基于卷积神经网络的脚部关键参数计算方法

针对传统制鞋业定制化程度低,无法适应足部多样性、舒适性,提出了一种基于卷积神经网络的脚型关键参数计算方法。首先对图像进行透视变换等预处理,然后使用fine-tune的迁移学习方法,通过修改VGG神经网络源模型全连接分类层,将高层卷积权重进行微调,优化网络模型并提取特征值进行特征分类,从图像中识别出脚的轮廓。最后通过设计的算法把脚型特征值计算出,并与实际测量的脚长、腰窝宽度、脚宽等做对比。实验表明,改进后的模型对脚部识别的准确率达到96.8%,输出结果与测量的真实数据相比误差不超过3%,可作为鞋底制作的重要依据。

Calculation Method of Key Parameters of Foot Based on Convolutional Neural Network

Aiming at the low degree of customization of the traditional footwear industry and the inability to adapt to the diversity and comfort of the foot, a calculation method for the key parameters of the foot based on convolutional neural network is proposed. Firstly, the image is subjected to pre-processing such as perspective transformation. Then, using the migration-learning method of fine-tune, the VGG neural network source model is fully connected to the classification layer, the high-level convolution weights are fine-tuned, the network model is optimized, and the eigenvalues are extracted for feature classification. , the outline of the foot is identified from the image. Finally, the foot shape feature value is calculated by the designed algorithm, and compared with the actual measured foot length, waist width, and foot width. Experiments show that the improved model has an accuracy of 96.8% for foot recognition, and the output is less than 3% compared with the measured real data, which can be used as an important basis for shoe sole production.