In order to solve the nonlinear relationship between the output voltage and quality of the weighing sensor in the micro-package system and improve the weighing accuracy, a nonlinear compensation method is designed based on the improved BP neural network. The nonlinear compensation principle of resistance strain type weighing sensor is described. According to the relationship between input and output of weighing sensor, a neural network compensator is designed. In order to improve the neural network control performance, a penalty factor is introduced to solve the problem of excessive error caused by insufficient training. By comparison, it is found that the improved BP neural network has faster convergence speed and higher precision, and can improve the control performance of the micro-weighing packaging system.In high-speed mode, the weighing error can be controlled within 0.5%, and the actual weighing result is relatively ideal. This method can improve the dynamic performance of the system, improve the measurement accuracy, and meet the requirements of weighing and packaging industry.