Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/20197
Title: | 卷積神經網路應用於鞋底磨損影像分析 Convolution Neural Network Applied to Sole Abrasion Image Analysis |
Authors: | Fu-Sheng Jiang 蔣富昇 |
Advisor: | 郭柏齡 |
Keyword: | 卷積神經網路,鞋底磨損影像,田口方法,人工智慧,鞋墊, Convolution Neural network,Sole Abrasion image,Taguchi Method,Artificial intelligence,Insoles, |
Publication Year : | 2018 |
Degree: | 碩士 |
Abstract: | 足部不適為常見病症,而近年來,功能性鞋墊的興起使國人對足部健康意識感到重視。為避免足部不適進一步影響到膝蓋偏斜、甚至脊椎側彎等問題,需做提早預防性治療。現行挑選功能性鞋墊做法,由復健治療師審視足部健康狀況後提供選擇建議為主。另外常用做法為使用壓力測試墊片或是壓力量測儀器,進行足壓分析後給出相應的足墊選擇建議。本研究以鞋底磨損資訊取代一次性壓力測試墊片或是由價格高昂的壓力量測儀器所量測的數據。在經濟與時間的考量下,鞋底磨損的資訊較一次性量測資訊,可提供更多關於使用者長時間運動資訊,較符合使用者實際狀況。另外鞋底磨損也帶有足部功能性或結構性問題,以及使用者的運動習慣資訊,值得進一步研究。
本研究收集鞋底磨損影像,並以卷積神經網路來分類鞋底磨損程度。經往復訓練得到與人類專家判定一致性結果,訓練機器成為挑選合適鞋墊的好幫手,來解決不同人類專家可能導致判讀上不一致的問題,並節省高昂的量測費用。正確選擇鞋墊參數,可提供使用者更舒適足部體驗。本研究使用了57 筆鞋底影像資料,整理收集鞋底磨損影像並處理為人工智慧使用之資料集,以供卷積神經網路分類訓練。 研究顯示,以人類專家判定結果為標準答案,使用卷積神經網路自行取出的特徵做分類結果,可達到準確率80%以上 ,可有效輔助人類專家的判斷,以找出相匹配的功能性鞋墊。 Foot discomfort is a common disease. In recent years the functional insoles make people feel the importance of foot health. In order to avoid foot discomfort and further affect the knee deflection and even scoliosis, it’s need to do early preventive treatment. The current selection of functional insoles which is reviewed foot health status by the rehabilitation therapist. In addition there are commonly used pressure test pad or pressure gauge, foot pressure analysis gives the corresponding choice of functional insoles. This study replaced one-time pressure test pads with shoe-wear information or measured data from costly pressure gauges. Under the consideration of economy and time the information of the sole abrasion provides more information about the user's exercise. The sole abrasion is a longer period of time than the one-time measurement. It is more fit the user actual situation. In addition to the soles abrasion with functional or structural foot problems and the user's exercise habits information, it is worth further study. In this study the images of abrasion were collected and the degree of sole abrasion was classified by convolution neural networks. The artificial intelligence and human experts determine the consistency of the results. The machine becomes a good helper for choosing the right insole. To solve the differences between different human experts are saving expensive measurement costs. It provides users with a more comfortable foot experience. In this study, 57 pieces of image data were used to complete the image acquisition of soles abrasion and the processing of artificial intelligence data set, which was used to classify the training of convolution neural network. The research shows that taking human experts' judgment as the standard answer. The feature extracted by convolution neural network as the classification result can achieve an accuracy of 80%, It can help human to judge the matching insoles. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/20197 |
DOI: | 10.6342/NTU201800327 |
Fulltext Rights: | 未授權 |
Appears in Collections: | 生醫電子與資訊學研究所 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
ntu-107-1.pdf Restricted Access | 2.82 MB | Adobe PDF |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.