In the traditional focimeter measurement, due to the change of the curvature radius of the rear surface of the lens to be measured, the rear vertex of the lens and the diaphragm surface cannot be completely coincident, resulting in the situation that the power measurement error increases with the increase of the lens power. Aiming at the insurmountable measurement error in traditional measurement, a method of extracting different features from lens images with different powers is proposed, and the feature data sets of lenses with different powers are obtained. k-Clustering algorithm (KNN algorithm) for classification. Select single focus lenses -20 ~+20m-1 with a refractive index of 1.551 every 1m-1, and -15~+8m-1 with a refractive index of 1.56 and 1.60 every 0.5m-1 for lens image acquisition. The experimental results show that the classification Precision, Recall rate and F1 score of this method are all 100%, the three models can correctly identify all the corresponding test lenses and overcome this measurement error.