Off-axis afocal system misalignment detection based on wavelet transform and improved ShuffleNetV2

In optical systems, misalignment of optical components caused by various factors such as gravity, the environment, and platform structure jitter can reduce the imaging quality of light spot images. The non-rotationally symmetric optical path structure and afocal characteristics of off-axis reflective afocal systems make these systems more sensitive to misalignment errors. Through neural networks and defocused spot images, it is possible to solve for the misalignment of the secondary mirror without using a Hartmann sensor. However, the current research difficulty is that when there is a decenter misalignment error on the Z-axis of the secondary mirror in the off-axis system, the accuracy of solving for the misalignment of the remaining four degrees of freedom of the secondary mirror will decrease, and correcting the Z-axis decenter misalignment first and then correcting other misalignments will affect the real-time performance of the algorithm. Therefore, this paper proposes an algorithm for solving misalignment based on wavelet transformation and an improved ShuffleNetV2. The misalignment is solved by establishing a mapping relationship between the wavelet transformation features of the defocused spot image and the misalignment through a lightweight deep learning model. Some types of aberrations have obvious directional characteristics, and wavelet transform can extract multi-directional texture features from defocused spot images to reduce the difficulty of decoupling the neural network. The Squeeze-and-Excitation Module is added to the original ShuffleNetV2 to improve the performance of the network, thereby improving the accuracy of misalignment solving. Finally, experiments were conducted using an actual off-axis two-mirror afocal system to prove the feasibility and superiority of this algorithm.