Application of Neural Network and Moment Invariants in Expert System of Pavement Distress Diagnosis

應用神經網路與動差不變量於專家系統之鋪面裂縫診斷

對現代化機場而言，鋪面裂縫診斷對維修方式選擇是極為重要的工作，然而卻只能仰賴人力判斷，本研究的主要目的，是希望透過類型辨識及神經網路技術，進行偵測與分類以提供專家系統之鋪面裂縫診斷，並建構自動化之鋪面管理系統;首先，擷取裂縫的數位影像，並將原彩色影像經影像處理後，轉換成裂縫與非裂縫的二元影像，然後利用傳統的幾何形狀量測及動差不變量理論，將影像加以數值化，以獲取裂縫特徵值，再透過神經網路針對裂縫影像種類進行分類，最後引用鋪面裂縫影像為例，實際將裂縫影像資料進行處理，並以傳統的幾何形狀量測，與加入動差不變量後的分類辨識率結果加以比較與討論，確認該等技術可有效達成自動鋪面裂縫診斷。

Pavement distress diagnosis is an important task for modern airport management, however, the strenuous routine check and diagnosis works are still executed by labor. The main purpose for this paper is to present an automatic expert system to detect and classify the airport pavement distress by using technologies of pattern recognition and neural network to enhance pavement distress diagnosis. First of all, we investigate pavement by digital camera or video to capture the crack images. Second, we use technique of image processing to transfer the original color images into binary images of distress and non-distress. Next, by means of the theories of traditional geometric measurement and moment invariant, we analyze the images to generate characteristic values. Finally, by using neural network algorithm to process the classification of pavement distress images, we took practical pavement distress images for example, and complete processing image data with traditional geometric measurement and moment invariant. The experimental results indicate that the system classification with both geometric measurement and moment invariant provide better accuracy than that of only geometric measurement.