描述了神经网络的非线性能力，提出边沿提取的神经网络应用新算法。在理解Marr算子意义的基础上，将Marr特征融入神经网络结构，提高该算法的整体性能。同时，对不同边沿提取算法在不同噪声环境下进行比较，结果表明，新算法的网络结构具有较强的抗器材怕能力，可以在实际环境的机器人视觉领域得到很好的应用。

This paper presents the powerful ability of the neural network, and puts forward a new structure of the network in edge detection. On the basis of the understanding of Marr algorithm, the characteristics of Marr are taken great consideration into the structure of the network so as to improve the performance of algorithm. Meanwhile, under different noisy environment, among different algorithms in edge detection, comparative analyses are made to reveal that new method owns the powerful ability of anti noise, and that more applications are done well in the robot vision.