Instance Segmentation for Feature Recognition on Noncooperative Resident Space Objects

Abstract

Active debris removal and unmanned on-orbit servicing missions have gained interest in the last few years, along with the possibility to perform them through the use of an autonomous chasing spacecraft. In this work, new resources are proposed to aid the implementation of guidance, navigation and control algorithms for satellites devoted to the inspection of non-cooperative targets before any proximity operation is initiated. In particular, the use of Convolutional Neural Networks (CNNs) performing object detection and instance segmentation is proposed and its effectiveness in recognizing components and parts of the target satellite is evaluated. Yet no reliable training images dataset of this kind exists to date. A tailored and publicly available software has been developed to overcome this limitation by generating synthetic images. Computer Aided Design models of existing satellites are loaded on a 3-D animation software and used to programmatically render images of the objects from different point of views and in different lighting conditions, together with the necessary ground truth labels and masks for each image. The results show how a relatively low number of iterations is sufficient for a CNN trained on such datasets to reach a mean average precision value in line with state-of-the-art-performances achieved by CNNs in common datasets. An assessment of the performance of the neural network when trained on different conditions is provided.To conclude, the method is tested on real images from the MEV-1 on-orbit servicing mission, showing that using only artificially generated images to train the model does not compromise the learning process.

Characterizing Satellite Geometry via Accelerated 3D Gaussian Splatting

“…In recent years, neural network-based computer vision techniques have shown highly promising results for performing in non-cooperative environments and are actively being used for autonomous navigation in robotics and the automobile industry, for example. Extensive research is being performed to identify and track [14] components of non-cooperative known spacecraft [15] and unknown spacecraft [10,[16][17][18], such as solar panels, antennas, bodies, and thrusters. Knowing the presence of the components and their relative positions could assist with autonomously rendezvousing to an RSO, as shown in [12,19].…”

A satellite component contour extraction method for lightweight space mobile platforms

“…Real satellite component data sets are costly, but the good news is that some researchers have broken the limitation of data set scarcity to some extent by synthesizing realistic data sets. These data sets can be used for pose estimation (Kisantal et al , 2020; Bechini et al , 2022) and even instance segmentation (Faraco et al , 2022; Proença and Gao, 2020). Considering the availability of depth information, a public data set AFDL-SCD [1] is selected as the main testing data.…”

“…So, in this paper two open-source algorithms were selected for Scheme 5 and 6, namely, the two-stage Mask R-CNN (He et al , 2020) and the single-stage SOLOv2 (Wang et al , 2020), which have 37.1 and 41.7 mask average precision true(APIoU(0.5:0.95)masktrue) on COCO test-dev, respectively. In addition, they are classic benchmark methods with stable segmentation performance (Gu et al , 2022), which are easily reproducible, widely implemented and widely used for comparison (Faraco et al , 2022). Recently, YOLO-based lightweight instance segmentation method has begun to appear, but currently its mask prediction performance fluctuates relatively large under different data sets (Hurtik et al , 2020); therefore, YOLO series approaches were not selected as a comparison method in this paper.…”

Small-Body Segmentation Based on Morphological Features with a U-Shaped Network Architecture

“…The background, surface, craters, and boulders masks are easily generated using the Cycles ray-tracing rendering engine in Blender and exploiting different identifiers assigned to the surface, craters, and boulders. This approach ¶ https://it.mathworks.com/products/matlab.html has been inspired by the work of [19]. Craters masks are obtained from the crater model while surface and boulders are obtained from the full model.…”