系统根据用户对检索结果的评价，动态构造神经网络，描述图像之间的相似性；图像间的这种相似性以及本次检索结果可以作为以后检索的历史信息保存在神经网络中，从而提高下一次检索的效率。实验表明，该方法嵌入到典型的图像检索系统中，改善了图像检索性能。-System based on user evaluation of search results, dynamic structural neural network, described the similarity between images　this similarity between images and the search results can be used as future retrieval of information stored in the history of neural networks, thereby enhancing the under a retrieval efficiency. Experiments show that the method is embedded in a typical image retrieval system, improve the image retrieval performance.