The measurement of the combustion flame has been one of the important research directions in the field of combustion. This paper designs a flame combustion state detection system based on FPGA of the Altera's DE2 board with CMOS image sensors to capture images of the flame. The algorithms are implemented on the FPGA such as the center point extraction, color extraction, temperature calculation, area calculation and characteristics extraction which will be as the input of BP neural network from the flame image. The current state of the flame combustion will be calculated by making use of the trained neural network. The hardware description of the system is completed by Verilog language, including device initialization, data acquisition and storage and other parts. The design of soft core is implemented in SOPC Builder. Software algorithms are completed by using C language programming environment in the Nios II. The experimental results show that the system could obtain ideal results.