基于BP神经网络的印刷色彩质量综合评价  

Printing Color Quality Comprehensive Evaluation Based on BP-ANN Method

作　　者：初红艳[1] 秦合营[1] 蔡力钢[1] 李鹏[1] 李风光[1] 

机构地区：[1]北京工业大学北京市先进制造技术重点实验室,北京100124

出　　处：《包装工程》2009年第10期97-100,共4页Packaging Engineering

基　　金：国家十一五科技支撑计划课题(2006BAF03B01);北京市教委科研计划项目(KM200910005006);北京工业大学青年科研基金(97001011200701)

摘　　要：应用BP人工神经网络模型对印刷色彩质量进行评价。对BP网络模型进行了改进:对网络模型的专家样本进行改进,即在国家对印刷色彩质量相关规定的基础上,通过大量的试验和分析构造出训练样本,从而提高了模型的鲁棒性和识别的准确性;采用基于黄金分割理论的优化算法对BP网络模型的隐含层节点数进行优选,使模型在进行较少次的循环迭代后训练误差收敛到最小,从而提高网络模型的性能。实例表明,此印刷色彩质量评价方法是切实可行的,评价结论具有客观性和实用性,能够细致划分印刷品等级。Back propagation artificial neural networks （BP ANN for short） model was applied to evaluate printing color quality. The model was improved in two aspects. Firstly, the expert samples of BP ANN model were extended to improve model＇s robustness and veracity of distinguishing. That is, created more training samples based on color standards instituted by government and lots of experiments. Secondly, the amount of nodes in network＇s covert layer was optimized using an arithmetic based on golden section theory to make training error of the model minimized after less times cyclic iteration, and to improve the performance of BP--ANN model. As an example, several printings were classified as four grades, very good, good, moderate and bad. The study showed that the method of BP network in printing color quality evaluation was feasible and reasonable, and the result of evaluation was objective and practical. The grades of printings can be classified finely.

关 键 词：印刷 色彩评价 BP神经网络 黄金分割 

分 类 号：TS801.3[轻工技术与工程] TS807