Detection and Recognition of Key Parts of Space Targets Based on Vision

The detection and identification of key components of space targets is the key of in-orbit vehicle repair, maintenance and refueling, and is also the premise of a large number of in-orbit vehicle maintenance and inspection in the future. In this paper, a vision-based detection and recognition method for key parts of space targets is designed. According to the data features, the traditional Convolutional Neural Network (CNN)-based method is improved, and the target area is preliminarily eliminated by combining the statistical characteristics of histogram, which ensures that most of the starry background is not calculated in the subsequent operation. After eliminating the remaining effective area, the target is detected and identified first, and the key components are identified after the size of the target is determined, which effectively reduces the redundant calculation of the data that is not suitable for component identification. At the same time, the component recognition algorithm and the aircraft target recognition algorithm adopt a unified computational architecture, which reduces the complexity of the engineering design process. In this paper, the peripheral equipment and data testing framework in the engineering process are given. Digital Signal Processor (DSP) and Field-programmable gate array (FPGA) chip are used to complete the whole calculation. The test results show that the recognition accuracy is more than $94 \%$, which is comparable to the previous design methods. At the same time, special tests of low signal-to-noise ratio (SNR), small target and different poses are carried out to obtain the recognition accuracy under different parameters. Through the experimental results and design framework, it can be seen that the design method effectively realizes the detection and recognition of key components of space targets, and the algorithm has good adaptability and small computation.