Abstract: According to the problem of intelligent decision-making for design scheme image based on deep learning algorithm under the condition of the small sample data set,a product modality design decision model based on Capsule Network(CapsNet)was proposed.A multi-angle image data set based on product morphology semantics was built,and it was pre-processed and feature extracted using artificial intelligence method.The image feature characteristics were obtained by using convolutional layer,and several image features from different convolutional layers were divided into a group in order to generate a master capsule with rich semantic features.A set of digital capsules was obtained and the capsule network model was set up utilizing a dynamic routing algorithm.The performance of the model was improved by training the data set in order to improve the decisions accuracy of the product form design.A small sample data set of the intelligent escort robot was constructed,and the effectiveness and feasibility of the model established were verified.

Key words: deep learning, capsule network, design decision, digital capsule, robot morphological characteristics, convolutional neural network

摘要： 针对小样本数据集条件下深度学习算法的方案图像智能决策限制问题,提出一种基于胶囊网络的产品形态设计决策模型。应用人工智能方法搭建基于产品形态语义的多视角图像数据集,并将数据集图像进行预处理和特征提取。再利用卷积层学习得到图像特征,将不同卷积层中的若干特征划分为一组,生成具有丰富语义特征的主胶囊。利用动态路由算法获取一组数字胶囊,完成整个胶囊网络模型。最后,通过对训练数据集提高模型识别性能,从而提高产品形态设计决策准确率。通过构建智能陪护机器人的小样本实例数据集,验证了所建立模型的有效性与可行性。

关键词: 深度学习, 胶囊网络, 设计决策, 数字胶囊, 机器人形态特征, 卷积神经网络