目的 基于改进遗传算法优化 BP 神经网络, 研究对显示器色彩空间转换的预测准确性的方法。 方法 通过改进数据归一化处理和遗传算法的适应度函数,来优化 BP 神经网络的权值和阈值,以将它们的分布范围缩小,再用 BP 算法进行精确求解,并将该方式与常规方式作对比。 结果 训练优化后的 BP 神经网络预测模型 20 次,测试色块平均色差为 2 . 9353 ,最小平均色差为 1 . 9467 。 结论 该方法大大降低了 BP 神经网络预测模型陷入局部极小值的可能性,对显示器色彩空间转换具有较好的非线性拟合能力和更高的预测准确性。

Objective To study the method for improving forecasting accuracy of monitor color space conversion based on genetic algorithm to optimize the BP neural network. Methods The weights and threshold of the BP neural network were optimized mainly through the improvement of data normalization and the fitness function of genetic algorithm, to narrow down their distribution range, and then BP algorithm was solved exactly. This mode was then compared to the normal mode. Results After 20 times training of the optimized BP neural network prediction model, the average color difference in test blocks reached 2. 9353, and the minimal average color difference reached 1. 9467. Conclusion The results showed that the method can greatly reduce the possibility of falling into local minima of the BP neural network prediction model, with good non-linear fitting capability and higher prediction accuracy of color space conversion.