Speaker
Description
This study aims to optimize the energy resolution of the Topmetal-II pixel detector through the utilization of neural networks and curve-fitting strategies. The Topmetal-II plays a crucial role in capturing soft X-ray signals, enhancing its energy resolution is crucial for accurate measurements.
We employed neural networks, specifically Convolutional Neural Networks (CNNs), to train on the energy data of Topmetal-II, aiming to enhance its performance. Simultaneously, we employed curve-fitting techniques to model the response of Topmetal-II. Through the fitting of mathematical curves to the detector's energy response, our objective was to optimize its performance and improve energy resolution.
We employed a 12-bit ADC for the continuous acquisition of the energy channel of the Topmetal-II. The final statistical results indicate that adopting a neural network structure, with encoding and decoding layers each comprising 5 layers and 2 fully connected layers, along with the use of a Denoising AutoEncoder (DAE), yielded promising results. A comparison between DAE outputs and inputs showed a reduction in RMSE from 0.0447 to 0.0132, a decrease by a factor of 3.38, resulting in a 9.54% improvement in energy resolution. In curve fitting, the output function of the charge-sensitive amplifier was employed for data fitting. The RMSE decreased from 0.05385 to 0.01423, a reduction by a factor of 3.78, leading to an 11.1% enhancement in energy resolution. These results robustly support the progression of soft X-ray detection technology, indicating the potential for delivering more precise and reliable measurement outcomes in low-energy polarization detector (LPD) experiments.
| Minioral | No |
|---|---|
| IEEE Member | No |
| Are you a student? | Yes |
