| The deep space exploration mission requires the space imaging system to keep the boresight in a stable state during its exposure time.However,due to the vibration of various subjective and objective factors such as the attitude of the satellite turntable,the boresight is shifted,which in turn leads to a decrease in the imaging quality of the detector.In order to improve the accuracy of the space detection system and improve the imaging quality of star points,a high-precision image stabilization control system must be designed for the telescope system.Based on the above,this paper studies the high-precision image stabilization related algorithms of the space astronomical telescope precision image stabilization system.For the high real-time and highprecision requirements of the space telescope,the image stabilization hardware system and software processing algorithm are designed.The main research contents are as follows:(1)By analyzing the imaging characteristics of the star map,a 3×3 FIR low-pass filter is designed to improve the positioning accuracy of the star points,and a doublethreshold star point extraction method is designed for the image stabilization mode,which can effectively extract the stars in the star map.Bright stars.(2)For low signal-to-noise ratio star point extraction,a deep learning-based SEAC-UNet star point extraction algorithm is proposed.The algorithm uses the SEAC module for encoding,that is,the SE-Net module is connected in series after the AC-Net module uses asymmetric convolution to increase the feature extraction capability,and the SE-Net module is used to replace the span connection after feature extraction,which effectively improves the network performance.3000 simulated star map datasets with noise and uneven background are constructed,and the simulation results show that the algorithm has higher detection rate and lower false alarm rate.(3)The simulation compares the existing mainstream sub-pixel centroid extraction algorithms,and analyzes the systematic error of the centroid algorithm.The 5×5calculation window square weighted centroid method with balanced accuracy and realtime performance is selected,and the detector is analyzed.The influence of noise on the positioning accuracy,the GA-GRNN algorithm is proposed to compensate the system error,and the simulation experiment proves that the error after compensation reaches 3.0 × 10-5,which is higher than the polynomial fitting compensation and BP compensation algorithm,and when the SNR is 3,the centroid positioning accuracy reaches 0.0762 pixel,which is significantly higher than the accuracy before compensation;the triangulation recognition algorithm is improved according to hash search,which effectively improves the recognition speed.(4)According to the requirements of high sampling rate and high precision,the hardware system of the precision image stabilization system is designed and the device selection is carried out;DDR image cache and embedded implementation of highprecision optical axis shift detection algorithm.(5)Test the CMOS image acquisition and buffering of the image stabilization detector.Set the exposure time to 20 ms.After receiving the image through the ground detection,the display frame rate on the host computer reaches 50 FPS,which meets the sampling requirements of the optical stabilization system.By comparing the DDR After the image data is cached,it is exactly the same as the image data saved after acquisition,which proves the correctness of the image data caching;the simulation test and real star map test of the real-time optical axis offset detection algorithm for precise image stabilization are carried out,and the detection accuracy reaches 0.04 The pixel(1/25pixel)is higher than the system requirements,and the correctness of the algorithm implemented on the hardware platform is verified by experiments,and the processing time of the optical axis offset detection algorithm on the hardware platform reaches7.98 ms,which meets the real-time image stabilization system.sexual requirements. |
PDF Full Text Request