2D cluster finding with the help of enhanced deep residual networks

In recent years, advancements in image processing technology have made cluster detection a vital area of research across various scientific fields, from particle physics to astronomy. Accurately identifying clusters in images is essential for enhancing experimental outcomes and applications. In this paper, a novel algorithm for two-dimensional cluster finding using Enhanced Deep Residual Network (EDSR) is proposed. The algorithm solves the challenge of accurately identifying points in an image by processing the pixel matrix to efficiently locate clusters. The algorithm uses a three-step process: identifying local maxima, clustering around these points, and calculating a weighted average to determine the cluster center. While traditional methods have difficulty in solving the problem of equality of adjacent maxima, our method solves this problem by redefining local maximum detection to minimize bias. By considering a 5×5 matrix around each local maximum, clustering was achieved, capturing over 98.73% of the peak values and ensuring comprehensive clustering formation without overlapping issues. Before clustering analysis, the EDSR algorithm is used to improve image resolution and reduce errors in identifying cluster centers. The model evaluation results indicate that the accuracy of cluster center determination has been improved and the error margin has been reduced, highlighting the effectiveness of the EDSR enhancement method.