黃文吉Wen-Jyi Hwang范哲誠Zhe-Cheng Fan2019-09-052015-08-062019-09-052012http://etds.lib.ntnu.edu.tw/cgi-bin/gs32/gsweb.cgi?o=dstdcdr&s=id=%22GN0699470137%22.&%22.id.&http://rportal.lib.ntnu.edu.tw:80/handle/20.500.12235/106892本論文提出以Recursive Least Mean Square為基礎，結合Fuzzy c-Means分群演算法實作出Radial Basis Function類神經網路之紋理圖辨識系統。在本論文中，Fuzzy c-Means計算紋理圖的質量中心點，Recursive Least Mean Square計算類神經網中的權重係數，希望利用硬體的特性來實現快速運算、低資源消耗、低功率消耗以及擁有良好的效能之硬體架構。 最後我們所提出的硬體架構會在以FPGA為基礎的可程式化系統晶片設計(System On a Programmable Chip，SOPC)之平台上作實際的效能測試。根據使用不同的紋理圖作為測試資料，實驗結果顯示本架構對於紋理圖辨識有良好的分類正確率，且此硬體架構提供了日後高度的延伸性。This paper presents a real time RBF training hardware architecture for texture recognition which is based on recursive least mean square method and fuzzy c-means algorithm. We use fuzzy c-means algorithm to calculate centers in the hidden layer and use recursive least mean square method to estimate connecting weights in the output layer. Experimental results show that the proposed architecture is a effective hardware for real time training with low computational cost, low power consumption and high performance.可程式化系統晶片資料分群FCM演算法Recursive Least Mean Square紋理圖辨識系統程式晶片設計FPGAdata clusteringFCM algorithmRecursive Least Mean Squaretexture recognitionsystem on programmable chip