Abstract:

At present, the determination of segmentation scale in the object-oriented seafloor acoustic image classification is empirical and significantly influenced by human factors. A spatially adaptive segmentation scale determination method using the confusion index as an objective index is proposed. The mean value and standard deviation of echo intensity corresponding to the segmentation objects are calculated by giving a set of segmentation scales. The unsupervised K-means clustering algorithm is then adopted to calculate the confusion indexes pf classification results at different segmentation scales, and the segmentation scale corresponding to the minimum confusion index is selected as the optimal scale to extract the seafloor image features. Based on the seafloor image features extracted at the optimal scale, a supervised classification model is established by combining the sampled data to predict the distribution of sediments in the whole surveying area. Experimental results prove that the spatially adaptive segmentation scales can be used to improve the classification accuracy significantly. The effectiveness of the proposed method is verified by cross-check in the experiment. Moreover, for thesegments that are with the relatively consistent the echo intensity characteristics, the classification accuracy can be further improved by introducing the terrain features.

Key words: seafloor sediment classification, spatially adaptive, segmentation scale, acoustic image, confusion index