CN117517299B - H2S colorimetric/electrical sensor and deep learning colorimetric/electrical dual sensing system - Google Patents

Abstract

The invention particularly relates to an H ₂ S colorimetric/electrical sensor and a deep learning colorimetric/electrical dual-sensing system, which are synthesized by combining an H ₂ S sensitive nanofiber membrane with a PET flexible interdigital electrode, and can simultaneously have two response characteristics of colorimetric and resistance change. The deep learning colorimetric/electrical dual-sensing system which visually displays visual response and simultaneously records resistance response is combined with the color-changing monitoring platform and the resistance response module, the effect of 'instant shooting instant measuring' is achieved, the gas concentration is rapidly and accurately detected, the detection advantages of strong applicability, wide application range, visualization and high precision can be achieved through a dual-sensing response complementary mechanism, the detection of the concentration H ₂ S in a large range can be achieved, the resolution is high, the response time is short, the selectivity is high, the low detection limit of 0.1ppm is provided, the stability is good, and the concentration detection accuracy can reach 99.8%.

Description

H2S colorimetric/electrical sensor and deep learning colorimetric/electrical dual sensing system

Technical Field

The present invention belongs to the technical field of gas sensors, and in particular relates to a H ₂ S colorimetric/electrical sensor and a deep learning colorimetric/electrical dual sensing system.

Background Art

Hydrogen sulfide ( _H2S ) is a colorless, flammable and toxic gas with a rotten egg smell. Prolonged exposure to _H2S at low concentrations (15 to 50ppm) can irritate mucous membranes and cause headaches, dizziness and nausea, _while high concentrations (>200ppm) can cause suffocation, coma or loss of consciousness. Therefore, it is critical to develop a monitoring system that can respond to hydrogen sulfide gas with high sensitivity.

Colorimetric gas sensors are sensors that generate visual sensing signals through color changes to identify the gas to be tested. They have the advantages of low power consumption, low detection limit, small size, convenient operation, and visual response. Currently, commonly used colorimetric sensors are based on thin films, gels, nanofibers, fabrics, etc. to detect various harmful gases, such as ammonia, phosgene, hydrogen sulfide, and volatile organic compounds.

CN109856122A discloses a colorimetric hydrogen sulfide sensor, which uses green synthetic gellan gum-silver nano-solution and agar solution to synthesize solid gel, has a high sensitivity response to hydrogen sulfide gas, and when reacting with hydrogen sulfide, the color gradually changes from yellow to colorless. However, the color change of the current _H2S colorimetric gas sensor needs to rely on large instruments for observation. If it is only recognized by the naked eye, it is difficult to capture the information generated by the color change, resulting in low detection accuracy. Therefore, the current colorimetric gas sensor has only completed part of the detection work, and has not formed a complete _H2S monitoring system. It is difficult to break through the use restrictions of certain environments, especially in strong light interference and dark and lightless environments, the colorimetric sensing function is limited. Electrical sensing has good stability and selectivity in different light environments. Combining colorimetric sensing with electrical sensing can solve the problem of difficult detection work in extreme light environments. At the same time, the dual sensing system combined with deep learning can avoid the reliance on high-power instruments during the detection process, realize an ultra-low energy consumption (i.e. no need for large spectrometers, colorimeters, etc.), high-precision colorimetric-electrical dual-function integrated sensing system, and thus form a complete gas sensing system.

Summary of the invention

In view of the shortcomings of the prior art, the present invention provides an H ₂ S colorimetric/electrical sensor and a deep learning colorimetric/electrical dual sensing system. The present invention synthesizes an H ₂ S colorimetric/electrical sensor by combining an H ₂ S sensitive nanofiber membrane with a PET flexible interdigital electrode, and combines a color change monitoring platform and a resistance response module to obtain a deep learning colorimetric/electrical dual sensing system that can intuitively display visual responses, achieving the effect of "shooting and measuring", rapid and accurate detection of gas concentration, and realizing the advantages of true visualization, ultra-low power consumption, and high precision detection.

In order to achieve the above-mentioned object, the first aspect of the present invention provides a H ₂ S colorimetric/electrical sensor, comprising a H ₂ S sensitive nanofiber membrane and a PET flexible interdigital electrode, wherein the H ₂ S sensitive nanofiber membrane is fixed on the PET flexible interdigital electrode by an electrospinning process, and the H ₂ S sensitive nanofiber membrane is Pb(CH ₃ CO ₂ ) ₂ /PVA nanofiber. The H ₂ S sensitive nanofiber membrane is combined with the PET flexible interdigital electrode, so that the sensor can improve its applicability and wide application according to its colorimetric-resistance dual response characteristics.

Furthermore, the H ₂ S sensitive nanofiber membrane is a dense network structure formed by interlaced nanofibers, the diameter of the nanofibers is 250-400 nm, and the thickness of the H ₂ S sensitive nanofiber membrane is 0.5-0.7 mm.

Specifically, the hydrogen sulfide sensing sensitivity of the nanofiber membrane is closely related to its microstructure. The large specific surface area and the pore structure formed by the interlaced fibers are the key factors affecting the gas sensitivity of the nanofiber membrane. Smaller fiber sizes will produce more pore structures, making the reaction faster and more complete, thereby improving the sensitivity of the device. In addition, the fiber membrane has a certain thickness, which can ensure that the color change process is not interfered by the color of the PET flexible interdigital electrode itself, and at this thickness, the gas sensing performance of H ₂ S is optimal.

The second aspect of the present invention provides a method for preparing a H ₂ S colorimetric/electrical sensor, comprising the following steps:

(1) dissolving lead acetate trihydrate, NaF and sodium dodecyl sulfate in deionized water to obtain a lead acetate solution;

(2) slowly adding polyvinyl alcohol into the lead acetate solution, and heating and stirring in a water bath to obtain an electrospinning solution;

(3) The electrospinning solution is fixed on a PET flexible interdigital electrode in the form of a H ₂ S sensitive nanofiber membrane through an electrospinning process to obtain the H ₂ S colorimetric/electrical sensor.

Furthermore, the mass ratio of lead acetate trihydrate: NaF: sodium dodecyl sulfate: polyvinyl alcohol is (0.8-1.2) g: (25-35) mg: (10-20) mg: (2.2-2.6) g, and the heating and stirring temperature in step (2) is 80-90° C. and the time is 8-10 h.

Furthermore, the electrospinning process in step (3) is as follows: using a 20, 22 or 24 gauge metal needle, applying 20±1.5 kV static voltage at the needle, the propulsion speed of the microinjection pump is 1±0.2 mL/h, and the receiver is 10±0.5 cm away from the needle. During the spinning process, the ambient humidity is 35%±5%, the temperature is 25±2°C, and the spinning time is 8~10 h. Adjusting the preparation process and parameters of the sensor can change various surface micromorphologies of nanofibers, such as mesh, porous and open structures, and the micromorphology of the nanofiber membrane affects the gas-sensitive performance of the sensor.

A third aspect of the present invention provides a deep learning colorimetric/electrical dual sensing system, comprising a gas chamber, a H ₂ S colorimetric/electrical sensor, a resistance response module, a color change monitoring platform, an analysis module, and a terminal;

The H ₂ S colorimetric/electrical sensor is the aforementioned H ₂ S colorimetric/electrical sensor, located inside the gas chamber, and used to detect H ₂ S gas in real time to obtain a gas concentration signal, and convert the gas concentration signal into a color change signal and a resistance change signal;

The resistance response module is connected to the H ₂ S colorimetric/electrical sensor and is used for real-time acquisition, storage and display of the resistance change signal output by the sensor;

The color change monitoring platform is used to collect, store and display the color change signals output by the sensor in real time;

The analysis module is loaded on the terminal, and is used to analyze and process the color change signal collected by the color change monitoring platform, and output the H ₂ S concentration based on the analysis and processing result; the terminal is used to display the H ₂ S concentration detection result.

Furthermore, the color change monitoring platform includes a fill light unit, a collection unit, a storage unit and a display unit, wherein the fill light unit is an LED fill light, the collection unit is an integrated camera module, which is used to collect the color change signal output by the sensor in real time, the storage unit is used to store the collected color change signal, and the display unit is used to display the color change signal. The specific implementation method is that the color change monitoring platform uses an integrated camera module to capture the color change image of the sensor in the _H2S environment, shoots the color change sample under a uniform light source, and controls the sample specifications to be consistent, and then uses the user graphical interface designed by the Tkinter library provided by Python to return the shooting view in real time, and can manually or automatically shoot the color change image and store it in a local database.

Furthermore, the analysis module uses a color segmentation algorithm to segment the color change signal collected by the color change monitoring platform, and calculates the color difference value ΔE of the color change signal by introducing a color difference formula. The H ₂ S concentration prediction model designed based on the multi-layer perceptron algorithm gives the H ₂ S concentration after measurement. In the specific implementation process, the color segmentation algorithm is introduced to further segment the color change signal, that is, the color response result, and the degree of color change in the sensor area is visible to the naked eye. The CIE color difference formula is introduced to digitize the color response result without an external colorimeter, and the measured H ₂ S gas concentration is quickly given through the multi-layer perceptron prediction model.

Furthermore, the deep learning colorimetric/electrical dual sensing system utilizes a resistance response module to form a colorimetric-electrical dual-function integrated sensing system, which can simultaneously detect the resistance response value change of the _H2S colorimetric/electrical sensor and then detect _H2S gas, and can avoid the detection work from being stopped in special environments of strong light interference and darkness.

Furthermore, the deep learning colorimetric/electrical dual sensing system has a detection range of 0-100ppm, a detection limit and resolution of 0.1ppm, a colorimetric response time of less than 4s, and maintains stable performance for 180 days and can work normally even in strong light interference and dark scenes.

Compared with the prior art, the present invention has the following beneficial effects:

(1) The present invention synthesizes an _H2S colorimetric/electrical sensor by combining an _H2S sensitive nanofiber membrane with a PET flexible interdigital electrode, and obtains a deep learning colorimetric/electrical dual sensing system that can intuitively display visual responses by combining a color change monitoring platform and an analysis module, thereby achieving the effect of "shoot and measure" and rapid and accurate detection of gas concentration, and realizing the advantages of true visualization, ultra-low power consumption, and high-precision detection; and a resistance response module is added as a response compensation of the deep learning colorimetric/electrical dual sensing system to ensure the completeness of the deep learning colorimetric/electrical dual sensing system.

(2) The deep learning colorimetric/electrical dual sensing system of the present invention can detect a wide range of H ₂ S concentrations, has a low detection limit, a short response time, high selectivity, a resolution of 0.1 ppm, good stability, and a concentration detection accuracy of up to 99.8%.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1a is a manufacturing process flow of the deep learning colorimetric/electrical dual sensing system of the present invention; FIG1b is a concentration detection process flow of the deep learning colorimetric/electrical dual sensing system of the present invention;

FIG2a is a SEM image of the H ₂ S sensitive nanofiber membrane at a magnification of 1000; FIG2b is a SEM image of the H ₂ S sensitive nanofiber membrane at a magnification of 10000; FIG2c is a SEM image of the H ₂ S sensitive nanofiber membrane after being placed in an environment with a relative humidity of 50% for one week;

FIG3 is a schematic diagram of concentration detection results of the deep learning colorimetric/electrical dual sensing system of the present invention;

FIG4a is a display result of an image processed based on the Euclidean distance segmentation algorithm; FIG4b is a display result of an image processed based on the Manhattan multi-channel distance segmentation algorithm; FIG4c is a display result of an image processed based on the Manhattan single-channel distance segmentation algorithm;

FIG5 shows the color change results of the H ₂ S colorimetric/electrical sensor under different H ₂ S concentration environments under LED, natural light and dim light at night;

FIG6 is a color difference ΔE variation curve of the H ₂ S colorimetric/electrical sensor under different H ₂ S concentration environments;

Figure 7a shows the training process of the H ₂ S concentration prediction model designed based on the 3-layer perceptron algorithm; Figure 7b shows the gas concentration prediction results of the cold model and the warm model based on the Manhattan multi-channel distance color segmentation algorithm; Figure 7c shows the test accuracy of the cold model; Figure 7d shows the test accuracy of the warm model; Figure 7e shows the loss comparison curve of the cold model and the warm model;

FIG8a shows the color difference ΔE change results of the H ₂ S colorimetric/electrical sensor in NH ₃ , SO ₂ , NO ₂ , H ₂ and H ₂ S gas environments; FIG8b shows the color difference ΔE change results of the H ₂ S colorimetric/electrical sensor in two mixed gas environments; FIG8c shows the color difference ΔE change results of the H ₂ S colorimetric/electrical sensor within 180 days;

FIG9a is the resistance response result of the H ₂ S colorimetric/electrical sensor in different H ₂ S concentration environments; FIG9b is the resistance response result of the H ₂ S colorimetric/electrical sensor in NH ₃ , SO ₂ , NO ₂ , H ₂ and H ₂ S gas environments; FIG9c is the resistance change result of the H ₂ S colorimetric/electrical sensor within 180 days; FIG9d is a comparison diagram of the resistance response and color difference ΔE response results of the H ₂ S colorimetric/electrical sensor;

Figure 10a is the color difference ΔE change result of the H ₂ S colorimetric/electrical sensor in a 0-5ppm H ₂ S environment, and Figure 10b is an enlarged schematic diagram of the color difference ΔE change result of Figure 10a in a 0.1-0.5ppm H ₂ S environment; Figure 10c is a schematic diagram of the difference in color difference ΔE change of the H ₂ S colorimetric/electrical sensor in a 50.1-50.9ppm H ₂ S environment; Figure 10d is the resistance change result of the H ₂ S colorimetric/electrical sensor in a 0-5ppm H ₂ S environment, and Figure 10e is an enlarged schematic diagram of the resistance change result of Figure 10d in a 0.1-0.5ppm H ₂ S environment; Figure 10f is a schematic diagram of the resistance change difference of the H ₂ S colorimetric/electrical sensor in a 50.1-50.9ppm H ₂ S environment.

DETAILED DESCRIPTION

The principles and features of the present invention are described below with reference to examples. The examples are only used to explain the present invention and are not used to limit the scope of the present invention. Figure 1a is a manufacturing process flow of the deep learning colorimetric/electrical dual sensing system of the present invention; Figure 1b is a concentration detection process flow of the deep learning colorimetric/electrical dual sensing system of the present invention.

Example 1

A H ₂ S colorimetric/electrical sensor comprises a H ₂ S sensitive nanofiber membrane and a PET flexible interdigital electrode, wherein the thickness of the H ₂ S sensitive nanofiber membrane is 0.6 mm.

The preparation method comprises the following steps:

S1 Add 1 g (CH ₃ COO) ₂ Pb·3H ₂ O, 30 mg NaF, and 15 mg sodium dodecyl sulfate (SDS) into 20 mL of deionized water and stir magnetically at 27 °C for 1.5 h to obtain a clear lead acetate solution;

S2 Slowly add 2.4g polyvinyl alcohol (PVA) into the lead acetate solution, stir in a water bath at 85 ^° C for 9h, take out and let stand at room temperature for 30 minutes after stirring to obtain the electrospinning solution;

S3 The obtained electrospinning liquid is sucked into a disposable syringe and electrospinning is performed using a 22-gauge metal needle: a 20 kV static voltage is applied to the needle, the propulsion speed of the microinjection pump is 1 mL/h, the receiver is 10 cm away from the needle, the ambient humidity during the spinning process is 35%, the temperature is 25 ^o C, and the spinning time is 9 hours. The H ₂ S sensitive nanofiber membrane is fixed on the PET flexible interdigital electrode through an electrospinning process to obtain the H ₂ S colorimetric/electrical sensor.

Figure 2 is a SEM image of the H ₂ S sensitive nanofiber membrane. As can be seen from Figure 2a, the nanofibers are randomly distributed and the fibers are interlaced to form a dense network structure. As can be seen from Figure 2b, the diameter of the nanofibers is 250-400nm. Since (CH ₃ COO) ₂ Pb·3H ₂ O is added to the raw material, the ion concentration in the spinning solution is increased, the solution conductivity is increased, and the instability of the solution jet in the electric field is increased, so that the sub-nanofibers are split from the main fibers, thereby forming a spider web structure in the electrospinning process. The smaller fiber size will produce more pore structures, making the reaction faster and more complete, thereby improving the sensitivity of the device. In addition, in order to verify the stability of the H ₂ S sensitive nanofiber membrane, the H ₂ S sensitive nanofiber membrane was placed in an atmospheric environment (relative humidity 50%) for 1 week. As can be seen from Figure 2c, no (CH ₃ COO) ₂ Pb particles were precipitated on the nanofibers, indicating that the H ₂ S sensitive nanofiber membrane has good stability.

Example 2

A H ₂ S colorimetric/electrical sensor comprises a H ₂ S sensitive nanofiber membrane and a PET flexible interdigital electrode. The H ₂ S sensitive nanofiber membrane is a dense mesh structure formed by interlacing nanofibers, and the membrane thickness is 0.5 mm.

The preparation method comprises the following steps:

S1 0.8 g (CH ₃ COO) ₂ Pb·3H ₂ O, 25 mg NaF, and 10 mg SDS were added to 20 mL of deionized water and stirred magnetically at 25 °C for 1.5 h to obtain a clear lead acetate solution;

S2 Slowly add 2.2g PVA into the lead acetate solution, stir in a water bath at 80 ^° C for 8h, take out and let stand at room temperature for 25min after stirring to obtain the electrospinning solution;

S3 The obtained electrospinning liquid is sucked into a disposable syringe and electrospinning is performed using a 20-gauge metal needle: an electrostatic voltage of 18.5 kV is applied to the needle, the propulsion speed of the microinjection pump is 0.8 mL/h, the receiver is 9.5 cm away from the needle, the ambient humidity during the spinning process is 30%, the temperature is 23 ^o C, and the spinning is performed for 8 hours. The H ₂ S sensitive nanofiber membrane is fixed on the PET flexible interdigital electrode through an electrospinning process to obtain the H ₂ S colorimetric/electrical sensor.

Example 3

A H ₂ S colorimetric/electrical sensor comprises a H ₂ S sensitive nanofiber membrane and a PET flexible interdigital electrode. The H ₂ S sensitive nanofiber membrane is a dense mesh structure formed by interlacing nanofibers, and the membrane thickness is 0.7 mm.

The preparation method comprises the following steps:

S1 1.2 g (CH ₃ COO) ₂ Pb·3H ₂ O, 35 mg NaF, and 20 mg SDS were added to 20 mL of deionized water and stirred magnetically at 25 °C for 2 h to obtain a clear lead acetate solution;

S2: 2.6 g PVA was slowly added into the lead acetate solution, and stirred in a water bath at 90 ^° C for 10 h. After the stirring was completed, the solution was taken out and allowed to stand at room temperature for 35 min to obtain an electrospinning solution.

S3 The obtained electrospinning liquid is sucked into a disposable syringe and electrospinning is performed using a 24-gauge metal needle: a static voltage of 21.5 kV is applied to the needle, the propulsion speed of the microinjection pump is 1.2 mL/h, the receiver is 10.5 cm away from the needle, the ambient humidity during the spinning process is 40%, the temperature is 27 ^o C, the spinning is 10 hours, and the H ₂ S sensitive nanofiber membrane is fixed on the PET flexible interdigital electrode through the electrospinning process to obtain the H ₂ S colorimetric/electrical sensor.

Deep learning colorimetric/electrical dual sensing system

A deep learning colorimetric/electrical dual sensing system is constructed using the colorimetric/electrical sensor in Example 1, which mainly includes a gas chamber, an H ₂ S colorimetric/electrical sensor, a color change monitoring platform, an analysis module, and a terminal. The H ₂ S colorimetric/electrical sensor is located inside the gas chamber, and is used to detect H ₂ S gas in real time to obtain a gas concentration signal, and convert the gas concentration signal into a color change signal and a resistance change signal; the color change monitoring platform is used to collect, store, and display the color change signal output by the sensor in real time; the analysis module is loaded on the terminal, and is used to analyze and process the color change signal collected by the color change monitoring platform, and output the H ₂ S concentration based on the analysis and processing results; the terminal is used to display the H ₂ S concentration detection result. FIG3 is a schematic diagram of the concentration detection result of the deep learning colorimetric/electrical dual sensing system of the present invention.

The deep learning colorimetric/electrical dual sensing system also includes a resistance response module, which is connected to the H ₂ S colorimetric/electrical sensor and is used to collect, store, and display the resistance change signal output by the H ₂ S colorimetric/electrical sensor in real time to detect H ₂ S gas.

The color change monitoring platform includes a fill light unit, a collection unit, a storage unit and a display unit. The fill light unit is an LED fill light, the collection unit is an integrated camera module, which is used to collect the color change signal output by the sensor in real time, the storage unit is used to store the collected color change signal, and the display unit is used to display the color change signal. Specifically, the color change monitoring platform uses an integrated camera module to capture the color change image of the sensor in the _H2S environment. The camera is kept at a distance of 15cm from the color change sample, and the color change sample is photographed under a uniform light source. The sample specifications are controlled to be consistent at 2cm×2cm. Then, the user graphical interface designed using the Tkinter library that comes with Python can return the shooting view in real time, and can manually or automatically shoot the color change image and store it in the local database, which is convenient for the later algorithm to call and process the data.

In the specific operation process, the H ₂ S colorimetric/electrical sensor is placed in the gas chamber, and H ₂ S gas with quantitative increasing concentration is introduced in sequence. The color change of the sensor is observed in real time through the color change monitoring platform interface, and the color change results of the sensor at different concentrations are photographed. These color change samples are uniformly stored in the local database, and they all have labels corresponding to the H ₂ S gas concentration, that is, the current known gas concentration. For the color change sample images without labels, it is necessary to immediately identify the H ₂ S gas concentration in the environment where the sample is located. In order to achieve the "shoot and get" measurement effect, the color expression is converted into intuitive and accurate gas concentration data to inform the user, and the analysis module is required to achieve rapid gas concentration measurement.

The analysis module uses a color segmentation algorithm based on Euclidean distance or Manhattan distance (single channel or multi-channel) to segment the color change signal collected by the color change monitoring platform, and calculates the color difference value ΔE of the color change signal by introducing a color difference formula such as CIE 1976, CIE 1995 or CIEDE2000. The H ₂ S concentration prediction model designed based on the multi-layer perceptron algorithm gives the H ₂ S concentration after measurement. Among them, the color difference value ΔE can distinguish the degree of chromaticity change of the colorimetric/electrical sensor under different gas concentration environments, and the color segmentation algorithm processes the color change signal, i.e., the color change sample image, with a cold color set or warm color set model.

In a specific implementation, the local database of the color change monitoring platform is called, and 200 images that have changed color are divided as the training set of the later prediction model, and 50 color change images are used as the test set to verify the accuracy of the model. The sample image is preprocessed using the color segmentation algorithm based on Euclidean distance and Manhattan distance. The pixel mean calculation can obtain the (R, G, B) mean vector of the color change image, and the sample RGB value is converted into the L*a*b* color space applicable to the colorimeter. The color difference value ΔE of the image can be calculated by the CIE color difference formula to distinguish the chromaticity change of the colorimetric/electrical sensor under different gas concentration environments. The color change from white to dark brown can prove the presence of H ₂ S. The H ₂ S concentration prediction model designed based on the multi-layer perceptron algorithm can quickly give the final measurement result, and finally can achieve high-precision, low-cost, cross-platform applicable H ₂ S detection, and truly achieve the measurement effect of "shoot and get".

The color-changing samples collected during the experiment need to undergo image preprocessing before they can be used to train the model. A color-changing sample image needs to be processed into a (R, G, B) vector, and this vector is used as a "pointer" to the gas concentration value corresponding to the color-changing sample and used as input to train the concentration prediction model. When calculating the mean of the color-changing sample image, each pixel of the image is segmented and the value of a single pixel (R _i , G _i , B _i ) is judged. The (R _i , G _i , B _i ) value of each pixel is calculated cyclically and accumulated, respectively, to obtain the pixel mean vector (R, G, B) of the color-changing sample. It is specifically calculated using the following formula:

The color segmentation algorithm is an important step in image preprocessing. It classifies the pixels of the color image based on the RGB space, that is, extracts the region of the color-changing sample image, and then assigns the average color to the segmented region based on the given color sample set, so that the reaction degree of each region on the sample image can be clearly distinguished. In order to achieve effective color segmentation, the similarity metric is introduced. The calculation methods of Euclidean distance, Manhattan single-channel distance and Manhattan multi-channel distance are as follows:

D(z,α)=ǁz-αǁ=[(z _R -α _R ) ² +(z _G -α _G ) ² +(z _B -α _B ) ² ], D(z,α)≤D ₀

D _MR (Z _R ,α _R )=ǀZ _R -α _R ǀ

D _M (z,α)=ǁz-αǁ _L1 =ǀz _R -α _R ǀ+ǀz _G -α _G ǀ+ǀz _B -α _B ǀ, D _M ≤ηǁσ‖L1

Among them, Z is the (R _i , G _i , B _i ) value of the current segmented pixel, α is the color average (R, G, B) of the segmented color area, the subscript is RGB component, D ₀ is the segmentation threshold, ǁσ‖L1 is the standard deviation vector of each of the three channels, and η is the standard deviation coefficient, usually 1.25.

When processing images by color segmentation, the selection of different color sample sets has a significant impact on the final segmentation effect, and different color block segmentation results will be presented, and the difference is large. Taking the discolored sample image under 20ppm _H2S environment as an example, the image is processed by the segmentation algorithm based on Euclidean distance (the segmentation threshold is 1500), and the given color sample sets are black and white set, warm color set, and cold color set, as shown in Figure 4a; the image is processed by the segmentation algorithm based on Manhattan distance (ǁσ‖L1 is taken as 1.25), and the given color sample sets are black and white set, warm color set, and cold color set, as shown in Figure 4b. It is obvious that the color segmentation based on the Manhattan algorithm is more sensitive and the pixel classification is more refined. At the same time, the color segmentation based on Manhattan single channel (given a single channel grayscale color set, a single channel single color set) is shown in Figure 4c. It can be seen that the "Blue" color corresponding part has a better segmentation effect on the discolored sample image. Since the color of the discolored sample itself tends to be darker (white gradually transitions to black and brown), the cold color sample points have a better effect in the selection of color sample sets.

In order to adapt to various complex environments, accurately identify _H2S gas, balance the perceptual consistency of numbers and colors from the perspective of the human eye, and prove the single selectivity of colorimetric/electrical sensors, the present invention introduces the CIE color difference calculation formula (CIE 1976, CIE 1995 or CIE DE2000), and adds the function of color difference calculation, the purpose of which is to replace the work of the colorimeter with large power consumption in the colorimetric experiment, thereby reducing the detection process and calculating the color change degree of the observed sample in real time.

In order to flexibly and quickly measure the color change degree of the colorimetric/electrical sensor under a certain H ₂ S gas concentration, it is placed in a gas chamber equipped with a color change monitoring platform, and the color change of the sensor is monitored in real time through a camera. The initial color of the colorimetric/electrical sensor is defined as the standard color, and an image of the sensor color change is collected once every second as the sample color. In order to ensure the detection effect of the system under different lighting environments as much as possible, the camera captures the image of the sensor color change under the H ₂ S environment of LED fill light, natural light, and dim light at night, as shown in Figure 5, and then uploads the image data to the smart terminal. Using Python as the development language, in the Spyder compilation environment, the RGB pixel mean of the image is calculated, and the RGB value conversion L*a*b* color space expression function written in advance is called. Its core calculation formula (1)(2)(3) is shown below. Because the RGB color space cannot be directly converted to the L*a*b* color space, it is necessary to use the XYZ color space to convert the RGB color space to the XYZ color space, and then convert the XYZ color space to the L*a*b* color space to obtain the Lab numerical expression of all image colors. Finally, the function of the CIE DE2000 color difference calculation formula is called to obtain the color difference ΔE change curve of the sensor as shown in Figure 6.

(1)

(2)

(3)

After image preprocessing, the conversion between image data and digital data is completed. After data cleaning, a data set that can be used to train the gas concentration prediction model is obtained. Subsequently, the multi-layer perceptron algorithm (Multi-Layer FeedForward Neural Networks) is used to predict the unknown concentration of H ₂ S gas. It shows good stability in processing multi-dimensional nonlinear data and can quickly give prediction results. Figure 7a shows the training process of the H ₂ S concentration prediction model designed based on the 3-layer perceptron algorithm. The Adam optimizer is selected during the training process. Its function is simple to implement, computationally efficient, requires less memory, and the parameter update is not affected by the scaling transformation of the gradient.

R = (R ₁ ,R ₂ ......R ₂₀₀ ) ^T ,G = (G ₁ ,G ₂ ......G ₂₀₀ ) ^T ,B = (B ₁ ,B ₂ ......B ₂₀₀ ) ^T is the vector of the RGB mean values of 200 training samples, which is used as the input of the input layer of the three-layer perceptron. X ₁ ,X ₂ ,X ₃ are the three neuron models of the input layer. From the input layer to the first hidden layer, it is a multivariate linear regression process. The initial weights from the input layer to the first hidden layer are set to W ₁ ,W ₂ ,W ₃ . At this time, the input of the neurons in the first hidden layer can be calculated by formula (4). The process from the first hidden layer to the second hidden layer is a nonlinear regression process. Due to its calculation stability and avoiding the gradient vanishing during the stochastic gradient descent optimization calculation, the ReLU function (Rectified Linear Unit) is selected as the activation function. Combined with the activation function, the output of the first hidden layer neurons can be obtained, as calculated by formula (5), and the input of the second hidden layer neurons can be calculated by formula (6). Where m = (100, 50, 20) are the dimensions of the three hidden layers, v ₁ , v ₂ ... v _m are the weight values of each hidden layer to the next layer. Repeated calculations can obtain the output gas concentration prediction result of the output layer, referred to as Yp.

(4)

(5)

(6)

The trained model needs to be further tested to determine whether it can be put into operation in an unknown H ₂ S gas environment. Based on the Manhattan multi-channel distance color segmentation algorithm (where the standard deviation σ vector of each of the three channels is set to 1.25), the cold and warm color sets are selected to process the discolored sample images. The image data processed by the cold and warm color sets are used to train two original models respectively, and the two trained models are named cold model and warm model. A test set is set to input the trained model, and the model immediately gives the gas concentration prediction result. As shown in Figure 7b, it can be intuitively found that the concentration prediction curve of the cold model shows a better fit, while the concentration prediction curve of the hot model has a larger relative deviation from the center line. In order to evaluate the prediction accuracy of the gas prediction model, the determination coefficient R ² is introduced as an indicator to evaluate the quality of the model. The value of R ² fluctuates between [0,1]. The higher the R ² , the closer the predicted gas concentration value is to the true value, the higher the model prediction accuracy is, and the better the performance is. Figures 7c and 7d show the test accuracy of the cold model and the warm model respectively. The accuracy of the hot model is relatively low, only reaching 82.3%, while the accuracy of the cold model can reach up to 99.8%, which is significantly higher than the cold model, showing excellent H ₂ S gas concentration prediction ability. During the model training process, the predicted result Yp and the known data Y are used as (yp _i , y _i ) inputs to calculate the model loss. The loss function is defined to calculate the loss once every 500 iterations, as shown in formula (7), that is, to calculate the mean square error MSE (MeanSquared Error). The loss comparison curves of the two models are shown in Figure 7e. It can be clearly seen that the cold model has less loss and can reach the minimum loss value of 0.00012 faster.

(7)

Therefore, the deep learning colorimetric/electrical dual sensing system of the present invention can realize the short-time safe detection of H ₂ S, and combined with the real-time color change monitoring platform and analysis module, it can continuously observe the color change degree of the sample and accurately determine the H ₂ S gas concentration.

Although the aforementioned color change monitoring platform can continuously observe the color change process, flexibly record the color change of the sensor, and the analysis module can quickly give the gas concentration prediction result for the color change of the sensor. However, as an independent gas sensing system, when facing possible system detection failures (such as damage to the color change monitoring platform, or being affected by very strong light interference), it is necessary to provide a "second solution" for gas detection as a response compensation to ensure the completeness of the _H2S gas sensitive sensing system. Therefore, the deep learning colorimetric/electrical dual sensing system of the present invention also includes a resistance response module, which is connected to the _H2S colorimetric/electrical sensor, and is used to collect, store, and display the resistance response value change signal output by the _H2S colorimetric/electrical sensor in real time to detect _H2S gas. The PET flexible interdigital electrode leads out the wires by soldering, and the resistance test can be carried out synchronously.

Deep learning colorimetric/electrical dual sensing system performance test

In order to verify the single selectivity of the sensor color response, the colorimetric/electrical sensor was placed in a gas chamber, and five gases that may exist in industrial sites, NH ₃ , SO ₂ , NO ₂ , H ₂ , and H ₂ S, were introduced into the gas chamber. The gas concentration was controlled to 100ppm, and the color change monitoring platform was used to observe under the conditions of the same concentration and different gas types. It can be clearly seen that the colorimetric/electrical sensor in the H ₂ S gas environment has obvious color difference changes, but the sensor color in the NH ₃ , SO ₂ , NO ₂ , and H ₂ gas environments has almost no change, and the color difference results calculated by the analysis module also support this view (as shown in Figure 8a), proving that the colorimetric/electrical sensor has a different color change sensitivity to H ₂ S gas than other gases. At the same time, H ₂ S and NH ₃ , SO ₂ , NO ₂ , H ₂ coexisted in pairs (gas concentration was 100ppm) to simulate the color change difference of the sensor under the interference of complex gas environment, as shown in Figure 8b. It is obvious that the influence of non-H ₂ S gas on the color change result of the sensor is almost zero, reflecting the high selectivity of the colorimetric nanofiber gas sensor to H ₂ S gas in a complex gas environment. At the same time, the long-term stability of the sensor was tested. The color response of the sensor at 20ppm was tested and recorded every five days in the first three months, and measured once a month in the next three months. The results showed that the color change performance of the sensor remained stable within 180 days, as shown in Figure 8c.

FIG9a shows the resistance change of the colorimetric/electrical sensor in Example 1 under the environment of increasing H ₂ S gas concentration. It can be seen that as the H ₂ S gas concentration increases from 0 ppm to 100 ppm, the resistance of the colorimetric/electrical sensor decreases in stages. This is because the H ₂ S gas reacts with the H ₂ S sensitive nanofiber membrane, and large PbS particles are precipitated. The original nanofiber gaps are torn and expanded, and the membrane structure becomes loose. The metal sulfide has good conductivity, which reduces the material resistance. And according to FIG9a, it can be obtained that when the resistance observation time of the control sensor in each stage (i.e., under different H ₂ S concentration environments) is within 30s, the change of resistance is stable at this time. Excluding the period when the resistance value remains stable, the resistance response time of the sensor is within 10-15s. In order to analyze the selectivity of the sensor and reflect the difference in the resistance response of the sensor to different gases, it is exposed to 1000ppm NH ₃ , SO ₂ , NO ₂ , H ₂ gas and 100ppm H ₂ S gas environment, and the results are shown in FIG9b. The response value of the colorimetric/electrical sensor to H ₂ S gas is significantly higher than that of other gases. The long-term stability of the sensor's resistance response (H ₂ S concentration is 20ppm) was tested, and the results showed that the sensor's resistance performance remained stable within 180 days, as shown in Figure 9c. Through experimental observation, the color response of the sensor is significantly more flexible. It can be clearly seen from Figure 9d that when the color change approaches stability, the resistance value is still decreasing, which means that the color response speed is faster (response time is less than 4s), and the measurement is faster and the energy consumption is lower. Therefore, in the H ₂ S gas measurement process, the color response result is mainly based on the color response result, and the resistance response result is only used as the automatic interlocking result of the colorimetric/electrical dual sensing system.

Figures 10a and 10d show the differences in color response and resistance response at lower H ₂ S gas concentrations. By enlarging Figures 10b and 10e, it can be seen that the detection limit of the sensing system can be as low as 0.1ppm. Figures 10c and 10f show the differences in color response and resistance response at higher H ₂ S gas concentrations. Obviously, the sensing system can sensitively detect subtle concentration differences with a resolution of up to 0.1ppm.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (7)

1. A deep learning colorimetric/electrical dual sensing system, characterized in that it includes a gas chamber, a H ₂ S colorimetric/electrical sensor, a resistance response module, a color change monitoring platform, an analysis module, and a terminal; The H ₂ S colorimetric/electrical sensor comprises an H ₂ S sensitive nanofiber membrane and a PET flexible interdigital electrode. The H ₂ S sensitive nanofiber membrane is fixed on the PET flexible interdigital electrode by an electrostatic spinning process. The sensitive nanofiber membrane is Pb(CH ₃ CO ₂ ) ₂ /PVA nanofiber. The H ₂ S sensitive nanofiber membrane is a dense mesh structure formed by interlacing nanofibers. The diameter of the nanofibers is 250-400nm. The thickness of the H ₂ S sensitive nanofiber membrane is 0.5-0.7mm. The H ₂ S colorimetric/electrical sensor is located inside the gas chamber and is used for real-time detection of H ₂ S gas to obtain a gas concentration signal, and convert the gas concentration signal into a color change signal and a resistance change signal. The resistance response module is connected to the H ₂ S colorimetric/electrical sensor and is used for real-time acquisition, storage and display of the resistance change signal output by the sensor; The color change monitoring platform is used to collect, store and display the color change signals output by the sensor in real time; The analysis module is loaded on the terminal, analyzes and processes the color change signal collected by the color change monitoring platform, and outputs the H ₂ S concentration based on the analysis and processing result; the analysis module uses the color segmentation algorithm to segment the color change signal collected by the color change monitoring platform, calculates the color difference value ΔE of the color change signal by introducing the CIE color difference formula, and gives the H ₂ S concentration based on the multi-layer perceptron algorithm prediction model; The analysis module uses a color segmentation algorithm based on Euclidean distance or Manhattan distance to segment the color change signal collected by the color change monitoring platform, and calculates the color difference value ΔE of the color change signal by introducing the color difference formula CIE 1976, CIE 1995 or CIE DE2000. The H ₂ S concentration prediction model designed based on the multi-layer perceptron algorithm gives the H ₂ S concentration through measurement; the multi-layer perceptron algorithm is a 3-layer perceptron algorithm; The color segmentation algorithm classifies the pixels of the color image based on the RGB space, that is, extracts the region of the color-changing sample image, and then specifies the average color for the segmented region according to the given color sample set; in order to achieve effective color segmentation, the similarity metric is introduced, and the calculation methods of Euclidean distance, Manhattan single-channel distance and Manhattan multi-channel distance are as follows: ; Where Z is the (R _i , G _i , B _i ) value of the current segmented pixel, α is the color average (R, G, B) of the segmented color area, the subscript is RGB component, D ₀ is the segmentation threshold, is the standard deviation vector of each of the three channels, η is the standard deviation coefficient, usually 1.25; The sample image is preprocessed using the color segmentation algorithm based on Euclidean distance and Manhattan distance. The pixel mean calculation can obtain the (R, G, B) mean vector of the color-changing image, and convert the sample RGB value into a colorimeter suitable for use. Color space, using the CIE color difference formula, calculate the color difference value ΔE of the image to distinguish the chromaticity change of the colorimetric/electrical sensor under different gas concentration environments. The color change from white to dark brown can prove the presence of H ₂ S; When calculating the mean of the color-changing sample image, each pixel of the image is segmented and the value of a single pixel (R _i , G _i , B _i ) is judged. The value of each pixel (R _i , G _i , B _i ) is calculated cyclically and accumulated, thereby obtaining the pixel mean vector (R, G, B) of the color-changing sample, which is calculated by the following formula: ; The terminal is used to display the H ₂ S concentration detection result. 2. The deep learning colorimetric/electrical dual sensing system according to claim 1, wherein the H ₂ S colorimetric/electrical sensor preparation method comprises the following steps: (1) dissolving lead acetate trihydrate, NaF and sodium dodecyl sulfate in deionized water to obtain a lead acetate solution; (2) slowly adding polyvinyl alcohol into the lead acetate solution, and heating and stirring in a water bath to obtain an electrospinning solution; (3) The electrospinning solution is fixed on a PET flexible interdigital electrode in the form of a H ₂ S sensitive nanofiber membrane through an electrospinning process to obtain the H ₂ S colorimetric/electrical sensor. 3. The deep learning colorimetric/electrical dual sensing system according to claim 2 is characterized in that the mass ratio of lead acetate trihydrate: NaF: sodium dodecyl sulfate: polyvinyl alcohol is (0.8-1.2) g: (25-35) mg: (10-20) mg: (2.2-2.6) g, and the heating and stirring temperature in step (2) is 80-90°C and the time is 8-10h. 4. The deep learning colorimetric/electrical dual sensing system according to claim 2 is characterized in that the electrospinning process in step (3) is as follows: using a 20, 22 or 24 gauge metal needle, applying a static voltage of 20±1.5 kV at the needle, the propulsion speed of the microinjection pump is 1±0.2 mL/h, and the receiver is 10±0.5 cm away from the needle; the ambient humidity during the spinning process is 35%±5%, the temperature is 25±2°C; and the spinning time is 8~10 h. 5. The deep learning colorimetric/electrical dual sensing system according to claim 1 is characterized in that the color change monitoring platform includes a fill light unit, a collection unit, a storage unit and a display unit, the fill light unit is an LED fill light, the collection unit is an integrated camera module, which is used to collect the color change signal output by the sensor in real time, the storage unit is used to store the collected color change signal, and the display unit is used to display the color change signal. 6. The deep learning colorimetric/electrical dual sensing system according to claim 1 is characterized in that the deep learning colorimetric/electrical dual sensing system uses a resistance response module to form a colorimetric-electrical dual-function integrated sensing system, which can simultaneously detect the resistance response value change of the _H2S colorimetric/electrical sensor and then detect _H2S gas, and can avoid the detection work being stopped in strong light interference and dark environment. 7. The deep learning colorimetric/electrical dual sensing system according to claim 1 is characterized in that the detection range of the deep learning colorimetric/electrical dual sensing system is 0-100ppm, the detection limit and resolution are 0.1ppm, the colorimetric response time is less than 4s, and the performance remains stable within 180 days, and it can work normally even in strong light interference and dark scenes.