A novel automated pipeline for meteor classification employing the artificial intelligence-based Convolutional Neural Networks (CNNs) is presented alongside practical applications.

We present a novel automated pipeline for meteor classification, employing Convolutional Neural Networks (CNNs). This system addresses the challenge of differentiating between meteor and non-meteor images, a process traditionally performed manually. Our methodology demonstrates significant efficacy in identifying meteors amidst diverse backgrounds. Key features of this pipeline include automated classification using CNNs for efficient and accurate meteor identification, enhanced model performance through the incorporation of transfer learning with ResNet-34, and precise meteor tracking using Gradient-weighted Class Activation Mapping (Grad-CAM) for internal layer communication. The system achieves a 98% success rate in meteor detection, validated on a dataset from the Spanish Fireball and Meteorite Recovery Network (SPMN-CSIC). The introduction of this automated pipeline represents a significant advancement in meteor studies, contributing to a deeper understanding of meteoroid influx and aiding in the recovery of fresh meteorites. Its high precision and efficiency promise to alleviate the workload of researchers and station operators, while improving the accuracy of meteor tracking and classification.

This work has been published in Planetary and Space Science.

Link to the article as open access in Planet. Space Sci.: https://www.sciencedirect.com/science/article/pii/S003206332300171X?via%3Dihub