Segmentation Methods Of Plantar Pressure Imaging On Shoe Last Optimization Design

The optimized design of the shoe last is an essential part of the shoe production process,and the optimization of the curved shape of the shoe last has an important influence on the comfort of the finished shoe.In the design of the shoe last,through the analysis and processing of the plantar pressure data,the shape,force,and movement trajectory of the foot when the human body is standing or walking can be obtained,and an ergonomic design can be designed accordingly.In plantar pressure analysis,using the plantar pressure imaging analysis method to realize the optimization design of shoe last is still relatively preliminary.The study and utilization of imaging data still lack processing methods,incomplete information acquisition,and poor robustness of the processing model.The research work of the thesis is mainly to research the extraction method of key regions related to the last of comfortable shoes for the image generated by the data of plantar pressure;using intelligent information processing methods to study the more accurate segmentation model;and the segmentation efficiency evaluation method can obtain accurate area division,extract the features required for the optimal design of the shoe last,increase segmenting effect and efficiency of plantar pressure image,and provide data services for the optimal design of the shoe last.The specific research content of the dissertation mainly includes:(1)Based on the analysis of plantar pressure image processing requirements,the plantar pressure imaging experimental equipment is used to collect the data from plantar pressure sensors;completed the test images to achieve the acquisition of plantar pressure imaging data;used denoising and dimensionality reduction to construct plantar pressure imaging the basic data set for the key characteristics of plantar pressure;provided a fundamental data-set for the segmentation of feature regions and the extraction of key features.(2)Aiming at the influence of selecting the segmentation area on the segmentation effect in the plantar pressure image segmentation issues,a mean shift segmentation model of plantar pressure image based on local preservation morphology is proposed.In this model,the image is segmented from a global perspective.The mean shift algorithm’s real-time feature is used to realize image filtering and dimensionality reduction,which reduces the computational complexity.The image is segmented into several local maximum uniform regions by using domain pixel preservation morphology technology,and the segmentation template of the initial target is extracted by morphological reconstruction.The histogram analysis method is used to validate the efficiency of image segmentation.The comparison results of segmentation evaluation indicators show that the model has strong robustness and has a good effect on pixel accuracy,frequency-weight intersection ratio,over-segmentation rate,and DICE coefficient.(3)Aiming at the problem of over-segmentation of traditional image segmentation algorithms,such as region growing based,level set based segmentation,and semantic segmentation,a deep learning network-based method seg Net combing with full connection layer network for plantar pressure image segmentation is proposed.The pooled index calculated in the maximum pooling step to perform the non-linear up-sampling he is used,and the integrity of high-frequency detail segmentation is also maintained.The validity of the semantic segmentation algorithm of the proposed method is performed in much higher effectiveness by comparing other general platforms through indices of global accuracy,average accuracy,and mean intersection over union,etc.(4)Full convolutional network is used to reduce the dimensionality of plantar pressure imaging data through the deep neural network platform FCN-Alex Net-8s,and a pre-processing and segmentation method for the plantar pressure imaging dataset is developed.The proposed method is superior to previous studies in mean square error and peak signal noisy rate.Besides,comparisons are made with a region-based convolutional neural network and a fast region-based convolutional neural network in terms of layer index,maximum step,and time consumption.(5)Based on the above research,a non-uniform rational B-spline surface is generated using the key point set of the image segmentation area as the support point and boundary point to solve the key problem of generating from the discrete point set to the bottom surface;Define a calculation method for the running-in degree of control points and fitting points,and evaluate the point data sets of different data sources with the artificial psychology scale method,to realize the effect evaluation of the data sets generated by different mining methods on the application of optimal last design.