Neural network which can adapt the sample data by training has good fault-tolerance and can be used in the field of intelligence widely. In the embedded system, restricted to the resources and the capacity of processor, the neural network application has a series of problems, such as losing timelines and the system could be collapsed easily. This article discusses how to use limited memory, processor and external equipment resources to achieve the neural network algorithm for improving the universality of detection system and adaptive ability in the embedded intelligent measuring system.