Research On Splicing Method Of Cultural Relic Fragments Based On Swarm Intelligence Optimization And Deep Learning

Ceramic cultural relics are cultural treasures left over from the long history of the Chinese nation,carrying a huge historical significance.However,limited by its environment and its own characteristics,ceramic cultural relics are not easy to be preserved and easily damaged,so the restoration of cultural relics is a major problem to be solved urgently.Due to the limitations of high cost and low accuracy of traditional manual splicing,the restoration of cultural relics with computer-aided technology is the key research direction.Deep learning model has outstanding performance in feature extraction and subsequent stitching work,but it also has the disadvantages of information redundancy,overfitting,and computational complexity.Therefore,this thesis introduces a swarm intelligence algorithm to solve large-scale,nonlinear,multiple constraints and other complex problems.Experiments show that the improved sparrow search algorithm and whale optimization algorithm in this thesis have improved convergence accuracy and optimization speed.Details are as follows:(1)In view of the overfitting problem of EfficientNet,this thesis proposes a sparrow search optimization algorithm based on rotation control strategy to optimize its structure so that the network structure is more reasonable.Firstly,the initial population with uniform distribution is generated by Halton sequence,and then the population quality is improved by Tent mapping.At the same time,the inertia weight factor is designed to accelerate the algorithm optimization by adjusting the position of the discover.Secondly,on the basis of Levy flight,a random migration mechanism is proposed to update the position of the scouts and improve the local escape ability of the algorithm.Finally,a self-warning simulation mechanism is designed to ensure the optimization ability.The experiment shows that this algorithm has significantly improved the optimization accuracy and confidence.On this basis,the algorithm is embedded into EfficientNet network to classify ceramic fragments,and the classification accuracy rate is 95.57%.(2)Aiming at the problem of complex branches of unsupervised deep learning network,this thesis proposes an intuitionistic fuzzy entropy whale algorithm with multi-stage collaborative optimization to adjust network factors and improve the success rate of splicing.Firstly,the Sobol sequence and the designed dynamic random opposition learning strategy are used to achieve the dual optimization of population quality.Secondly,the composite nonlinear adaptive inertia weight is constructed,and the sinusoidal intuitionistic fuzzy entropy with periodic characteristics is designed for global optimization.The Beta function is used as a disturbance factor to explore the possible global optimal solution region.At the same time,the nonlinear adjustment factor is designed to conduct local fine search for the potential global optimal solution region;since Cauchy mutation strategy can enhance the reproduction ability of surviving excellent individuals after disorderly competition,it is combined with the algorithm in this thesis to accelerate the ability to capture the global optimal value and the convergence accuracy in multi-stage iterative iteration.The results show that the algorithm can speed up the location of the optimal solution and apply it to the image stitching task.The unsupervised deep learning network image stitching model based on whale optimization algorithm established in this thesis effectively improves the success rate of stitching.(3)A cultural relic fragment splicing system based on the swarm intelligence optimization algorithm is realized.The system can realize the functions of fragment classification and splicing,and provide theoretical and technical support for the virtual restoration of cultural relics,so as to improve the splicing efficiency of damaged cultural relics.