Research And Engineering Application Of Gas Recognition Algorithm Based On Porphyrin Gas Detection System

In recent years, air pollution is getting worse. The sandstorm and haze, whichcontain pathogenic toxic gases, may cause respiratory disease, even cancer. The clinicalmedical reports have pointed out that cancer patients can get early diagnosis bydetecting their exhaled gases. The gas detection system with porphyrin as sensors canreact with gas quickly and accurately. Because porphyrin is a high sensitivity material ofgas sensor, it may bring the discrepancies between the results of parallel experimentswhen there are slight changes of environment or gas concentration. This phenomenon iscalled “divergent problem”, which can reduce the recognition accuracy. Therefore, anintelligent algorithm that can identify gases by processing data of difference maps isneeded to keep good recognition rate of gas detection system.In this study, a gas species recognition algorithm that was employed in porphyrinsensor array (PSA) gas detection system had been proposed. It was applied to identifygases by processing the differences of RGB channels of each dye spot in differencemaps (the “eigenvalues”), using the rough set reduction and backpropagation neuralnetwork. After the availability and accuracy had been validated, this new algorithmwould be made into a C++program on the platforms of VC++and Qt respectively. Then,this program had been added to the software installed in PC of PSA gas detectionsystem, and ported to the embedded device. The content of this study mainly includesthe following aspects:(1) The necessary information of eigenvalues had been extracted by the rough settheory, and then be the input/output vector of the BP-ANN. it was helpful to avoidthe local minimum value and improve the training efficiency.(2) The convergences of training, recognition accuracy and generalization of BP-ANNhad been validated well. By comparing the recognition results of cluster analysisand the new algorithm respectively, we had proved that the algorithm perform wellenough to identify gases even in the divergent problem.(3) The functional requirements analysis and architecture design for the recognitionprogram and training program of the new algorithm had been made respectively.The programmatic implementation of the algorithm using VC++had completed,and the algorithm was successfully added to the software of PSA system.(4) Rewritten the recognition program for adapting the embedded C programming standards, and ported to the embedded device. The training results generated by thetraining program in PC were ported to the embedded device as the parameters ofthe recognition program to complete the gas recognition function. The problem ofhow neural network training program porting and applying in embedded device wassolved.