^{1}

^{1}

The electronic nose with chemical dyes as sensor can react with target gas and have specific color changes. In general, RGB camera collects a group of images to record these changes used for pattern recognition. RGB filters are not sensitive to the slight color changes, which limits the performance of this kind of electronic nose. This paper demonstrates using quantum dot spec-troscopy technology to solve this problem. Multiple quantum dot filters are placed on the surface of image sensor. When capturing images, there are more response channels of the same incident light than RGB filters. Simulation and experiment both prove that quantum dot filters with appropriate processing are more sensitive to color changes than RGB filters.

An electronic nose is an equipment that imitates the olfactory system of mammals. In general, it consists of two parts. One is an array of gas sensors acting like olfactory receptor cells of creatures, which can convert the gas information to easy detected variables, such as voltage, resistance. The other is a pattern recognition system to recognize the type, concentration or acidity of target gas according to the measured variables like creature brains [

Quantum dot spectroscopy is a kind of emerging technology, which use a variety of quantum dot colloids to make films as filters. Films with different light transmission curves filter the incident light spectrum. The image sensor can record the transmitted intensities of different filters simultaneously. The origin spectrum can be reconstructed by resolving a group of filter transmission equations [

This paper demonstrates using a group of quantum dot filters to capture the slight color changes of electronic nose sensors and trying to get better results than RGB filters by appropriately processing on the quantum dot data.

In the experiment, acetic acid with different concentrations generates target gas by volatilization. Specific sensors made of pH indicator Bromothymol Green (with NaOH solution) react with target gas in a closed container for a certain time. The changes before and after the reaction are measured by quantum dot camera, color camera and spectrometer respectively. Quantum dot camera, which has nine filters on the image sensor surface, captures the transmitted intensity of signal light. Color camera with RGB filters on the surface of the image sensor, collects data for comparison. Spectrometer records reflection spectrum of sensors for simulation.

For data processing, acetic acid with five different concentrations (0.0%, 0.6%, 1.2%, 1.8% and 2.4%) correspond to class labels 1 - 5. Thus, pattern recognition method can be used for quantitative analysis. The recognition rate, false alarm rate can be used to judge the performance of RGB filters and quantum dot filters.

Spectrum data can simulate the outputs of RGB filters and quantum dot filters. In the simulation, it is convenient to find appropriate method for processing quantum dot data with the help of spectrum. Finally, actual data is used to verify the simulation and differences are analyzed.

Regardless of the quantum efficiency of image sensor, the transmitted intensities of RGB filters and quantum dot filters are described as Equation (1)

where

As the origin state of different groups is not exactly the same, changes of the reaction cannot be shown simply by after-reaction data. Therefore, feature vectors are demanded to represent the changes. The ratio of the intensity is capable for feature vector.

R_{s}, R_{rgb} and R_{qd} are feature vectors of spectrum, RGB and quantum dot. As in Equation (2), spectrum is discrete from 360 nm - 700 nm with 1 nm interval. Support vector machine (SVM) is used to train and test feature vectors [

Spectrum has the best results due to high dimension and completeness of target wavelength band. Direct projection of spectrum under RGB and quantum dot base result bad for not sensitive to color changes. Delicate sensitive base is needed to achieve better result, which can be devised using existing quantum dot base.

Down sampling is equivalent to a group of filters with square band-pass curves whose bandwidth is n nm. Spectrum does an average every n nm. The amount of data becomes 1/n of the original.

Quantum dot filters are used to fit band-pass filter. Target transmission base is a series of Gaussian band pass curves. The center wavelength of these curves covers from 400 nm - 640 nm with 20 nm interval and full width at half maximum is 20 nm. As in Equation (5),

The quantum dot base can generate new base by linear combination.

New transmission base can be as close as possible to the target base by adjusting the linear coefficients. The optimization equation is to minimize the square error between target base and new base.

Feature Vector | Recognition Rate | False Alarm Rate |
---|---|---|

0.8273 | 0.0423 | |

0.6263 | 0.0935 | |

0.5501 | 0.1128 |

Compression Rate | 1 | 4 | 10 | 20 | 68 | 85 |
---|---|---|---|---|---|---|

Recognition Rate | 0.8273 | 0.8187 | 0.8540 | 0.8636 | 0.7511 | 0.7184 |

Simplified solution is Equation (8).

The outputs of reconstructed target base can be generated by linear combination with coefficients

New feature vector R_{a} is as Equation (10).

The target base and reconstructed base are shown in

Feature Vector | Recognition Rate | False Alarm Rate |
---|---|---|

0.6263 | 0.0935 | |

0.5501 | 0.1128 | |

0.7702 | 0.0583 |

Feature Vector | Recognition Rate | False Alarm Rate |
---|---|---|

0.6351 | 0.0912 | |

0.6222 | 0.0944 | |

0.7111 | 0.0722 |

Quantum dot filters are more sensitive to slight color changes and have a higher dimension compared to RGB filters. With appropriate data processing method, the information of a certain wavelength band can be enlarged. If the number of quantum dots is enough, the origin spectrum can be even resolved and used for classification. Furthermore, quantum dot base has larger area of integration than RGB transmission base so weak light can be captured.

Li, Z.N. and Bao, J. (2017) Electronic Nose Technology Based on Quantum Dot Filters. Optics and Pho- tonics Journal, 7, 26-32. https://doi.org/10.4236/opj.2017.78B005