- 學位論文
模糊類神經網路影像校準方法應用在精密樣板比對
A Fuzzy Neural Network Image Alignment Method for Precision Pattern Matching
摘要
近年來自動視覺檢驗技術已廣泛的為工業界所運用，而影像比對技術在許多自動視覺檢驗方法上更扮演一個重要的角色，如零件的外型尺寸檢測、分類或識別等品質的檢驗及控管，而應用於線上即時檢驗，速度及精度均有一定的要求。傳統的影像比對技術對於比對的過程經常需耗費大量的運算時間並產生較大的誤差，尤其對於較大的影像檢測範圍。本研究對精確及快速的樣板比對需求，提出一個新的模糊類神經網路 (Fuzzy Neural Network) 校準方法，可達到快速並精確的樣板校準目標。所推薦的模糊類神經網路校準系統係利用倒傳遞網路(Back-propagation algorithm) 運作，以樣板與待測影像間的比對推論誤差值與期望誤差值所建立的能量函數，運用模糊演算及監督式的學習法則配合梯度坡降法進行線上學習及調整，使得推論誤差與期望誤差之間愈接近，而達到影像比對校準的目的。在實驗方面，以四個實驗來論證所推薦方法的效能並實際應用在精密的齒輪楊測上。結果顯示，模糊類神經網路校準方法能夠在所要求的時間範圍內達到精確的樣板定位。
並列摘要
Image-matching plays an important role in many general processes of image inspections, such as dimension inspection and/or part classification. For those requiring fast speed and high precision, the step of image-alignment in image-matching often causes most of computation time and key errors, especially when the image space considered is large. A newly-designed fuzzy neural network (FNN) alignment scheme to achieve a fast and precise pattern alignment is proposed in this study to provide a means for favorable and fast inspection. The proposed FNN alignment system utilizes the supervised gradient decent method and the back-propagation algorithm to accomplish the alignment based on the minimization of a prescribed energy function that is defined by the matching error between the feature points of the pattern and the test image. Four experiments are conducted to validate the effectiveness of the proposed method and applied to inspect precision gears practically. The results show that the FNN alignment method is capable of performing precision positioning within a required time span.