论文部分内容阅读
该文提出一种由多层神经网络与自组织神经网络相结合进行多类别遥感图象分类的复合神经网络分类方法。第1步将训练样本按其统计特征分成若干组,用不同组别的训练样本分别训练BP网络。第2步将这些训练好的BP网络并联构成有监督分类器,对遥感图象进行有监督分类。第3步用BP网络的分类结果对Kohonen网络进行自组织训练,用训练好的Kohonen网络构造无监督分类器,对遥感图象进行细分。通过对SPOT遥感图象的分类实验表明,该方法对多类别遥感图象很适用,能显著提高分类的数目和精度,对一幅SPOT遥感图象进行的分类实验,结果可分类别数高达48类,对其16大类的有监督分类的精度可达91.6%。
This paper proposes a new classification method based on multi-layer neural network and self-organizing neural network to classify multi-class remote sensing images. Step 1 Divide the training samples into several groups according to their statistical characteristics, and use different groups of training samples to train the BP network respectively. Step 2 The trained BP networks are connected in parallel to form a supervised classifier for supervised classification of remote sensing images. The third step is to classify the Kohonen network using the classification results of the BP network and construct an unsupervised classifier using the trained Kohonen network to segment the remote sensing image. The experiments on the classification of SPOT remote sensing images show that this method is suitable for multi-class remote sensing images and can significantly improve the number and accuracy of classification. The classification experiment on a SPOT remote sensing image can be divided into 48 The accuracy of supervised classification of its 16 major categories can reach 91.6%.