Alzheimer’s disease will lead to the atrophy of the Hippocampus. In order to recognize the position changes of Hippocampus, improve the image contour quality, further improve the accuracy of convolution, and achieve the purpose of accurately extracting image information, convolutional neural network is introduced to recognize the Hippocampus region with brain magnetic resonance imaging, and a method combining multi-level 3D U-NET is proposed based on single-stage U-NET. The results showed that this model could enhance the segmentation performance, significantly improved the segmentation accuracy which had certain clinical significance for the brain to recognize the Hippocampus and the automatic discrimination of Alzheimer’s disease.