基于BP神经网络的化妆品色彩配方设计

doi:10.3969/j.issn.1001-1803.2020.07.006

摘要/Abstract

摘要：

为探讨基于BP神经网络模型预测化妆品色彩配方的可行性,以口红为研究对象,按不同比例色素制备100个样本。采用节点数均为15个的2层隐含层、1层输出层的3层网络结构,通过Matlab R2016a软件构建BP神经网络模型,形成颜色RGB参数与口红色素质量配比间的非线性映射关系。在训练次数为10 000次,学习率为0.5时,各色素误差参数均小于0.6,预测配方成品与真实配方成品无明显色差。基于BP神经网络模型的化妆品配色方法,可以直接给出色彩配方,为配色工程师提供了一种快速简单的参考工具。

关键词: BP神经网络, 配色, 色彩预测, Matlab, 口红

Abstract:

To explore the feasibility of prediction of color formulas of cosmetics based on BP neural network model, one hundred lipstick samples of different proportions of pigments were prepared. The BP neural network model was constructed by MATLAB R2016a software, which was composed of two hidden layers with 15 nodes and one output layer with three network structures. The nonlinear mapping relationship between the RGB parameters and the mass ratio of lipstick pigments were formed. When the training times were 10 000 and the learning rate was 0.5, the color error parameters were all less than 0.6, and there was no significant color difference between the predicted formula and the real formula. Hence the color matching method based on BP neural network model could directly generate the color formula and provide a quick and simple reference tool for color engineers.

Key words: BP neural network, color matching, color prediction, Matlab, lipstick

中图分类号:

TQ658

吴少娟,郭清泉. 基于BP神经网络的化妆品色彩配方设计[J]. 日用化学工业, 2020, 50(7): 464-469.

WU Shao-juan,GUO Qing-quan. Color formulation design of cosmetics based on BP neural network[J]. China Surfactant Detergent & Cosmetics, 2020, 50(7): 464-469.

图/表 12

图 1

表 1

表 2

图 2

图 3

表 3

表 4

图 4

表 5

图 5

表 6

图 6

参考文献 20

[1]	Li Chaojing, Tang Youchun, Huang Xia. A survey of the application of BP neural network[J]. Labor Security World, 2012,34(8) : 71-74.
[2]	Yi Lijuan, Li Ya, Zou Sulan, et al. Optimization of water extraction technology of Yiqi Huoxue prescription based on orthogonal test design and BP neural network[J]. Chinese Traditional and Herbal Drugs, 2019,50(18) : 4305-4306.
[3]	Feng Shaojun, Li Chunli, Ren Zhenzhong. Application of artificial neural network model in concrete field[J]. China Southern Agricultural Machinery, 2019,50(9) : 206-207.
[4]	Yasser Vasseghian, Negin Heidari, Mojtaba Ahmadi, et al. Simultaneous ash and sulfur removal from bitumen: Experiments and neural network modeling[J]. Fuel Processing Technology, 2014,125:79-85.
[5]	Jun Cheng, Xin Wang, Si Tingting, et al. Ignition temperature and activation energy of power coal blends predicted with back-propagation neural network models[J]. Fuel, 2016,173:230-238.
[6]	Zhang Defeng. Application design of MATLAB neural network [M]. Beijing: China Machine Press, 2012: 30- 38, 136-158.
[7]	Tian Limei, Zhang Qiang, Shi Jianquan, et al. Design and optimization of cigarette fragrance formula based on BP neural network[J]. Journal of Yunnan University, 2010,32(S1) : 76-80.
[8]	Wang Ying. Research on color matching algorithm of fabric dyeing based on improved BP neural network[D]. Shandong: Qingdao University, 2008.
[9]	Huang Ye. Color prediction of original liquid colored cellulose yarn based on GA-BP neural network[D]. Tianjin: Tianjin Polytechnic University, 2019.
[10]	Fu Li. Algorithm improvement and function expansion of paint color matching system[D]. Shanxi: Northwestern Polytechnical University, 2002.
[11]	Xie Lamei. Formula and preparation of a moisturizing lipstick: 109846738A [P]. 2019 -06-07.
[12]	Lin Xihui. A formula of moisturizing lipstick: 109260042A [P]. 2019 -01-25.
[13]	Ma Hongxia, Wang Yu, Chen Chao, et al. Preparation of purple grass moisturizing color lipstick[J]. Guangdong Chemical Industry, 2019,46(9) : 126-127.
[14]	Strahl André, Gerlich Christian, Alpers Georg W, et al. An instrument for quality assurance in work capacity evaluation: development, evaluation and inter-rater reliability[J]. BMC Health Services Research, 2019,19(1) : 1-11. doi: 10.1186/s12913-018-3827-x pmid: 30606168
[15]	Zhong Wen, Deng Yunchuan, Tenreiro Machado J A, et al. Strength prediction of similar materials to ionic rare earth ores based on orthogonal test and back propagation neural network[J]. Soft Computing, 2019,23(19) : 9429-9437. doi: 10.1007/s00500-019-03833-7
[16]	Luo Li. Matlab realization of BP neural network digital recognition[J] . Electronic Technology & Software Engineering, 2019,20:174-175.
[17]	Wang Peng, Wei Jiaqiang, Li Qing, et al. Color matching of porcelain powder mixture based on BP neural network[J]. Medical Journal of Wuhan University, 2014,35(3) : 386-389.
[18]	Yao Litong, Zhong Yifan, Wu Jingyang, et al. Multivariable logistic regression and back propagation artificial neural network to predict diabetic retinopathy[J]. Diabetes, Metabolic Syndrome and Obesity: Targets and Therapy, 2019,12(12) : 1943-1951. doi: 10.2147/DMSO
[19]	Kazuhiro Nakano. Application of neural networks to the color grading of apples[J]. Computers and Electronics in Agriculture, 1997,18(2) : 105-116.
[20]	Cao Congjun, Sun Jing. Study on color space conversion between CMYK and CIE Lab based on generalized regression neural network [C]// IEEE Computer Society. 2008 International Conference on Computer Science and Software Engineering. Hubei: IEEE, 2008: 275-277.

组分	w/%
组分	样品1	样品2	样品3	样品4
蓖麻油	53	58	61	63
角鲨烷	8	9	9	8
巴西棕榈蜡	12	9	6	6
白蜂蜡	11	10	11	10
小烛树蜡	8	6	5	5
维生素E乙酸酯	0.9	0.9	0.9	0.9
CI77202	7	7	7	7
香精	0.10	0.10	0.10	0.10

评分标准	质地	硬度	滋润度	气味	色泽
很好（8~10分）	表面光滑,无气孔,膏体质地细腻,无颗粒	不易弯曲折断,不过软且不过硬	涂后不干燥,无结块	无刺鼻香精味、石油味或厚重的蜡味	容易上色,颜色均匀自然
较好（5~7分）	表面光滑,基本无气孔,无肉眼可见颗粒	不易弯曲折断,有点软/硬	涂后4 h内不干燥,无结块	基本无刺鼻香精味、石油味或厚重的蜡味	比较好上色,颜色较均匀自然
一般（2~4分）	有少量气孔,有少许颗粒	较易弯曲折断,比较软/硬	涂后4 h内干燥,有轻微结块	含有轻微的香精味、石油味或厚重的蜡味	不易上色,颜色基本均匀
较差（0~1分）	含有较多气孔,颗粒感明显	轻易弯曲折叠,非常软/硬	涂后干燥,有明显结块	有比较重的香精味、石油味或厚重的蜡味	难上色,颜色不均匀

项目	参数
个案数	10
Kendall’s W系数	0.912
卡方	27.371
自由度	2
渐进显著性	0

编号	w/%					R	G	B
编号	色素1	色素2	色素3	色素4	色素5	R	G	B
1	3.00	1.00	0	3.00	0	187	185	93
2	0	0	3.00	4.00	0	132	49	49
3	0	1.00	0	6.00	0	153	168	163
4	2.00	0	0	5.00	0	170	170	101
5	0.50	0	1.30	5.10	0.10	96	39	45
6	0	0.50	0	5.00	1.50	71	80	78
7	0.50	0.50	4.00	2.00	0	65	25	25
8	0	0	6.00	0	1.00	103	42	37
9	4.00	0	0	2.00	1.00	129	127	65
10	0.50	0	1.70	4.80	0	134	51	58
…	…	…	…	…	…	…	…	…
90	0	0.30	1.00	5.00	0.70	102	51	64
91	0	2.00	5.00	0	0	45	24	25
92	1.70	0.10	2.50	2.70	0	112	38	35
93	3.00	0	1.00	2.00	1.00	94	42	36
94	2.50	1.00	0.50	3.00	0	92	67	61
95	2.50	0	0.50	4.00	0	172	95	76
96	2.00	0	3.00	0	2.00	75	35	31
97	0.32	0.80	1.50	4.38	0	99	56	58
98	4.00	0	0	0	3.00	47	49	27
99	0	3.00	1.70	2.30	0	37	28	49
100	1.00	2.00	4.00	0	0	69	56	60

名称	MSE	RMSE	MAE	MBE
色素1	0.15	0.39	0.28	-0.01
色素2	0.03	0.18	0.14	0.06
色素3	0.30	0.55	0.32	-0.11
色素4	0.30	0.55	0.36	0.15
色素5	0.17	0.41	0.30	-0.09