The adaptive selection of the laser centre positioning algorithm based on convolutional neural network

This paper proposes an adaptive selection of a laser centre positioning algorithm based on a Convolutional Neural Network. By simulating the transmission process of Gaussian beams in the atmospheric turbulence environment, the laser spot images with noise were generated successfully. To precisely handle these images, the four laser spot centre positioning algorithms are utilised, namely the centroid method, the circular fitting method, the Hough transform method and the maximum chord length method. Meanwhile, the CNN model was constructed and trained to achieve the classification and regression tasks of laser spot images. The experimental results clearly claim that, compared with the other two classical CNN models, the model in this paper demonstrates outstanding accuracy and robustness in the classification of the laser spot images.