Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep machine learning for meteor monitoring: advances with transfer learning and gradient-weighted class activation mapping

Eloy Peña-Asensio, Josep M. Trigo-Rodríguez, Pau Grèbol-Tomàs, David Regordosa-Avellana, Albert Rimola

TL;DR

The study addresses the data bottleneck in meteor monitoring by developing a CNN-based pipeline that uses transfer learning with a ResNet-34 backbone to detect meteors and a Grad-CAM–based method to localize them within frames. By fusing ROI information from Grad-CAM on the last convolutional layer with high-resolution activations from the initial layer, the approach achieves precise meteor tracking, validated on SPNM data with a meteor precision of 0.98 and overall test accuracy of 0.96. Training on a relatively small, carefully balanced dataset demonstrates strong generalization, aided by data augmentation. This work significantly reduces manual workload for meteor scientists and enables automatic extraction of velocity curves and potential orbital elements from large meteor datasets.

Abstract

In recent decades, the use of optical detection systems for meteor studies has increased dramatically, resulting in huge amounts of data being analyzed. Automated meteor detection tools are essential for studying the continuous meteoroid incoming flux, recovering fresh meteorites, and achieving a better understanding of our Solar System. Concerning meteor detection, distinguishing false positives between meteor and non-meteor images has traditionally been performed by hand, which is significantly time-consuming. To address this issue, we developed a fully automated pipeline that uses Convolutional Neural Networks (CNNs) to classify candidate meteor detections. Our new method is able to detect meteors even in images that contain static elements such as clouds, the Moon, and buildings. To accurately locate the meteor within each frame, we employ the Gradient-weighted Class Activation Mapping (Grad-CAM) technique. This method facilitates the identification of the region of interest by multiplying the activations from the last convolutional layer with the average of the gradients across the feature map of that layer. By combining these findings with the activation map derived from the first convolutional layer, we effectively pinpoint the most probable pixel location of the meteor. We trained and evaluated our model on a large dataset collected by the Spanish Meteor Network (SPMN) and achieved a precision of 98\%. Our new methodology presented here has the potential to reduce the workload of meteor scientists and station operators and improve the accuracy of meteor tracking and classification.

Deep machine learning for meteor monitoring: advances with transfer learning and gradient-weighted class activation mapping

TL;DR

The study addresses the data bottleneck in meteor monitoring by developing a CNN-based pipeline that uses transfer learning with a ResNet-34 backbone to detect meteors and a Grad-CAM–based method to localize them within frames. By fusing ROI information from Grad-CAM on the last convolutional layer with high-resolution activations from the initial layer, the approach achieves precise meteor tracking, validated on SPNM data with a meteor precision of 0.98 and overall test accuracy of 0.96. Training on a relatively small, carefully balanced dataset demonstrates strong generalization, aided by data augmentation. This work significantly reduces manual workload for meteor scientists and enables automatic extraction of velocity curves and potential orbital elements from large meteor datasets.

Abstract

In recent decades, the use of optical detection systems for meteor studies has increased dramatically, resulting in huge amounts of data being analyzed. Automated meteor detection tools are essential for studying the continuous meteoroid incoming flux, recovering fresh meteorites, and achieving a better understanding of our Solar System. Concerning meteor detection, distinguishing false positives between meteor and non-meteor images has traditionally been performed by hand, which is significantly time-consuming. To address this issue, we developed a fully automated pipeline that uses Convolutional Neural Networks (CNNs) to classify candidate meteor detections. Our new method is able to detect meteors even in images that contain static elements such as clouds, the Moon, and buildings. To accurately locate the meteor within each frame, we employ the Gradient-weighted Class Activation Mapping (Grad-CAM) technique. This method facilitates the identification of the region of interest by multiplying the activations from the last convolutional layer with the average of the gradients across the feature map of that layer. By combining these findings with the activation map derived from the first convolutional layer, we effectively pinpoint the most probable pixel location of the meteor. We trained and evaluated our model on a large dataset collected by the Spanish Meteor Network (SPMN) and achieved a precision of 98\%. Our new methodology presented here has the potential to reduce the workload of meteor scientists and station operators and improve the accuracy of meteor tracking and classification.
Paper Structure (7 sections, 5 figures, 4 tables)

This paper contains 7 sections, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Artificial intelligence for meteor detection
  3. CNN, transfer learning, and Grad-CAM
  4. Detection
  5. Tracking
  6. Training and results
  7. Conclusions

Figures (5)

  • Figure 1: A basic-block building block of ResNet-34.
  • Figure 2: The general process of class activation mapping method. Adapted from Jiang2021ITIP.
  • Figure 3: Training and validation loss during model training.
  • Figure 4: Confusion matrix of the trained model with normalized values in parentheses.
  • Figure 5: Top panel: False positive of a SpaceX Starlink satellite track as it exhibits similar characteristics as a meteor trail. Recording obtained from the Alpicat SPMN station under the operation of Marc Corretgé-Gilart. Bottom panel: False negative of SPMN070523G superbolide recorded near the full moon with a cloudy sky. Recording obtained from Bartolo-Castelló SPMN station under the operation of Vicente Ibañez