油漆调色系统的改进算法

On Improving Learning Speed and Learning Precision of BP Algorithm for Oil Paint Color Design

根据现有 CAID系统中色彩设计和色彩生产难以协调统一的问题以及油漆厂的实际需求 ,在原有的“基于神经网络的油漆调色系统”基础上 ,创建了 BP网络样本数据库 ,找到了一种适用于实际应用的有效的 BP改进算法。研究测试的结果表明 ,BP改进算法能够弥补原有算法精度速度不够理想的不足 。

The BP (back propagation) neural network algorithm we developed in 1999 for oil paint color design was deficient in learning speed and learning precision. Subsection 1.2 explains that our new BP neural network alogrithm is different from the old one in two respects: (1) the step in our new algorithm is variable instead of the constant step used in the old algorithm; (2) the sample databases in the new and old algorithms are different. Table 1 shows that learning speed of new algorithm is much higher than that of the old one. Fig.2 shows the learning curve for the old algorithm, and Figs.3 and 4 show the learning processes of the new algorithm. Figs.2, 3, and 4 show that the learning precision of the new algorithm is higher than that of the old one.