Method and system for detecting iron in down feather based on spectral analysis

Abstract

The invention relates to the technical field of down feather detection, in particular to a method and system for detecting iron in down feather based on spectral analysis. The method comprises the following steps: pretreating to-be-detected down feather to obtain a to-be-detected down feather sample; performing dual-mode spectrum collaborative excitation on the down feather sample to be detected, and performing signal acquisition to obtain detection data of the down feather sample to be detected; performing background correction operation on the detection data of the down feather sample to be detected to obtain correction data of the down feather sample to be detected; obtaining an iron detection result of the down feather sample to be detected according to the correction data of the down feather sample to be detected; therefore, the iron in the down feather can be accurately detected.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of down detection, and particularly relates to a method and system for detecting iron in down based on spectral analysis. BACKGROUND

[0002] Down is a natural thermal material, and its safety directly affects the health of consumers. In the production, storage or transportation of down, iron impurities may be introduced due to contact with metal equipment (such as sorting machines and dryers), environmental corrosion or human intervention. Excessive iron can cause down to oxidize and discolor (such as yellowing), catalyze the growth of microorganisms, and even cause allergies or inflammation through skin contact. Therefore, domestic and foreign standards (such as GB / T 17685-2016 "Down Feathers" and FZ / T 81002-2016 "Washed Down Feathers") usually stipulate that the iron content in down should be less than a certain limit value (such as 0.3 mg / g or lower), and if the threshold is exceeded, it is determined to be unqualified.

[0003] In the prior art, there is a scheme for detecting down by spectral analysis to determine whether the iron content exceeds the standard. However, the uneven dispersion of down suspension is one of the main interference factors affecting the accuracy of spectral detection. Due to the natural agglomeration of down fibers, there may be local differences in fiber density on the surface of the suspension (such as fiber agglomerate protrusions or sparse area depressions). If the laser focus is fixed at an initial position, when the light spot falls on the fiber agglomerate, the detection signal may be too high due to excessive accumulation of fibers; on the contrary, if the light spot falls in the sparse area, the signal may be too low due to insufficient fibers; this may eventually cause the detection result to deviate significantly from the actual value. In addition, in the prior art, when the sampling image is used to locate the down in the down suspension, the micro features of the down suspension are significantly different due to differences in sample source (duck down / goose down) and storage conditions, resulting in poor accuracy of traditional image positioning schemes and affecting accurate positioning of the down in the down suspension. SUMMARY

[0004] To solve the above technical problems, the present application provides a method and system for detecting iron in down based on spectral analysis to solve the problems in the prior art.

[0005] The present application provides a method for detecting iron in down based on spectral analysis, comprising the following steps: S1: pretreating the down to be detected to obtain a down sample to be detected; S2: performing dual-mode spectral cooperative excitation on the down sample to be detected and collecting signals to obtain detection data of the down sample to be detected; An adaptive focusing step based on image feedback is used to compensate for focusing deviation caused by uneven sample dispersion or microfluidic chip surface fluctuation in real time. Specifically, a CMOS camera is used to capture microscopic image data of the sample pool surface in real time, and the image data is transmitted to an image processing unit through a USB 3.0 interface. An image acquisition unit calculates the actual focusing position of the sample to be detected based on a hybrid algorithm of an edge detection algorithm and a gray center of gravity method. The edge detection algorithm is an improved Canny operator algorithm, and the improved Canny operator algorithm uses an adaptive threshold to realize edge detection of the microscopic image data. Specifically, Sa: calculate the gradient amplitude of each pixel point of the microscopic image data and generate a gradient histogram; Sb: count the peak and valley structure and cumulative distribution of the gradient histogram; Sc: determine the number and value of the Canny operator threshold according to the peak and valley structure and cumulative distribution. S3: performing background correction on the detection data of the sample to be detected to obtain corrected data of the sample to be detected; S4: obtaining the iron detection result of the sample to be detected according to the corrected data of the sample to be detected.

[0006] Preferably, in the Sb, the peak and valley structure includes two indicators of a main peak position and a peak width. The main peak position μ is a concentrated interval of the gradient amplitude in the microscopic image data, which is used to reflect the dominant contrast of the fiber edge in the microscopic image data. The peak width σ is the standard deviation of the main peak in the microscopic image data, which is used to reflect the dispersion degree of the gradient distribution. The cumulative distribution is reflected by a background noise level N indicator. The background noise level N is the pixel ratio of the low gradient interval, which is used to reflect the purity of the sample to be detected.

[0007] Preferably, the Sc is specifically: If the peak width σ of the gradient histogram of the microscopic image data is ≤15 and the background noise level N is ≤0.3, a double-threshold detection algorithm is used. And the calculation formula of the double threshold is: ; ; In the formula, T high is the high threshold of the improved Canny operator algorithm, and T low is the low threshold of the improved Canny operator algorithm.

[0008] If the peak width σ of the gradient histogram of the microscopic image data is >15 and the background noise level N is ≤0.3, a three-threshold detection algorithm is used. And the calculation formula of the three thresholds is: wherein, T high is a high threshold value of the improved Canny operator method, T mid is a medium threshold value of the improved Canny operator method, the T low is a low threshold value of the improved Canny operator method.

[0009] If the background noise level N of the gradient histogram of the microscopic image data is greater than 0.3, at this time, the suspension of the sample contains a large amount of bubbles or oil, the gradient distribution low gradient interval proportion is high, and the background noise interference is strong, then a four-threshold detection algorithm is used. And the calculation formula of the four thresholds is: wherein, T high is a high threshold value of the improved Canny operator method, T mid is a medium threshold value of the improved Canny operator method, the T low is a low threshold value of the improved Canny operator method, and the T low is an extremely low threshold value of the improved Canny operator method.

[0010] Preferably, the S1 is specifically: S1.1: sample weighing and initial dispersion are performed on the down-filled down to obtain an initial dispersion suspension; S1.2: a surfactant is added to the initial dispersion suspension for auxiliary dispersion to obtain an auxiliary dispersion suspension; S1.3: the auxiliary dispersion suspension is transferred to a sample pool of a microfluidic chip, and magnetic stirring is performed to obtain a down-filled down sample to be detected.

[0011] Preferably, the down-filled down sample to be detected is subjected to dual-mode spectrum synergistic excitation, and signal acquisition is performed to obtain detection data of the down-filled down sample to be detected, which is specifically: ​​​​​​​The dual-wavelength composite laser outputs 488nm laser beam and 785nm laser beam simultaneously, which are combined by a dichroic mirror and focused to the sample cell of the microfluidic chip through the same optical fiber probe; then time-sharing detection is performed, when the 488nm laser beam is turned on, the 785nm laser beam is turned off synchronously, the fluorescence signal is collected by the optical fiber probe and transmitted to the fluorescence spectrometer through the filter set, so as to collect the fluorescence data of the to-be-detected down sample; when the 488nm laser beam is turned off, the 785nm laser beam is turned on synchronously, the Raman signal is collected by the same optical fiber probe and transmitted to the Raman spectrometer through the long-pass filter, so as to collect the Raman spectrum data of the to-be-detected down sample, and the fluorescence collection and the Raman spectrum collection are alternately performed, and the alternating period is 1s.

[0012] Preferably, the S2 further comprises: capturing microimage data of the surface of the sample cell in real time at a rate of 30 frames / s by using a CMOS camera, and transmitting the image data to the image processing unit through a USB3.0 interface; The image processing unit calculates the actual focusing position of the to-be-detected down sample based on a mixed algorithm of edge detection algorithm and gray gravity center method.

[0013] Preferably, the S3 specifically comprises: S3.1: blank background spectrum acquisition and data preprocessing, to obtain preprocessed blank background spectrum data; S3.2: performing dynamic background deduction on the detection data of the to-be-detected down sample based on the preprocessed blank background spectrum data, to obtain corrected data of the to-be-detected down sample.

[0014] Preferably, the dynamic background deduction on the fluorescence data of the to-be-detected down sample based on the preprocessed blank background spectrum data specifically comprises: introducing a scattering light proportion factor k scat performing fluorescence intensity correction, wherein the scattering light proportion factor is the background intensity ratio below 400nm of the preprocessed blank background spectrum data and the detection data of the to-be-detected down sample; according to the scattering light proportion factor k scat performing dynamic background deduction on the detection data of the to-be-detected down sample, to obtain corrected fluorescence data of the to-be-detected down sample; The dynamic background deduction on the Raman spectrum data of the to-be-detected down sample based on the preprocessed blank background spectrum data specifically comprises: outside the iron characteristic peak interval, selecting a region with obvious matrix characteristic peak to fit polynomial baseline data B matrix ( λ);background deduction is performed in the Raman spectrum data of the to-be-detected down sample, to obtain corrected Raman spectrum data of the to-be-detected down sample; Preferably, S4 is specifically: S4.1: extracting fluorescence features and Raman features in the corrected data of the to-be-detected down sample; S4.2: establishing a deep learning model architecture; S4.3: training the deep learning model architecture; S4.4: inputting the fluorescence features and Raman features of the to-be-detected down sample into the deep learning model architecture, to obtain an iron detection result of the to-be-detected down sample.

[0015] According to another aspect of the present application, a down iron detection system based on spectral analysis is provided, which adopts the down iron detection method based on spectral analysis of embodiment 1, and the system comprises: A sample pretreatment module is configured to pretreat the to-be-detected down, to obtain a to-be-detected down sample; A detection data acquisition module is configured to perform dual-mode spectral synergistic excitation on the to-be-detected down sample, and perform signal acquisition, to obtain detection data of the to-be-detected down sample.

[0016] A detection data correction module is configured to perform background correction on the detection data of the to-be-detected down sample, to obtain corrected data of the to-be-detected down sample; An iron detection module is configured to obtain an iron detection result of the to-be-detected down sample according to the corrected data of the to-be-detected down sample.

[0017] The embodiments of the present application have the following technical effects: The present application pretreats the to-be-detected down, to obtain a to-be-detected down sample; performs dual-mode spectral synergistic excitation on the to-be-detected down sample, and performs signal acquisition, to obtain detection data of the to-be-detected down sample; performs background correction on the detection data of the to-be-detected down sample, to obtain corrected data of the to-be-detected down sample; and obtains an iron detection result of the to-be-detected down sample according to the corrected data of the to-be-detected down sample; thereby realizing accurate detection of iron in down; Meanwhile, according to the characteristics of the down suspension liquid, the adaptive threshold detection scheme is adopted, the threshold value is automatically matched through the gradient distribution characteristics, manual intervention is not needed, all scene samples such as high uniformity, high aggregation-sparse mixture and high noise are covered, the double threshold simplifies the calculation, avoids the weak edge missing caused by excessive subdivision, the three thresholds retain the medium contrast edges and improve the fiber coverage, the four thresholds + strict connection rules have a noise false detection rate <0.5% (traditional method >5%), the gradient distribution characteristics of the down suspension liquid are dynamically analyzed, the threshold value is intelligently matched, the noise interference is significantly inhibited while the integrity of the real edge is ensured. The scheme solves the limitation of the traditional fixed threshold method, and provides more accurate and robust down suspension liquid image positioning technology support for the rapid detection of iron in down. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed in the specific embodiments or the prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0019] Figure 1 is a flowchart of a down iron detection method based on spectral analysis provided by an embodiment of the present application; Figure 2 is a flowchart of edge detection of the microscopic image data achieved by using the adaptive threshold provided by an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical scheme and advantages of the present application more clear, the technical scheme of the present application will be clearly and completely described as follows. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of the present application.

[0021] Embodiment 1, attached Figure 1 shows a flowchart of a down iron detection method based on spectral analysis, as shown in the attached Figure 1 The down iron detection method based on spectral analysis includes the following steps: S1: pretreating the down to be detected to obtain a down sample to be detected; Traditional methods of homogenization of down samples (such as drying, acid digestion) can destroy the original morphology of the sample or introduce interference, while simple mechanical stirring or ultrasonic treatment can cause down fibers to agglomerate, affecting the uniformity of subsequent spectral detection. In view of the above situation, a rapid pretreatment method based on the synergistic effect of microfluidic chip and surfactant is proposed to realize the uniform dispersion of down samples under the condition of non-destructive and non-chemical pollution, and to ensure the repeatability and accuracy of iron detection.

[0022] Specifically, the S1 is specifically: S1.1: sample weighing and initial dispersion of the down to be detected to obtain an initial dispersion suspension; According to the fluffiness of the down sample, 0.1-0.5g of the down to be detected is weighed; If the sample fluffiness is high (such as down raw material without compression), the upper limit 0.5g is taken; if the sample has been compressed (such as down pillow core disassembled material), the lower limit 0.1g is taken.

[0023] The weighed down to be detected is transferred to the sample tube of the micro vortex mixer, 5mL of deionized water (temperature 25±1℃, pH 6.5-7.5) is added, and vortex mixing is carried out at a speed of 300rpm for 2 minutes; In this step, the partial agglomeration between down fibers is broken by gentle mechanical shearing force, so that the sample is preliminarily dispersed into a suspended state.

[0024] Among them, the temperature of the deionized water is 25±1℃, and the pH value is 6.5-7.5.

[0025] S1.2: adding a surfactant to the initial dispersion suspension for auxiliary dispersion to obtain an auxiliary dispersion suspension; 0.1% TritonX-100 surfactant is added to the initial dispersion suspension, and vortex mixing is continued at 300rpm for 1 minute; TritonX-100 is a non-ionic surfactant, the hydrophobic end of which can adsorb the grease on the surface of down fibers, and the hydrophilic end can reduce the surface tension of water, promoting the dispersion of fibers in water. The surface of down fibers often adheres to natural grease (such as wax) and processing residues (such as silicone oil), which can increase the friction between fibers, leading to agglomeration. TritonX-100 can "wrap" grease molecules, weaken the adhesion between fibers, and form hydrogen bonds with water molecules, so that the fibers are more easily dispersed in the water phase.

[0026] S1.3: transferring the auxiliary dispersion suspension to the sample pool of the microfluidic chip and performing magnetic stirring to obtain the down sample to be detected; The vortex-mixed suspension is transferred to a sample pool of a microfluidic chip based on the principle of "restricted flow", which can force the arrangement of down feather fibers in a single layer to avoid aggregation; a magnetic stirrer with a diameter of 3 mm is placed in a microcavity below the microfluidic chip, and the external magnetic stirrer is stirred at a speed of 500 rpm for 1 minute. The vortex generated by magnetic stirring can further break up the fiber aggregates, while avoiding damage to the fibers caused by direct mechanical shearing.

[0027] The channel size of the microfluidic chip is 50 μm x 100 μm, and the material is PDMS.

[0028] In this step, the combination of micro-vortex and magnetic stirring avoids the damage of ultrasonic or high-speed shearing to the iron form, ensuring the authenticity of the detection results. 0.1% Triton X-100 reduces the surface tension without introducing detectable chemical residues. The design of micron-scale channels and transparent materials realizes real-time monitoring of dispersion uniformity and efficient homogenization, solving the problem of "uneven dispersion → large detection error" in traditional methods.

[0029] S2: performing dual-mode spectral synergistic excitation on the to-be-detected down feather sample and collecting signals to obtain detection data of the to-be-detected down feather sample.

[0030] In this step, a dual-mode spectral detection module, i.e., a detection module integrating a fluorescence detection unit and a surface-enhanced Raman (SERS) detection unit, is used. Through synchronous excitation-time detection and adaptive focusing technology, efficient synergistic collection of fluorescence and Raman signals is realized, thereby obtaining the detection data of the to-be-detected down feather sample.

[0031] Specifically, S2 is specifically: A dual-wavelength compound laser is used to simultaneously output 488 nm laser beams and 785 nm (SERS excitation) laser beams. After being combined by a dichroic mirror (Dichroic Mirror), the laser beams are focused to the sample pool of the microfluidic chip through the same optical fiber probe. The 488 nm laser beams are used for fluorescence excitation, and the 785 nm laser beams are used for SERS excitation. The diameter of the optical fiber probe is 50 μm.

[0032] Then, time detection is performed. When the 488 nm laser beams are turned on, the 785 nm laser beams are turned off at the same time. After the fluorescence signal is collected by the optical fiber probe, it is transmitted to the fluorescence spectrometer through the filter set, thereby collecting the fluorescence data of the to-be-detected down feather sample.

[0033] The cut-off wavelength of the filter set is 500 nm, which is used to block the interference of 785 nm laser scattering, and the detection range of the fluorescence spectrometer is 500-700 nm, and the integration time is 100 ms.

[0034] When the 488 nm laser beam is turned off, the 785 nm laser beam is turned on synchronously, and after the Raman signal is collected by the same optical fiber probe, it is transmitted to the Raman spectrometer through a long-pass filter, thereby collecting the Raman spectrum data of the sample to be detected.

[0035] The cut-off wavelength of the long-pass filter is 750 nm, which is used to block the interference of 488 nm laser scattering; the detection range of the Raman spectrometer is 1000-1800 cm -1 , and the integration time is 500 ms.

[0036] The fluorescence acquisition and Raman spectrum acquisition are alternately performed, and the alternating period is 1 s.

[0037] In this step, a synchronous excitation-time-sharing detection mode is adopted, and the synchronous spatial acquisition of the dual-mode signals is realized through the same optical fiber probe, thereby eliminating the errors caused by sample displacement in the traditional separate optical path.

[0038] Furthermore, the uneven dispersion of the down feather suspension is one of the main interference factors affecting the spectral detection accuracy. Due to the natural aggregation of down feather fibers, there may be local fiber density differences on the surface of the suspension (such as fiber aggregation protrusions or sparse area depressions). If the laser focus is fixed at an initial position, when the light spot falls on the fiber aggregation, the detection signal may be too high due to excessive fiber accumulation; on the contrary, if the light spot falls in the sparse area, the signal may be too low due to insufficient fibers. This dynamic shift between the light spot and the sample surface can cause significant fluctuations in the repeated detection results of the same sample, seriously affecting the accuracy and reliability of the iron content detection. In view of the uneven dispersion of the down feather suspension, an adaptive focusing step based on image feedback is introduced in this step to compensate for the focusing deviation caused by the uneven dispersion of the sample or the fluctuation of the microfluidic chip surface in real time, thereby ensuring the spatial consistency of the dual-mode spectral signals.

[0039] Specifically: A CMOS camera is used to capture microscopic image data of the sample pool surface at a rate of 30 frames per second, and the image data is transmitted to the image processing unit through a USB 3.0 interface; The CMOS camera is integrated directly below the microfluidic chip and is vertically 500 pm away from the surface of the sample cell, and is fixed by a support to ensure that the optical axis is coaxial with the laser probe; the technical parameters of the CMOS camera are: resolution: 2 million pixels (1600*1200 pixels); frame rate: 30 frames / s to meet the real-time dynamic monitoring requirement; field of view: 200*200 pm to correspond to the pixel resolution of 0.125 pm / pixel, which can identify the micron-level morphological changes of the fiber agglomerates; exposure time: 1 ms to avoid image blur caused by the flow of the down feather suspension.

[0040] The image processing unit calculates the actual focusing position of the down feather sample to be detected based on a mixed algorithm of an edge detection algorithm and a gray center of gravity method; The edge detection algorithm is an improved Canny operator algorithm, and the Gaussian filter parameters of the improved Canny operator algorithm are: σ=1.0, which is used to balance noise suppression and edge retention; and the edge width is set to a single pixel level to avoid interference of false edges.

[0041] More importantly, the improved Canny operator method uses an adaptive threshold to realize edge detection of the microscopic image data, The micro features (such as fiber density, agglomeration degree, and oil content) of the down feather suspension are significantly different due to differences in sample sources (duck down / geese down) and storage conditions. The traditional fixed threshold (such as a double threshold or a triple threshold) segmentation method is difficult to adapt to such diversity: high agglomeration samples: the edge gradient amplitude of the fiber agglomerates is high and concentrated (a high threshold is needed to distinguish strong edges), but the fiber gradient in the sparse area is low (fixed high threshold is easy to cause edge missing in the sparse area); fat samples: the suspension background is clean and the noise interference is small (the low threshold can be relaxed to retain more weak edges), and the high oil sample background is complex (the low threshold needs to be tightened to suppress noise); therefore, the embodiment proposes an adaptive threshold detection algorithm, which dynamically determines the number and specific value of the threshold by analyzing the gradient distribution characteristics of the down feather suspension, to realize precise edge segmentation of the “one sample one strategy”.

[0042] As shown in the accompanying Figure 2 The adaptive threshold is used to realize edge detection of the microscopic image data, and specifically: Sa: Calculate the gradient amplitude of each pixel point of the microscopic image data and generate a gradient histogram; The horizontal axis of the gradient histogram is the gradient amplitude G , and the vertical axis is the pixel point number ratio.

[0043] Sb: Statistics of the peak and valley structure and cumulative distribution of the gradient histogram; The peak-valley structure comprises two indexes of main peak position and peak width; the main peak position μ is a main concentrated interval of gradient amplitude in the microscopic image data, and is used for reflecting a dominant contrast of the fiber edge in the microscopic image data; and the peak width σ is a standard deviation of the main peak in the microscopic image data, and is used for reflecting a discrete degree of the gradient distribution and being related to the uniformity of the fiber aggregation. The cumulative distribution is reflected by a background noise level N index, the background noise level N is a pixel proportion of a low gradient interval (G<20), and is used for reflecting the purity of the suspension.

[0044] Sc: determining the number and values of the threshold of the Canny operator method according to the peak-valley structure and the cumulative distribution; The Sc is specifically as follows: If the peak width σ of the gradient histogram of the microscopic image data is less than or equal to 15 and the background noise level N is less than or equal to 0.3, a double-threshold detection algorithm is adopted. The calculation formula of the double threshold is as follows: ; ; In the formula, T high is a high threshold of the improved Canny operator method, and T low is a low threshold of the improved Canny operator method.

[0045] If the peak width σ of the gradient histogram of the microscopic image data is greater than 15 and the background noise level N is less than or equal to 0.3, a three-threshold detection algorithm is adopted. The calculation formula of the three threshold is as follows: ; ; ; In the formula, T high is a high threshold of the improved Canny operator method, T mid is a middle threshold of the improved Canny operator method, and T low is a low threshold of the improved Canny operator method.

[0046] If the background noise level N of the gradient histogram of the microscopic image data is greater than 0.3, at this time, the suspension of the sample contains a large amount of bubbles or oil, the proportion of the low gradient interval of the gradient distribution is high, and the background noise interference is strong, a four-threshold detection algorithm is adopted. The calculation formula of the four threshold is as follows: ; ; ; ; In the formula, T high is the high threshold value of the improved Canny operator method, T mid is the medium threshold value of the improved Canny operator method, the T low is the low threshold value of the improved Canny operator method, the T low is the very low threshold value of the improved Canny operator method; At the same time, the connection rule is that for weak edges (T vlow <G≤T low ), it needs to be connected with medium / high edges at the same time and form a chain of ≥5 pixels to further eliminate noise.

[0047] The present application covers all scene samples such as high uniformity, high clustering-sparse mixing and high noise through the adaptive threshold detection scheme, automatic matching threshold strategy through gradient distribution characteristics, without manual intervention, and double threshold simplifies calculation, avoids weak edge missing caused by excessive subdivision; three threshold values retain medium contrast edges, improve fiber coverage; four threshold values + strict connection rule, noise false detection rate <0.5% (traditional method >5%); through dynamic analysis of the gradient distribution characteristics of the down feather suspension, intelligent matching of threshold value and number, while ensuring the integrity of the real edge, significantly suppresses noise interference. The scheme solves the limitations of the traditional fixed threshold method "one size fits all", and provides more accurate and robust image positioning technology support for rapid detection of iron in down.

[0048] After edge detection is performed on the microscopic image data, the actual focusing position of the down sample to be detected is calculated through the gray center method, and then the Z-axis height of the optical fiber probe is dynamically adjusted (step motor precision ±1μm) to ensure that the laser is always focused on the most dense area of the sample, i.e., the uniform distribution of down fibers.

[0049] S3: performing background correction operation on the detection data of the down sample to be detected to obtain corrected data of the down sample to be detected; The S3 is specifically: S3.1: blank background spectrum acquisition and data preprocessing to obtain preprocessed blank background spectrum data; Before each detection, 50μL of deionized water containing 0.1% TritonX-100 surfactant is injected into the microfluidic chip, and then spectrum acquisition is performed; Among them, the fluorescence channel is excited by 488nm laser, and the emission spectrum of 500-700nm is collected, the integration time is 100ms, and the average of continuous 3 times is taken; the Raman channel is excited by 785nm laser, and the emission spectrum of 1000-1800cm -1Raman spectrum, integral time 500 ms, 3 times repeated and averaged to obtain blank background spectrum.

[0050] Then, the blank background spectrum is baseline corrected. The blank background spectrum is polynomial fitted to subtract background noise, wherein the polynomial fitting is a 3-order polynomial fitting.

[0051] S3.2: Based on the pre-processed blank background spectrum data, the detection data of the to-be-detected down sample is dynamically background-subtracted to obtain the corrected data of the to-be-detected down sample. Wherein, based on the pre-processed blank background spectrum data, the fluorescence data of the to-be-detected down sample is dynamically background-subtracted, specifically as follows: The fluorescence signal of the down suspension is easily interfered by particle scattering, therefore, a scattering light proportion factor k is introduced scat The fluorescence intensity is corrected, and the scattering light proportion factor is the background intensity ratio below 400 nm of the pre-processed blank background spectrum data and the detection data of the to-be-detected down sample. The calculation formula is: ; In the formula, ∑ 400nm represents the integral intensity below 400 nm, F sample is the fluorescence data of the to-be-detected down sample, and F blank is the pre-processed blank background fluorescence data.

[0052] According to the scattering light proportion factor k scat The detection data of the to-be-detected down sample is dynamically background-subtracted to obtain the corrected fluorescence data of the to-be-detected down sample.

[0053] The specific formula is: ; In the formula, F corrected is the corrected fluorescence data of the to-be-detected down sample.

[0054] Wherein, based on the pre-processed blank background spectrum data, the Raman spectrum data of the to-be-detected down sample is dynamically background-subtracted, specifically as follows: The organic matters (such as protein, oil) in the down matrix will produce wide peaks (such as protein amide I band ~1650 cm -1 , oil C-H bond ~2900 cm -1 ) which overlap with the characteristic peaks of iron (such as Fe-O bond ~600 cm -1 ) in the Raman spectrum. Dynamic baseline fitting is adopted to eliminate background interference. In the iron characteristic peak interval (such as 200-800cm -1 ), the region with obvious matrix characteristic peaks (such as 1500-1700cm -1 ) is selected to fit the polynomial baseline data B matrix ( λ ); The specific formula is: ; In the formula, a 0, a 1, a 2, a 3 are coefficients, and λ is wavelength.

[0055] The background is deducted in the Raman spectrum data of the to-be-detected down sample, and the corrected Raman spectrum data of the to-be-detected down sample is obtained; The specific formula is: ; In the formula, R baseline_corrected is the corrected Raman spectrum data of the to-be-detected down sample, R sample is the Raman spectrum data of the to-be-detected down sample, B matrix ( λ ) is the polynomial baseline data.

[0056] S4: Obtain the iron detection result of the to-be-detected down sample according to the corrected data of the to-be-detected down sample.

[0057] The S4 is specifically: S4.1: Extract fluorescence characteristics and Raman characteristics from the corrected data of the to-be-detected down sample; Wherein, the fluorescence characteristics are the maximum intensity F peak of the fluorescence peak at 520nm and the half-width F width of the fluorescence peak at 500-700nm; the Raman characteristics are the peak area R -1 of the Fe-O bond Raman peak at 600cm area and the offset R -1 of the 600cm shift peak position relative to the standard value.

[0058] S4.2: Establish a deep learning model architecture; Wherein, the deep learning model architecture is a multi-layer perception (MLP) and attention mechanism fusion architecture; specifically: Input layer, used to receive 4-dimensional input features (F peak , Fwidth , R area , R shift ).

[0059] The hidden layer (MLP part) comprises: a first hidden layer: 16 neurons, an activation function ReLU (Rectified Linear Unit), which performs a preliminary nonlinear transformation on the input features; a second hidden layer: 8 neurons, an activation function ReLU, which further extracts high-order features; a Dropout layer: a dropout rate of 0.2, which prevents overfitting.

[0060] The attention mechanism layer is used to apply attention weights to the feature vectors extracted by the hidden layer, and dynamically adjust the contribution of different features to the prediction of iron content.

[0061] The output layer is used to output the iron detection result.

[0062] S4.3: training the deep learning model architecture; Wherein, the loss function of the deep learning model architecture is mean square error (MSE) loss function, the optimizer is Adam (adaptive moment estimation) optimizer, the initial learning rate is 0.001, the batch size is (batchsize) 16, and the training rounds (epochs) are 100.

[0063] S4.4: inputting the fluorescence features and Raman features of the to-be-detected down sample into the deep learning model architecture to obtain the iron detection result of the to-be-detected down sample; Wherein, the iron detection result includes: iron detection pass and iron detection fail.

[0064] Further, the output layer of the deep learning model architecture further comprises a threshold comparison module, which compares the content of the iron in the to-be-detected down sample predicted by the deep learning model architecture with a preset threshold value. When the content is greater than the preset threshold value, the output layer outputs the result of iron detection fail, otherwise, the output layer outputs the result of iron detection pass.

[0065] Embodiment 2, the present application also provides a down feather iron detection system based on spectral analysis, which adopts the down feather iron detection method based on spectral analysis in embodiment 1, and the system comprises: A sample pretreatment module is used for pretreating the to-be-detected down sample to obtain a to-be-detected down sample. A detection data acquisition module is used for dual-mode spectral synergistic excitation of the to-be-detected down sample, signal acquisition, and obtaining the detection data of the to-be-detected down sample.

[0066] The detection data correction module is configured to perform background correction on the detection data of the to-be-detected down sample to obtain corrected data of the to-be-detected down sample. The iron detection module is configured to obtain an iron detection result of the to-be-detected down sample according to the corrected data of the to-be-detected down sample.

[0067] In some embodiments, the electronic device can further include one or more processors and a memory.

[0068] The processor can be a central processing unit (CPU) or other form of processing unit that has data processing and / or instruction execution capabilities, and can control other components in the electronic device to perform desired functions.

[0069] The memory can include one or more computer program products, which can include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM), cache memory, and / or the like. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, and / or the like. One or more computer program instructions can be stored on the computer-readable storage media, and the processor can execute the program instructions to implement the method for detecting iron in down based on spectral analysis according to any of the embodiments of the present application and / or other desired functions. Various contents such as initial external parameters, threshold values, and the like can also be stored in the computer-readable storage media.

[0070] In one example, the electronic device can further include an input device and an output device, which are interconnected through a bus system and / or other forms of connection mechanism (not shown). The input device can include, for example, a keyboard, a mouse, and / or the like. The output device can output various information to the outside, including pre-warning prompt information, braking force, and the like. The output device can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, and / or the like.

[0071] Of course, components such as buses, input / output interfaces, and the like are omitted for simplicity. In addition, the electronic device can include any other appropriate components according to specific application cases.

[0072] In addition to the above method and device, the embodiments of the present application can also be a computer program product including computer program instructions, which, when executed by a processor, cause the processor to implement the functions of the method for detecting iron in down based on spectral analysis provided by any of the embodiments of the present application.

[0073] In addition, the embodiments of the present application can also be a computer readable storage medium having stored thereon computer program instructions, which, when executed by a processor, cause the processor to implement the method for detecting iron in down based on spectral analysis provided by any of the embodiments of the present application.

[0074] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the present application.

Claims

1. A method for detecting iron in down based on spectral analysis, characterized in that, The method comprises the following steps: S1: pretreating the to-be-detected down feather to obtain a to-be-detected down feather sample; S2: performing dual-mode spectrum synergistic excitation on the to-be-detected down feather sample and collecting signals to obtain detection data of the to-be-detected down feather sample; in this step, an adaptive focusing step based on image feedback is adopted to compensate for focusing deviation caused by uneven dispersion of the sample or surface fluctuation of the microfluidic chip in real time; Specifically, a CMOS camera is used to capture microscopic image data of the surface of the sample pool in real time, and the image data is transmitted to an image processing unit through a USB 3.0 interface; the image acquisition unit calculates the actual focusing position of the to-be-detected down feather sample based on a mixed algorithm of an edge detection algorithm and a gray gravity center method; the edge detection algorithm is an improved Canny operator algorithm, and the improved Canny operator method uses an adaptive threshold to realize edge detection of the microscopic image data; specifically, Sa: calculating the gradient amplitude of each pixel point of the microscopic image data and generating a gradient histogram; Sb: counting the peak-valley structure and cumulative distribution of the gradient histogram; Sc: determining the number and value of the threshold of the Canny operator method according to the peak-valley structure and cumulative distribution; S3: performing background correction on the detection data of the to-be-detected down feather sample to obtain corrected data of the to-be-detected down feather sample; S4: obtaining the iron detection result of the to-be-detected down feather sample according to the corrected data of the to-be-detected down feather sample.

2. The method according to claim 1, wherein: in Sb, the peak-valley structure comprises two indexes of a main peak position and a peak width; the main peak position μ is a concentrated interval of the gradient amplitude in the microscopic image data, and is used to reflect the dominant contrast of the fiber edge in the microscopic image data; the peak width σ is a standard deviation of the main peak in the microscopic image data, and is used to reflect the dispersion degree of the gradient distribution; the cumulative distribution is reflected by a background noise level N index, and the background noise level N is a pixel ratio of a low gradient interval, and is used to reflect the purity of the to-be-detected down feather sample.

3. The method according to claim 2, wherein: Sc is specifically: if the peak width σ of the gradient histogram of the microscopic image data is ≤15 and the background noise level N is ≤0.3, a double-threshold detection algorithm is used; the calculation formula of the double threshold is: ； ； In the formula, T high is the high threshold value of the improved Canny operator method, and the T low is the low threshold value of the improved Canny operator method; if the peak width σ of the gradient histogram of the microscopic image data is >15 and the background noise level N is ≤0.3, a three-threshold detection algorithm is used; the calculation formula of the three threshold is: ； ； ； wherein T high is a high threshold value for the improved Canny operator method, T mid is a middle threshold value for the improved Canny operator method, the T low is a low threshold value for the improved Canny operator method; if the background noise level N of the gradient histogram of the microscopic image data is >0.3, at this time, the suspension of the sample contains a large amount of bubbles or oil, the gradient distribution has a high proportion of low gradient interval, and the background noise interference is strong, a four-threshold detection algorithm is used; the calculation formula of the four threshold is: ； ； ； ； wherein T high is a high threshold value for the improved Canny operator method, T mid is a medium threshold value for the improved Canny operator method, the T low is a low threshold value for the improved Canny operator method, the Tv low is an ultra-low threshold value for the improved Canny operator method.

4. The method according to claim 1, wherein: S1 is specifically: S1.1: sample weighing and initial dispersion are performed on the down-filled eiderdown to be detected to obtain an initial dispersion suspension; S1.2: a surfactant is added to the initial dispersion suspension for auxiliary dispersion to obtain an auxiliary dispersion suspension; S1.3: the auxiliary dispersion suspension is transferred to a sample pool of a microfluidic chip, and magnetic stirring is performed to obtain the down-filled eiderdown sample to be detected.

5. The method according to claim 1, wherein: The down-filled eiderdown sample to be detected is subjected to dual-mode spectrum synergistic excitation, and signal acquisition is performed to obtain detection data of the down-filled eiderdown sample to be detected, specifically: A dual-wavelength compound laser is used to simultaneously output a 488 nm laser beam and a 785 nm laser beam, and after beam combination by a dichroic mirror, the laser beams are focused to the sample pool of the microfluidic chip through the same optical fiber probe; then time-sharing detection is performed, the 785 nm laser beam is synchronously turned off when the 488 nm laser beam is turned on, the fluorescence signal is collected by the optical fiber probe, and then transmitted to a fluorescence spectrometer through a filter group, so as to acquire fluorescence data of the down-filled eiderdown sample to be detected; the 488 nm laser beam is turned off when the 785 nm laser beam is turned on, the Raman signal is collected by the same optical fiber probe, and then transmitted to a Raman spectrometer through a long-pass filter, so as to acquire Raman spectrum data of the down-filled eiderdown sample to be detected, and the fluorescence acquisition and the Raman spectrum acquisition are alternately performed, and the alternating period is 1 s.

6. The method according to claim 1, wherein: The S2 further comprises: a CMOS camera is used to capture micro image data of the surface of the sample pool in real time at a rate of 30 frames / s, and the image data is transmitted to an image processing unit through a USB 3.0 interface; The image processing unit calculates the actual focusing position of the down-filled eiderdown sample to be detected based on a mixed algorithm of an edge detection algorithm and a gray gravity center method.

7. The method according to claim 1, wherein: The S3 specifically comprises: S3.1: blank background spectrum acquisition and data preprocessing are performed to obtain preprocessed blank background spectrum data; S3.2: dynamic background deduction is performed on the detection data of the down-filled eiderdown sample to be detected based on the preprocessed blank background spectrum data to obtain corrected data of the down-filled eiderdown sample to be detected.

8. The method according to claim 7, wherein: The dynamic background deduction performed on the fluorescence data of the down-filled eiderdown sample to be detected based on the preprocessed blank background spectrum data specifically comprises: Introducing a scattering light proportion factor k scat The fluorescence intensity is corrected, and the scattering light proportion factor is the background intensity ratio of the pretreated blank background spectrum data and the detection data of the down sample to be detected below 400 nm. According to the scattered light proportion factor k scat The detection data of the to-be-detected down sample is subjected to dynamic background deduction to obtain corrected fluorescence data of the to-be-detected down sample. The dynamic background deduction performed on the Raman spectrum data of the down-filled eiderdown sample to be detected based on the preprocessed blank background spectrum data specifically comprises: In the interval outside the iron characteristic peak, the area with obvious matrix characteristic peak is selected to fit the polynomial baseline data B matrix ( λ ) background subtraction is performed in the Raman spectrum data of the to-be-detected down sample to obtain the corrected Raman spectrum data of the to-be-detected down sample.

9. The method according to claim 1, wherein: The S4 specifically comprises: S4.1: fluorescence features and Raman features are extracted from the corrected data of the down-filled eiderdown sample to be detected; S4.2: a deep learning model architecture is established; S4.3: training the deep learning model architecture; S4.4: inputting the fluorescence feature and the Raman feature of the to-be-detected down sample into the deep learning model architecture to obtain an iron detection result of the to-be-detected down sample.

10. A system for detecting iron in down based on spectral analysis, characterized by, The system adopts the method for detecting iron in down based on spectral analysis according to any one of claims 1-9, and the system comprises: a sample pretreatment module configured to pretreat the to-be-detected down to obtain a to-be-detected down sample; a detection data acquisition module configured to perform dual-mode spectral synergistic excitation on the to-be-detected down sample, and perform signal acquisition to obtain detection data of the to-be-detected down sample; a detection data correction module configured to perform background correction on the detection data of the to-be-detected down sample to obtain corrected data of the to-be-detected down sample; an iron detection module configured to obtain an iron detection result of the to-be-detected down sample according to the corrected data of the to-be-detected down sample.

Images