论文部分内容阅读
目的:探讨基于BP神经网络预测瓷粉配方的配色网络模型的可行性。方法:按照不同质量配比的混合松风Halo瓷粉制作32个瓷片样本,随机分为两组:训练组和测试组。用Matlab软件构建BP神经网络模型,用训练组瓷片训练并优化网络,形成颜色L*a*b*参数与瓷粉配方质量配比之间的非线性映射关系。用测试组试样检验构建的神经网络的精度。结果:测试组预测出的混合瓷粉质量配比与真实质量配比之间的准确率为80%(配比绝对误差±0.05)。按照测试组的预测和真实瓷粉质量配比制作的瓷片之间的平均色差为1.68(远小于临床可容忍阈值2.7)。结论:基于BP神经网络模型的瓷粉配色方法,直接给出了瓷粉配方,缩减了操作程序,减小色差,为当前口腔临床修复技术提供了一种新的比色途径。
Objective: To explore the feasibility of color matching network model based on BP neural network to predict porcelain powder formula. Methods: 32 samples of porcelain were made according to the different mixing ratio of pine-pozzolanic Halo porcelain powder, and were randomly divided into two groups: training group and test group. The BP neural network model was constructed by using Matlab software, and the training and optimization of the network was carried out by using the training set tiles to form the nonlinear mapping relationship between the color L * a * b * parameters and the porcelain powder recipe quality ratio. The test group test samples to verify the accuracy of the constructed neural network. Results: The accuracy of the ratio of the mass of the mixed porcelain powder to the real mass ratio predicted by the test group was 80% (the absolute error of the ratio was ± 0.05). The average color difference between tiles made according to the test group’s prediction and the real porcelain powder mass ratio was 1.68 (far less than the clinical tolerable threshold of 2.7). Conclusion: The porcelain powder color matching method based on BP neural network model directly gives the recipe of porcelain powder, reduces the operation procedure and reduces the chromatic aberration, and provides a new colorimetric approach for the current oral clinical repair technique.