Abstract: The paper proposes a new algorithm to obtain image semantic by neural network.By means of a designed RBF neural network,the paper establishes the mapping relationship between the low－level features such as color,texture and shape,and the high－level semantic.A new training method using genetic algorithm is presented,which can get all the parameters（such as quantity,centers,widths and connection weights of RBF neural network）.The successfully－trained neural network can obtain image semantic automatically.Experimental results indicate that the proposed image retrieval algorithm is effective in characterizing image semantic．

摘要： 提出一种利用神经网络获取图像语义的算法。通过构建一个RBF神经网络,在图像的颜色、纹理、形状等低层视觉特征和高层语义特征之间建立映射关系。利用遗传算法训练RBF网络,获得RBF网络的隐节点个数、中心、宽度和连接权值等参数值,训练成功后的神经网络能够自动获取图像的语义。实验结果表明,该算法具有较好的基于语义的检索效果,体现了人对图像内容的理解,符合人的思维习惯。