A Detection Algorithm Based on Improved Faster R-CNN for Spacecraft Components
Zongqi Wang, Yue Cao, Jun Li
Abstract
Due to the increasing need of science and technology, more and more spacecraft are in space. Once the spacecraft reaches its useful life, it will become floating junk in space. Active space debris removal technology is an effective method to solve space garbage. With the rapid iteration of deep learning methods, target detection technology in the field of computer vision has made a breakthrough. In this paper, a Faster R-CNN model based on RegNet as the main trunk network is proposed, which can reduce the loss of key information caused by R-CNN sampling and effectively improve the target detection effect. We will collect the photos of space spacecraft as training sets and verification sets sent to the network for iterative training. The test results of the data set in this paper show that: With RegNet as the main backbone network, the Faster R-CNN model has better test results than other target detection algorithms under the same parameters. Compared with other backbone networks added to Faster R-CNN, the mAP value is improved by at least 9.6%. Compared with other advanced target detection algorithms, mAP value increased by at least 26.3%. When processing low Earth orbit target identification tasks, the Faster R-CNN model based on RegNet as the main trunk network has a strong recognition ability for the image targets whose color is similar to the background color, relatively fuzzy and small scale.