Abstract: The automatic classification and recognition of conical yarn paper tubes has been a hot topic in the intelligent manufacturing of this component. A yarn paper tube classification method based on multiple fusion optimized template matching was proposed to address the problems that traditional image classification methods could not balance speed and accuracy, as well as high cost of deep learning, difficult deployment, and high hardware requirements. Several improved algorithms and strategies were adopted and three times data dimensionality reduction was used to accelerate the template matching speed. First, the optimization algorithm, successive elimination algorithm (SEA) used for motion estimation was used in template matching, and the threshold of this algorithm was improved to adaptive threshold for enhancing the robustness of the algorithm. Then, the wavelet pyramid was used to further reduce the amount of operations to improve its speed. Finally, the cross gray scale feature was used instead of the traditional sum of absolute differences (SAD) algorithm to calculate the performance index, and the strategy of stopping the iterative search in advance was used to further filter the data and set the cumulative error threshold to stop the search in advance. The matching experiments show that the improved algorithm guarantees the accuracy and the matching speed reaches about 0.126 s. The comparison and ablation experiments show that, under the premise of ensuring the accuracy, the speed of the algorithm is nearly 11 times higher than that of the traditional SAD algorithm, compared with some other classical methods in the speed are also improved, which verifies the effectiveness of the method.