e-dreamer-lqe-dreamer-lq
11

Astronomical_Target_Detection

screenshot of Astronomical_Target_Detection

This repository was created for the paper "Detection and Classification of Astronomical Targets with Deep Neural Networks in Wide Field Small Aperture Telescopes"

Github

Overview

The detection and classification of astronomical targets is a crucial aspect of data processing for wide field small aperture telescopes. Recent advancements using deep neural networks have demonstrated great potential in improving the accuracy and efficiency of these processes. This framework leverages technologies such as Faster R-CNN and a modified ResNet-50 backbone to enhance the capabilities of astronomical observations, particularly in identifying dim targets that traditional methods might miss.

By employing transfer learning and simulated images for initial training, this framework not only reduces the need for extensive datasets but also shows impressive results when tested against both simulated and real observation data. The idea of implementing this framework in embedded devices further allows for real-time detection, which could greatly benefit astronomers.

Features

  • Enhanced Detection: Utilizes advanced deep neural network architectures like Faster R-CNN to improve the detection capabilities of dim astronomical targets significantly.
  • Transfer Learning: Reduces the requirements for large training datasets by initially training with simulated images and refining with minimal real data, making it more accessible.
  • Modified Backbone: Incorporates a modified ResNet-50 network as its backbone, providing a robust foundation for effective target classification.
  • Feature Pyramid Network (FPN): This architecture enhances the model's ability to detect targets at different scales, making it versatile in various observational conditions.
  • Real-Time Processing: The potential installation of the framework in embedded devices allows for rapid processing and real-time detection of astronomical events.
  • Python Compatibility: Designed to run on Python 3.5 and compatible with libraries like PyTorch and OpenCV, ensuring ease of integration into existing workflows.
  • Validation Support: Provides clear guidelines on how to validate the model using real observation data, giving users confidence in its performance.