基于特征的模糊神经网络遥感图像目标分类识别

Feature-based fuzzy-neural network approach for target classification and recognition in remote sensing images

特征是图像处理中用于辨识目标的最基本属性。提出了利用模糊神经网络方法,针对舰船的几何特征、矩特征和纹理特征进行舰船目标识别处理。首先简单地描述了几何特征、矩特征尤其是Hu矩特征、一阶纹理特征和二阶纹理特征。然后分别对仿真数据、卫星观测数据中的舰船目标,以及自动检测处理获取的舰船目标的几何特征、Hu机特征和纹理特征进行了提取和分析。模糊神经网络方法可以综合模糊集理论和神经网络方法的优势,有效地实现基于特征的图像目标分类识别处理。文章首先描述了一种主从神经元结构的模糊神经网络分类识别方法,然后利用该方法对大型舰船进行分类识别,包括基于单类舰船特征的分类识别和基于多源(时相)数据融合的分类识别。实验结果表明,基于大型舰船的几何特征、矩特征和纹理特征,利用模糊神经网络方法可以实现对大型舰船目标的有效分类识别。通过多源数据融合处理,可以改善分类识别效果。

Feature is the most essential attribute for recognizing target in image processing. This paper proposes to recognize ship target by utilizing a fuzzy neural network processing on its geometry, moment and texture features. First, we simply depict geometry feature and moment feature especially Hu moment. After that, we respectively extract and analyze geometry, Hu moment and texture feature of ship target in simulated and satellite observed data, as well as ship target acquired by automatic target detection. By analyzing the ship target＇s features, the feature set（ or subset）, comprising geometry, Hu moment and texture feature, can be used to recognize ship target. Fuzzy-neural network method can combine fuzzy set＇s advantages with neural network＇s, by which feature- based classification and recognition for targets in images can be implemented validly. The paper depicts a fuzzy-neural network method with principal-subordinate neuro for classification and recognition at first, and then, utilize the method to classify and recognize, basing on single category feature and multi-source （multi- temporal） data fusion. Experiments＇ results indicates that classification and recognition for large ship can be implemented validly by utilizing fuzzy-neural network methods based on large ships＇ geometry features, moment features and texture features. Furthermore, using multi-source data fusion, the classification and recognition effect can be improved.