Abstract:Wide field small aperture telescopes are widely used for optical transient observations. Detection and classification of astronomical targets in observed images are the most important and basic step. In this paper, we propose an astronomical targets detection and classification framework based on deep neural networks. Our framework adopts the concept of the Faster R-CNN and uses a modified Resnet-50 as backbone network and a Feature Pyramid Network to extract features from images of different astronomical targets. To increase the generalization ability of our framework, we use both simulated and real observation images to train the neural network. After training, the neural network could detect and classify astronomical targets automatically. We test the performance of our framework with simulated data and find that our framework has almost the same detection ability as that of the traditional method for bright and isolated sources and our framework has 2 times better detection ability for dim targets, albeit all celestial objects detected by the traditional method can be classified correctly. We also use our framework to process real observation data and find that our framework can improve 25 % detection ability than that of the traditional method when the threshold of our framework is 0.6. Rapid discovery of transient targets is quite important and we further propose to install our framework in embedded devices such as the Nvidia Jetson Xavier to achieve real-time astronomical targets detection and classification abilities.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to effectively detect and classify astronomical objects in wide - field - of - view small - aperture telescopes (WFSATs). Specifically, the paper proposes a framework based on deep neural networks to handle this problem, in order to improve the detection ability of astronomical objects (especially those with low signal - to - noise ratio or extended objects), and perform automatic classification. Traditional methods have limitations in detecting low signal - to - noise ratio objects or extended objects, and this framework aims to overcome these limitations by combining deep - learning techniques, especially using the concept of Faster R - CNN, thereby enhancing the performance of WFSATs in transient celestial body detection tasks. ### Core Problems of the Paper 1. **Challenges in Detection and Classification**: - **Detection Problem**: Although traditional detection algorithms (such as SExtractor) are robust, they perform poorly in detecting low signal - to - noise ratio objects or extended objects, resulting in many potential astronomical objects being missed. - **Classification Problem**: Even if classification algorithms can achieve high precision and recall rates, if the detection step fails to identify the objects, these classification algorithms cannot play a role. 2. **Solutions**: - **Deep - Learning Framework**: The paper proposes a deep - learning framework based on Faster R - CNN. This framework uses a modified ResNet - 50 as the backbone network and combines the Feature Pyramid Network (FPN) to extract image features. - **Multi - scale Detection**: By using anchor boxes of different sizes and aspect ratios, the framework can detect objects at different scales, thereby improving the detection ability of low signal - to - noise ratio and extended objects. - **Real - time Processing**: In order to achieve real - time detection and classification, the paper suggests deploying this framework on embedded devices, such as Nvidia Jetson Xavier. ### Main Contributions - **Improved Detection Ability**: Experimental results show that this framework has twice the detection ability of traditional methods in detecting low signal - to - noise ratio objects. - **Classification Accuracy**: For the detected objects, this framework can provide high - precision classification results. - **Real - time Application**: Through deployment on embedded devices, this framework can achieve real - time astronomical object detection and classification, which is very important for the rapid discovery of transient celestial bodies. ### Technical Details - **Feature Extraction**: Use FPN to extract features from images of different scales. - **Region Proposal**: Generate candidate regions through RPN and adjust the position and size of these regions through bounding box regression. - **RoI Pooling/Alignment**: Map candidate regions of different sizes to the same size for classification and position regression. - **Classification and Regression**: Perform final classification and position regression through fully - connected layers and output the position and type of the object. ### Experimental Verification - **Data Generation**: Use Monte Carlo simulation to generate observational data of WFSATs, including point - like sources and streak - like sources. - **Data Labeling**: The generated data contains labels of the position and type of the object for training and testing the framework. - **Performance Evaluation**: Evaluate the performance of the framework through the mAP metric and compare it with traditional methods. The results show that this framework has a significant advantage in detecting low signal - to - noise ratio objects. In conclusion, this paper solves the key problems in the detection and classification of astronomical objects in WFSATs by proposing a deep - learning - based framework, especially in dealing with the deficiencies in handling low signal - to - noise ratio and extended objects, and provides a new technical means for the rapid discovery of transient celestial bodies.

Detection and Classification of Astronomical Targets with Deep Neural Networks in Wide Field Small Aperture Telescopes

Effective Framework for Space Target Detection Through Atmospheric Turbulence

The Deep Neural Network Based Photometry and Astrometry Framework for Wide Field Small Aperture Telescopes

Optical Transient Object Classification in Wide Field Small Aperture Telescopes with Neural Networks

Optical Transient Object Classification in Wide-field Small Aperture Telescopes with a Neural Network

A Deep Learning Based Astronomical Target Detection Framework for Multi-Colour Photometry Sky Survey Projects

Small and Dim Point Target Detection in Special Feature Space

Deep Neural Network closed-loop with raw data for optical resident space object detection

Smart Obervation Method with Wide Field Small Aperture Telescopes for Real Time Transient Detection

Fast Inspection and Accurate Recognition of Target Objects for Astronaut Robots Through Deep Learning

Rapid Automatic Multiple Moving Objects Detection Method Based on Feature Extraction from Images with Non-sidereal Tracking

StarNet: Convolutional Neural Network for Dim Small Target Extraction in Star Image

PNet -- A Deep Learning Based Photometry and Astrometry Bayesian Framework

Detection, instance segmentation, and classification for astronomical surveys with deep learning ( deepdisc ): detectron2 implementation and demonstration with Hyper Suprime-Cam data

Identifying Light-curve Signals with a Deep-learning-based Object Detection Algorithm. II. A General Light-curve Classification Framework

Target Detection Framework for Lobster Eye X-Ray Telescopes with Machine-learning Algorithms

Deep Learning-Based Small Target Detection for Satellite–Ground Free Space Optical Communications

Infrared Small Target Detection Based on a Temporally-Aware Fully Convolutional Neural Network

A Real-time Algorithm for Weak Small Targets Detection in Astronomical Opto-electronic Image Sequences

Detecting optical transients using artificial neural networks and reference images from different surveys