側掃聲納回散射訊號之海床地貌影像分析研究

The Study of Side-Scan Sonar Image for Sea Bed Propreties

陳昭儒(Jhao-Ru Chen)

指導教授 ： 宋國士

國立臺灣大學/理學院/海洋研究所/碩士(2009年)

https://doi.org/10.6342/NTU.2009.00744

摘要

本研究則利用側掃聲納之訊號，將訊號以影像方式呈現，並用幾種數位影像處理方式去分辨不同的地貌影像。資料來源是使用2007年6月與10月執行基隆港疏浚工程浚渫物海洋棄置計畫側掃聲納影像。判斷地貌影像方法主要有兩種：小波特徵值、灰度共現矩陣。小波特徵值是將影像訊號利用小波轉換分成七個頻段，計算七個頻段之統計值，根據統計曲線去判斷不同地貌影像。灰度共現矩陣是以參考像元為中心，針對各方向之鄰近像元間所產生的灰階組合，組成灰度共現矩陣，再以統計方式建立該影像的特徵，以描敘影像灰度分佈情形。最後以自動辨認方式取代人工的影像紋理辨認，對特定底質進行辨認。地貌影像辨認結果顯示影像的小波特徵質之頻段特徵，以及灰度共現矩陣之商與同質性分佈圖都能辨認地貌影像之不同；在自動辨認的結果上，都能找出特定底質的地點，並且對資料處理以及資料品質引起辨認誤差之討論。

關鍵字

側掃聲納影像處理 ； 自動辨認 ； 小波特徵值 ； 灰度共現矩陣

並列摘要

In this study, side-scan signals were displayed as sonograph, and different morphological images were classified using digital image processing methods. The data of side-scan sonar image was collected in the project of “Disposal plan of the Keelung Harbor dredging material” in June and October 2007, respectively. Two kinds of method to analysis image are: Wavelet Packet Feature (WPF) and Gray Level Co-occurrence Matrix (GLCM). WPF constructs the signals into seven bands to calculate their respective statistics, and each morphological feature poses their represented trend of statistics among the bands; GLCM produces 16 by 16 grayscale matrix based on the grayscale value which is derived from the pairs of each of 8 peripheral elements matching with the central pixel, respectively, in a sequence of 3 by 3 matrix. GLCM matrix is to establish the characteristics or distribution for each specified image. In the follow, this study also tries to recognize texture composed by side-scan image using automatically identification method instead of manually approach, especially to identify some specific sediment properties. Results of morphological recognition using side-scan images show WPF distribution among the bands and GLCM entropy and homogeneity patterns provide successful differentiation on different morphological images. In addition, for automatic identification, this method can identify their respective locations for those specific sediment properties. In the final, the study brings up some discussions regarding to error judgment to the images resulted from data processing and data quality.

並列關鍵字

Side-Scan Sonar Image Processing ； Automatic Identification ； Wavelet Packet Feature ； Gray Level Co-occurrence Matrix.

參考文獻

劉佩琨, 2007, 側掃聲納數位影像之解析與辨識分析研究, 台灣大學海洋研究所博士論文.

Beyer A., Chakraborty B., and Schenke H.W., 2007, Seafloor classification of the mound and channel provinces of the Porcupine Seabight: an application of the multibeam angular backscatter data, International Journal of Earth Sciences, v. 96, p. 11-20.

Blondel P., 2000, Automatic mine detection by textural analysis of COTS sidescan sonar imagery, International Journal of Remote Sensing, v. 21, p. 3115-3128.

Blondel P., Sempere J.C., and Robigou V., 1993, Textural Analysis and Structure-hacking for geological mapping: Applications to sonar images from Endeavour Segment, Juan de Fuca Ridge, IEEE Xplore, v. 3, p. 203-213.

Chang T., and Kuo C.C.J., 1993, Texture analysis and classification with tree-structured wavelet transform, IEEE transactions on image processing, v. 2, p. 429-441.