基于BP神经网络的海底地形复杂度自动分类方法研究

Study on Automatic Classification Method for Seafloor Terrain Complexity Based on BP Neural Network

针对海底地形复杂程度分类问题,在考虑传统水深均值的基础上引入坡度和起伏度两个地形因子作为表征海底地形复杂程度的分类指标并进行量化,对水深数据空间分辨率进行统一,建立包含18种典型海底特征的海底地形复杂度分类库,利用BP神经网络对建立的分类库进行训练学习。为验证该方法的有效性和适用性,选取地形复杂度不同的4块实验区分别采用统计学方法和BP神经网络算法进行海底地形复杂度进行分类,对比发现该方法可以实现海区海底平坦、一般、复杂三种地形的自动识别与分类,并保留实验区海底地形复杂度细节信息。

For the classification of seafloor terrain complexity, the slope and the relief degree, in addition to the mean water depth, are also introduced as the classification in- dexes for characterizing the seafloor terrain complexity and quantified. And the spatial resolution of water depth data is unified. Based on these, a seafloor terrain complexity classification library which includes 18 types of typical submarine features is established and trained by using BP neural network. For testing the validity and applicability of this method, 4 experimental areas with different seafloor terrain complexity are chosen and statistics method and BP nerve network algorithm are respectively applied for the classi- fication of seafloor terrain complexity. It is found by the comparison that by using the proposed method three types of seafloor terrain, i.e. flat seabed, general seabed and complex seabed, can be identified accurately, rapidly and automatically, and the details of the seafloor terrain complexity in the experimental areas can also be well preserved.