BP神经网络的宇航服多维意象造型评价方法

目标 为适应宇航服造型设计对“大国形象”多维意象识别的需要,提出了以“大国形象”为导向的宇航服造型设计评价模型构建方法。方法 在大量用户调研的基础上,对“大国形象”意象集进行筛选,并采用认知实验与聚类分析相结合的方法获得宇航服产品的代表样品。采用语义差异法,对“大国形象”感性意象集中初步筛选词和宇航服产品代表样本之间的映射数据进行检验和分析,对得到的数据结果进行主成分分析,以获得宇航服产品对“大国形象”感性意象的认知空间。同时,根据全局HIEs解构原则和宇航服产品的功能约束清单建构宇航服产品造型特征空间。采用语义差异法和认知测试,得到宇航服造型多维意象认知空间。利用BP神经网络,以样本关键HIEs评价经数字编码后作为输入层,以各意象词下样本的感性意象均值作为输出层,构建宇航服造型意象评价模型。随后,采用留一交叉训练法对评价模型的准确性进行验证。结论 该评价模型能够有效解决造型特征与多维意象之间的映射及匹配,论证了造型意象和认知空间之间存在的关联性,探索出设计目标和设计意象关联判断的实践方法。

Multi-dimensional Image Styling Evaluation Method of Space Suit Based on BP Neural Network

The work aims to propose an evaluation model construction method for space suit styling design oriented to "Great Power Image", in order to meet the need for multi-dimensional image recognition of "Great Power Image" in space suit styling design. On the basis of a large number of user research, the image set of the "Great Power Image" was screened, and the representative samples of the space suit products were obtained by combining cognitive experiments with cluster analysis. The mapping data between initially screened words in the perceptual image set of "Great Power Image" and representative samples of space suit were tested and analyzed by semantic difference method. Principal component analysis was carried out on the obtained data results in order to obtain the cognitive space of space suit products for the perceptual image of "Great Power Image". At the same time, according to the global HIEs deconstruction principles and the functional constraints of the space service products, the feature space of space suit product styling was constructed. The semantic difference method and cognitive test were adopted to obtain a multi-dimensional image cognitive space of space suits. By BP neural network, the key HIEs evaluation of the samples was digitally encoded as the input layer, and the average perceptual image of the samples under each image word was used as the output layer, to construct the evaluation model of space suit styling image. Finally, the accuracy of the evaluation model was verified by the leave-one-out cross validation. This evaluation model can effectively solve the mapping and matching between styling characteristics and multi-dimensional images, demonstrate the correlation between styling intention and cognitive space, and explore the practical methods of judging the correlation between design goals and design images.