An image classification algorithm for plantar pressure based on convolutional neural network

Human structure-based plantar pressure (PP) analysis has been widely used in medical, sports, footwear design, and footwear sales. The current studies mostly focus on the development of PP measuring technologies and the analysis of pressure distribution features based on sensing results. Relatively few scholars have tried to analyze PP through image processing. To bridge the gap, this paper tries to classify PP images based on convolutional neural network (CNN). Firstly, the authors prepared the zoning and center calculation for PP images, and established a PP image classification model. Then, the PP image features were selected dynamically based on sparse, low-redundancy feature subsets, and the results of principal component analysis (PCA) were combined with the CNN to realize dynamic extraction of features from PP images. Finally, an image classification algorithm was designed based on the inter-area difference in PP distribution. The proposed algorithm was proved feasible through experiments. The research findings provide a reference for processing pressure images in other scenarios.