Autonomous Detection for Spacecraft Assembly State Based on Deep Learning Method

Currently, the inspection of spacecraft assembly status primarily relies on manual inspection, which is inefficient and struggles to meet the demands of future satellite mass production. Real-time detection of assembly status through object detection models can effectively resolve this problem. In this study, existing assembly data were collected and processed through annotation, data augmentation, and other preprocessing steps to form a dataset. The YOLO11-I-OBB model was employed for transfer learning, incorporating the Attention module CBAM into its network architecture, resulting in a model named YOLO11-I-OBB-ht. The YOLO11-I-OBB-ht model demonstrates significant potential in detecting missing assembly components such as screw absence, adhesive presence verification on screws, and electrical connector mating status in spacecraft assembly, leveraging its superior bounding box detection capabilities and precise recognition of oriented objects. Experimental results indicate that the proposed spacecraft assembly status detection model achieves a precision (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{P}$</tex>) of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 2. 1 \%}$</tex>, a recall (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}$</tex>) of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 9. 1 \%}$</tex>, and a mean average precision (mAP50) of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 5. 2 \%}$</tex>. Compared to the YOLOv81-OBB model, the precision is improved by 4.2 %, recall by 1.8 %, and mAP50 by 2.3 %. Compared to the YOLO11-I-OBB baseline model, there is an improvement of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1. 7 \%}$</tex> in Precision (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{P}$</tex>), <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1. 5 \%}$</tex> in Recall <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(\mathrm{R})$</tex>, and 1.6 % in mAP50. The findings of this study demonstrate that the YOLO11-I-OBB-ht model exhibits significant performance advantages and robustness in detecting spacecraft typical assembly configuration, providing crucial technical support for rapid inspection of typical spacecraft assembly status. This research holds important reference value and practical significance for enhancing the efficiency of spacecraft production lines.