基于CNN-DW的遥感图像分类检索方法

Remote Sensing Image Classification and Retrieval Method Based on CNN-DW

针对传统图像检索方法存在的检索范围过大、检索效率低下的问题,提出了一种基于卷积神经网络和距离权重的图像检索方法(CNN-DW),该方法可以从海量遥感图像中检索出与查询图像具有相似或相同特征的检索图像,图像检索试验表明:CNN-DW检索法较传统卷积神经网络(Convolutional Neural Networks,CNN)检索法的分类检索效果有显著提升,仅需要更少数量的训练集就能达到良好的检索效果,可在遥感图像分类检索工作中予以合理运用。

In view of the problems of the traditional image retrieval methods,such as too large retrieval range and low retrieval efficiency,an image retrieval method based on convolution neural network and distance weight(cnn-dw method)was proposed.This method could retrieve the retrieval images with similar or the same characteristics as the query images from the massive remote sensing images.The image retrieval experiments showed that the cnn-dw retrieval method was effective Compared with the traditional CNN retrieval method,the classification retrieval effect had been significantly improved.Only a small number of training sets was needed to achieve good retrieval effect,which could be reasonably used in remote sensing image classification and retrieval work.