| Name: | Srinadh Reddy Bhavanam |
| Affiliation: | Indian Institute of Technology, Hyderabad |
| Conference ID : | ASI2022_312 |
| Title : | Attention Enhanced Deep Learning for Cosmic Ray Detection in Astronomical Images |
| Authors : | Srinadh Reddy Bhavanam, Sumohana S. Channappayya, P.K. Srijith, Shantanu Desai |
| Abstract Type: | Poster |
| Abstract Category : | Instrumentation and Techniques |
| Abstract : | Wide-field astronomical surveys are often plagued by unwanted artifacts that arise from a number of sources including imperfect optics, faulty image sensors, cosmic ray (CR) hits, and even airplanes and artificial satellites. The CR hits are caused due to high energy transfer from external particles in the environment to the detector electrons in a valence band, resulting in excess of charge in the affected regions and is difficult to avoid. The identification and mitigation of the CR artifacts is important to ensure top quality data is used for rigorous astronomical analysis. For this purpose, we have developed and tested a novel Deep Learning-based framework for the automatic detection of CR hits from astronomical imaging data from two imaging sources: Dark Energy Camera (DECam) and Las Cumbers Observatory Global Telescope (LCOGT). deepCR and Cosmic-CoNN are the current state-of-the-art learning-based algorithms for CR detection. These models were trained using Hubble Space Telescope (HST) ACS/WFC and LCOGT Network images respectively, and are used as the backbone in our study. Additionally, we have introduced attention modules into these backbones to evaluate their efficacy on CR detection performance. We show that the introduction of attention modules results in a consistent improvement over deepCR and Cosmic-CoNN. Further, the attention modules provide better explainability for the performance of the models. The models were trained on DECam data and validated on both DECam and LCOGT data. Our validation results show that the proposed model outperforms current state-of-the-art models. Through these experiments, we demonstrated the efficacy of attention modules in improving CR detection performance. |