level to ensure a good detection efficiency, but not too low in order not to overload the event-reconstruction algorithm with pixels dominated by noise. Pixels outside those limits have their intensities reset to zero.
4. Acceptance cuts are custom cuts in X and Y that allows the user to remove the noise edges of the images.
5. Images are then rescaled for CPU reasons, so that each ResxRes matrix, called macro-pixel, is assigned an intensity value corresponding to the average of the intensities Ii of the ResxRe pixels occupying the same area of the sensor.
6. The rescaled image goes then through a filtering stage based on a 2x2 median filter that replaces a given macro-pixel intensity by the median of all macro-pixels in its neighborhood.

Clustering

Finally, the coordinates (X, Y) and respective intensities (Z) of the pixels with non-zero Ii values are sent to the clustering algorithm whose output is used to extract clusters' features. In the algorithm the image is called Full Resolution (FR) until it passes through the Rescaled step (6). And it is called Zero Suppressed (ZS) after it passes through the Noise Thresholding (4). And we also have another step which is the vignette correction. This correction is applied to the FR and FR_ZS image and used only for energy calculation (Not to the clustering algorithm). And we will call it FR_VI and FR_ZS_VI.

The current clustering algorithm for "autumn21" branch of the code is Directional DBSCAN.