Abstract: An image acquisition system based on binocular visual sensors for robot mobile platforms was established, and an improved grey wolf algorithm combined with a minimization parameter strategy was studied to optimize support vector machines and achieve recognition of different weld types. Firstly, introducing the theory of the best point set into the Grey Wolf algorithm to generate an initial population, reducing the number of species in the Grey Wolf population and laying the foundation for the fast and stable global search of the algorithm. Then, a nonlinear convergence factor was introduced into the classifier SVM and combined with a strategy of minimizing parameters to enhance the generalization ability of the optimal parameters. Finally, an SVM model based on the optimal parameters was used for weld seam type recognition experiments. It is proved that the improved algorithm optimized SVM model has significantly improved recognition accuracy and optimization speed compared to particle swarm optimization, genetic algorithm, cuckoo bird algorithm, and basic grey wolf algorithm.