Abstract: In the embedded field represented by missile spaceflight, the traditional template matching method is often used to conduct the target location search, and the accuracy of the matching will be greatly reduced due to the complexity of the scene and the deformation or Angle transformation of the target. However, the algorithm that USES CNN network to directly detect the target is difficult to transplant to the embedded platform due to the huge network scale and computation. Siamese network matching algorithm) is presented in this paper, on the one hand, using the network feature extraction ability, improve the matching accuracy under complex scene, on the other hand using symmetry of Siamese, using graph matching process supervision, significantly reduced the size of the network, the algorithm can be transplanted to the embedded platform, have very strong practicability.