A Committee Machine Gas Identification System Based on Dynamically Reconfigurable FPGA

Abstract

Abstract-This paper proposes a gas identification system based on the committee machine (CM) classifier, which combines various gas identification algorithms, to obtain a unified decision with improved accuracy. The CM combines five different classifiers: K nearest neighbors (KNNs), multilayer perceptron (MLP), radial basis function (RBF), Gaussian mixture model (GMM), and probabilistic principal component analysis (PPCA). Experiments on real sensors' data proved the effectiveness of our system with an improved accuracy over individual classifiers. Due to the computationally intensive nature of CM, its implementation requires significant hardware resources. In order to overcome this problem, we propose a novel time multiplexing hardware implementation using a dynamically reconfigurable field programmable gate array (FPGA) platform. The processing is divided into three stages: sampling and preprocessing, pattern recognition, and decision stage. Dynamically reconfigurable FPGA technique is used to implement the system in a sequential manner, thus using limited hardware resources of the FPGA chip. The system is successfully tested for combustible gas identification application using our in-house tinoxide gas sensors.Index Terms-Committee machine (CM), dynamically reconfigurable field programmable gate array (FPGA), gas identification, pattern recognition.

Identification and quantification of mixed air pollutants based on homotopy method for gas sensor array

“…Sensor responses to CH 4 and SO 2 have been tested and collected in this paper. Other target gases such as NO 2 , CO, acetaldehyde and so on are still being tested for the purpose of minimizing the cross sensitivity among those targets.…”

“…These fingerprints serve as features and are usually projected to a lower dimensional space to reduce the computational complexity before classification. Such classification algorithm have been implemented in electronic devices with low power, low cost and portability [3][4]. However, classification techniques are only suitable for discriminating individual gases in a mixture and can only identify one unknown sample at a given time.…”

Gas classification using binary decision tree classifier

“…The positive values are diverted to one branch and the negative values are diverted to the second branch. Sensor couple (1,6) is selected at the root node for distribution of the gases in two branches. It diverts hydrogen, ammonia and propane in one branch, and carbon dioxide and carbon monoxide in the other branch.…”

“…This response vector is input to the pattern recognition system for classification. The K nearest neighbor (KNN) method, multilayer perceptron (MLP), Principal component analysis (PCA), and Linear discriminant analysis (LDA) have been used for odors identification [1], [2], [3], [4], [5], [6], [7].…”

Versatile Gas Monitoring System on the Heterogeneous Zynq SoC Platform

“…The commonly used algorithms for gas classification are K nearest neighbors (KNNs), multilayer perceptron (MLP), radial basis function (RBF), gaussian mixture model (GMM) and probabilistic principal component analysis (PPCA). These five algorithms were combined in a committee machine [4] for an improved accuracy and recognition rate, in this work the system was implemented on a dynamically reconfigurable field programmable gate array (FPGA). Decision tree (DT) classifier can also be used for gas identification such as in [5] where it was implemented on both FPGA and application specific integrated circuit (ASIC).…”