Design Of Electronic Nose Based On Hardware Implemented Bp Neural Network

In gas detection, compared with traditional detection methods, because of the cheap price, small, easy to carry and can be measured in real time, electronic nose has been widely used. In recent years, domestic and foreign research-related electronic nose technology is also increasing. The electronic nose system is mainly involved gas sensors, pattern recognition and other aspects of technology. The gas sensor is the basis for an electronic nose, the impact on the overall performance of the electronic nose. Existing cross-sensitive gas sensors exist, to identify the gas information, required an array of a plurality of sensors, and then pattern recognition algorithms to identify the array signal corresponding to the gas information.In this paper, eight gas sensors sensor array, the choice of BP neural network as a pattern recognition algorithm. Since the BP neural network software exists operation is slow, poor stability, although dedicated neural network chip operation speed but inflexible, and expensive, paper selects FPGA offline by BP neural network algorithm. Design and implementation of first hidden layer BP neural node key part --Sigmoid function and multiply-accumulate module, on this basis, to complete the design of the hidden layer and output layer nodes. Then, FPGA-based off-line learning BP neural network is implemented with the weight and threshold value storage ROM module、the nodes and necessary control logic. At the same time the design is complete electronic nose system hardware circuit and other auxiliary module, and write verilog driving uxiliary module.The electronic nose system design and hardware and software debugging are finished,then, verify the functions of the electronic nose. Using data acquisition acquire sensor array response signal in different atmospheres, then creation and training of BP neural network in MATLAB software.Then according to the trained network, before adjusting the design parameters of the neural network module. The results show that the electronic nose system can successfully identify qualitative methane and ammonia, but quantitative measurement of large errors.