中文摘要：

通过提取图像的底层特征,将特征输入BP神经网,应用共轭梯度法对网络进行有监督训练,即将先验知识加入神经网络中;一旦训练成功,依靠网络出色的泛化能力,系统就能正确识别图像中的物体和各物体的位置信息,这样就实现了对图像语义的理解。实验验证结果表明该系统对特定测试图像集的理解正确率达到了100%。

英文摘要：

The research of image semantics is an active field in present time. It remains to be unsolved on the problem of ＂semantic gap＂ between the low-level image features and image semantics. A new method for the object classification is proposed, in which low-level features are extracted from some image, then inputed to a BP neural network and trained in the teacher＇ s guide. Once the training effectively finishs, by the excellent extending capacity of the net, the system can classify the objects in the image correctly, and give the location information of every object. In this way, the semantics of the image can get . Experimental results show the effectiveness of the proposed method on the object classification.