Deuterium-deuterium fusion charged particle detection using CR-39 and deep learning model
Year: 2025
Type: article
Abstract: Columbia Resin #39 (CR-39) solid-state nuclear track detectors are widely used in fusion research for detecting charged particles produced in fusion reactions. However, analyzing increasingly complex and large-scale CR-39 track images to extract meaningful information is a tedious and time-consuming process, often prone to human errors and bias. To address these challenges, we developed an AI-based classification model capable of differentiating protons, tritons, and helions produced during D-D fusion, using CR-39 track images as input data. The CR-39 track images were processed and used to train a deep learning model. By preprocessing the track images for noise reduction and feature enhancement, we trained the You Only Look Once version 8 (YOLOv8) network to distinguish the three particle types with high accuracy. We designed a method that cuts large images into smaller pieces and then enlarges them again to achieve a lower density in each segmented piece to improve accuracy. The sliced images are then reconstructed using overlapping cutting and Non-Maximum Suppression (NMS) to handle the overlapping parts of the stitches. By adopting NMS, we retain only the bounding box with the highest confidence level, avoiding duplicate detections and achieving accurate counting. The proposed model achieved a classification accuracy of over 94%, demonstrating its potential for improving automated track analysis in CR-39 detectors. Additionally, the model precisely identifies particle coordinates and counts, enabling comprehensive particle analysis. This study highlights the application of AI in track detection and classification, offering a robust solution for particle identification in CR-39 detector-based experiments. • Combining deep learning model YOLOv8 with CR-39 for particle detection provides a more accurate approach compared to traditional methods like ImageJ, as it eliminates the need for manually setting thresholds and reduces the impact of human bias. • For overlapping particles where the edges are unclear and particle shapes vary, image enhancement techniques effectively resolve these challenges. By applying advanced image processing and enhancement methods, overlapping particles can be accurately distinguished and counted separately. This approach significantly improves detection accuracy compared to traditional methods. • By cropping and enlarging image segments, the particle density in each sub-image is reduced, making it easier to distinguish individual particles. During the image tiling process, overlapping regions are used to maintain continuity, and when stitching the results, only high-confidence detections are retained. This approach ensures accurate counting even in high-density images, effectively handling overlapping particles and reducing counting errors. • By applying advanced image preprocessing techniques, such as background subtraction, denoising filters, and adaptive thresholding, the impact of background noise on particle detection is significantly reduced. This enhances the model's ability to accurately distinguish particles from the background, leading to more precise particle counting and minimizing false detections caused by noise. • A user-friendly UI interface has been designed to simplify the process—users can simply select the image they need to analyze, and the system will automatically provide classification and counting results. This approach ensures high efficiency and speed, significantly reducing manual effort and time costs, making the detection process more accessible and streamlined.
Source: Radiation Measurements
Language: English
Subfield: Nuclear and High Energy Physics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,690
OpenAlex ID: W4410023150