Abstract:Detecting Resident Space Objects (RSOs) and preventing collisions with other satellites is crucial. Recently, deep convolutional neural networks (DCNNs) have shown superior performance in object detection when large-scale datasets are available. However, collecting rich data of RSOs is difficult due to very few occurrences in the space images. Without sufficient data, it is challenging to comprehensively train DCNN detectors and make them effective for detecting RSOs in space images, let alone to estimate whether a detector is sufficiently robust. The lack of meaningful evaluation of different detectors could further affect the design and application of detection methods. To tackle this issue, we propose that the space images containing RSOs can be simulated to complement the shortage of raw data for better benchmarking. Accordingly, we introduce a novel simulation-augmented benchmarking framework for RSO detection (SAB-RSOD). In our framework, by making the best use of the hardware parameters of the sensor that captures real-world space images, we first develop a high-fidelity RSO simulator that can generate various realistic space images. Then, we use this simulator to generate images that contain diversified RSOs in space and annotate them automatically. Later, we mix the synthetic images with the real-world images, obtaining around 500 images for training with only the real-world images for evaluation. Under SAB-RSOD, we can train different popular object detectors like Yolo and Faster RCNN effectively, enabling us to evaluate their performance thoroughly. The evaluation results have shown that the amount of available data and image resolution are two key factors for robust RSO detection. Moreover, if using a lower resolution for higher efficiency, we demonstrated that a simple UNet-based detection method can already access high detection accuracy.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the challenge of automatically detecting Resident Space Objects (RSOs) in single - frame space images. Specifically, due to the low frequency of RSOs in space images and scarce data, traditional detection methods are difficult to be effectively trained and evaluated. In addition, the lack of effective evaluation methods also affects the design and application of detection methods. To solve these problems, the authors propose a Simulation - Augmented Benchmarking framework for RSO Detection (SAB - RSOD), which supplements the shortage of actual data by simulating and generating space images containing RSOs, thereby improving the training effect and evaluation accuracy of detectors. ### Core Problems of the Paper 1. **Data Scarcity Problem**: The data containing RSOs in actual space images is very limited, which leads to the problem of insufficient data when training Deep Convolutional Neural Network (DCNN) detectors. 2. **Evaluation Difficulties**: The lack of effective evaluation methods makes it difficult to comprehensively evaluate the performance of different detectors in the RSO detection task. 3. **Detector Design and Application**: Due to the above problems, it becomes difficult to design and apply effective RSO detection methods. ### Solutions 1. **High - Fidelity Simulator**: A high - fidelity RSO simulator has been developed using the hardware parameters of real sensors, which can generate various realistic space images and enrich the training data set. 2. **Automatic Labeling**: The images generated by the simulator can be automatically labeled, reducing the workload of manual labeling. 3. **Mixed Data Set**: Combine the synthetic images with the existing real - world images to form a data set containing about 500 images for training detectors. 4. **Benchmarking Framework**: A Simulation - Augmented Benchmarking framework (SAB - RSOD) is proposed for training and evaluating different popular detectors, such as YOLO and Faster R - CNN. ### Main Contributions 1. **High - Fidelity Simulation System**: Based on the parameters of real sensors, a system that can simulate space images has been developed, in which RSOs are presented as stripes and these stripes can be automatically labeled. 2. **New Data Set and Benchmark**: A new simulation - augmented space - tracking data set and benchmark are proposed, which is the first time to use real - world Swarm Spaceborne Star Tracker (SST) images combined with synthetic images to train and validate DCNN detectors. 3. **Detector Evaluation**: Two popular detectors (Faster R - CNN and YOLO) are evaluated, and a simple UNet - based detector is further developed for comparative study. 4. **Performance Analysis**: Through simulated data, the performance of different detectors in the RSO detection task is revealed. The research shows that the amount of training data and image resolution are two key factors affecting performance. In addition, even on low - resolution images, the simple UNet - based detector can also show high accuracy. ### Formula Examples - **Gaussian Point Spread Function (PSF)**: \[ f(x,y)=\frac{A}{2\pi\sigma_x\sigma_y}\exp\left(-\frac{(x - x_i)^2}{2\sigma_x^2}-\frac{(y - y_i)^2}{2\sigma_y^2}\right) \] where \(A\) is the amplitude, \(x_i\) and \(y_i\) are the pixel positions obtained from formula (6), and \(\sigma_x\) and \(\sigma_y\) are the corresponding Gaussian variances. - **Electron Number Calculation**: \[ n_e=\frac{1}{h}I_{\text{Sun}}t_{\text{int}}\sigma_x\sigma_y\rho_x\rho_y\eta\times 10^{-(m^* - m_{\text{Sun}})/2.5} \] where \(h = 6.626\times 10^{-34}\) J·s is Planck's constant, \(t_{\text{int}}\) is the integration/exposure time of the star sensor, \(\rho_x=n_x x_p\) and \(\rh

A Simulation-Augmented Benchmarking Framework for Automatic RSO Streak Detection in Single-Frame Space Images

Simulation of RSO Images for Space Situation Awareness (SSA) Using Parallel Processing

Finding a Needle in a Haystack: Faint and Small Space Object Detection in 16-Bit Astronomical Images Using a Deep Learning-Based Approach

RSDS: A Specialized Loss Calculation Method for Dense Small Object Detection in Remote Sensing Images

A New Spatial-Oriented Object Detection Framework for Remote Sensing Images

SDSDet: A real-time object detector for small, dense, multi-scale remote sensing objects

RS-YOLOX: A High-Precision Detector for Object Detection in Satellite Remote Sensing Images

MOON: A Subspace-Based Multi-Branch Network for Object Detection in Remotely Sensed Images

A Refined and Efficient CNN Algorithm for Remote Sensing Object Detection

ORSI Salient Object Detection via Progressive Semantic Flow and Uncertainty-Aware Refinement

SOCDet: A lightweight and accurate oriented object detection network for satellite on-orbit computing

ORSI Salient Object Detection via Multiscale Joint Region and Boundary Model

One-stop multiscale reconciliation attention network with scribble supervision for salient object detection in optical remote sensing images

A Self-Supplementary and Revised Network for Remote Sensing Object Detection

ORSIm Detector: A Novel Object Detection Framework in Optical Remote Sensing Imagery Using Spatial-Frequency Channel Features

RSSOD-Bench: A large-scale benchmark dataset for Salient Object Detection in Optical Remote Sensing Imagery

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

Self-Training-Based Semantic-Balanced Network for Weakly Supervised Object Detection in Remote-Sensing Images

RSI-CB: A Large Scale Remote Sensing Image Classification Benchmark via Crowdsource Data

Concerning Imbalance and Bounding Box Loss to Detect Small Targets in Remote Sensing

Transcending Pixels: Boosting Saliency Detection via Scene Understanding from Aerial Imagery