Raman spectrum-based luohanguo agricultural residue fluorescence background suppression imaging method

Abstract

The present application relates to the technical field of intelligent sensor, specifically to a Raman spectrum-based monk fruit agricultural residue fluorescence background suppression imaging method, which comprises the following steps: recognizing the gap between the hairy tissues of monk fruit by using a distance sensor, locking a three-dimensional pixel coordinate matrix and extracting local curvature feature mapping as a capillary pressure gradient field, guiding noble metal nano-probes to enrich at the root of the gradient field to construct a local enhanced hot spot, and outputting an enhanced signal stream; adopting frequency shift excitation combined with polarization differential phase detection, using polarization difference to strip scattering interference caused by the hairy tissues, and performing differential processing through image subtraction algorithm to obtain an agricultural residue differential image; and establishing an attenuation correction model based on surface topography and normal angle to perform pixel-level gain compensation on the signal and restore the real distribution concentration of agricultural residues. The present application is suitable for agricultural product detection scenes with complex hairy structure and strong fluorescence background, and can significantly improve the signal sensitivity, spatial imaging resolution and quantitative analysis capability of agricultural residue detection in a strong interference environment.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensor technology, specifically to a method for suppressing the background fluorescence of pesticide residues in monk fruit based on Raman spectroscopy. Background Technology

[0002] The surface of monk fruit, due to its complex composition of natural flavonoids and dense villi, presents a significant technical bottleneck when detected using conventional optical methods. In the suppression of strong background interference, the natural organic components in the shell of monk fruit produce an extremely strong autofluorescence background under laser excitation, the intensity of which is usually 3 to 5 orders of magnitude higher than the Raman scattering signal of pesticide residue molecules. Existing narrowband filtering or differential smoothing algorithms are prone to causing weak signals to be submerged when dealing with such high signal-to-noise ratio differences, making it difficult to achieve accurate detection of trace pesticide residues.

[0003] In cases of signal distortion caused by dynamic focal drift, the monk fruit (Luo Han Guo) is an irregular ellipsoid, and the distance between the target surface and the detector changes in real time during high-speed rolling in the pipeline. Existing technologies mostly use laser scanning with a fixed focal length. When the surface deviates from the focal point, the energy density decays rapidly, and due to the change in the spherical reflection angle, the collected spectral information is mixed with a large amount of spatial speckle noise, making it impossible to obtain continuous and stable imaging data.

[0004] Due to limitations in occlusion and perspective, the velvety structure on the surface of monk fruit creates a microscopic occlusion effect, and conventional image detection can only capture surface information. During pesticide penetration, residues accumulate at the roots of the velvet or in the shallow epidermis of the fruit shell. Existing visual solutions lack penetrating detection capabilities and cannot perform longitudinal depth verification of the material phases under complex curved surfaces without damaging the fruit. Therefore, a technical solution is needed that utilizes three-dimensional spatial coordinates to drive capillary enrichment enhancement, coupled with frequency-shifting polarization differential background suppression and morphology adaptive attenuation correction, to address the problems of strong fluorescence masking, diffuse scattering interference, and spherical geometric attenuation under complex capillary backgrounds in multiple dimensions.

[0005] To address this, a method for suppressing the background fluorescence of pesticide residues in monk fruit based on Raman spectroscopy is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a method for suppressing the fluorescence background of pesticide residues in monk fruit based on Raman spectroscopy. By driving capillary enrichment enhancement and polarization difference background suppression through three-dimensional spatial coordinate data, a differential image stream of pesticide residues is generated and fused with curvature gradient. An attenuation correction model is established to perform pixel compensation on the differential image stream of pesticide residues, thereby achieving accurate imaging of pesticide residue concentration in monk fruit.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for suppressing the background fluorescence of pesticide residues in monk fruit based on Raman spectroscopy includes: The distribution data of hair tissue in the test area is obtained by using a distance sensor, and the spatial filtering characteristics of the confocal optical path are combined to drive the light beam through the gap in the test area and lock the three-dimensional pixel coordinate matrix. Based on the three-dimensional pixel coordinate matrix, the local curvature distribution features are extracted and mapped to the capillary pressure gradient field. The capillary action is used to promote the enrichment of noble metal nanoprobes, constructing a local enhanced hot spot. The Raman scattering signal is enhanced by the local electromagnetic field, and the enhanced Raman signal stream of pesticide residue components is output. Using the frequency-shifting excitation method, the laser temperature is controlled to output lasers of different wavelengths to irradiate the area under test. Based on the depolarization physical characteristics of the fuzzy fiber, polarization phase detection is performed on the enhanced Raman signal stream for different wavelengths of laser. The light intensity data under orthogonal polarization states is captured by the area array sensor. The polarization difference method is executed to remove scattering interference, obtain the pesticide residue signal matrix, and perform differential processing by the image subtraction algorithm to output the pesticide residue differential image stream. By using a distance sensor to extract the curvature gradient of the area to be measured, and combining it with the pesticide residue differential image stream, an attenuation correction model for the Raman signal is established. Pixel compensation is then performed on the pesticide residue differential image stream to complete pesticide residue concentration imaging.

[0008] Preferably, the process of locking the three-dimensional pixel coordinate matrix includes: using a distance sensor to acquire the distribution data of the hairy tissue in the area to be tested and identifying the connecting gaps through which the light beam can pass; driving the light beam through the connecting gaps and using the spatial pinholes in the optical path to perform spatial filtering, filtering out astigmatism and focusing on the surface of the fruit peel; reading the scanning state parameters at the time of focusing in real time, and mapping the scanning state parameters to three-dimensional physical space coordinates based on the geometric projection model, thereby locking the three-dimensional pixel coordinate matrix.

[0009] Preferably, the process of outputting the Raman signal stream of the pesticide residue component includes: extracting the three-dimensional pixel coordinate matrix for local surface fitting, calculating the curvature characteristic parameters of each pixel in the actual physical space, quantizing the local curvature parameters into capillary pressure values ​​according to the Yang-Laplace equation, and generating a capillary pressure gradient field pointing towards the depression area of ​​the root of the hair; utilizing the physical tendency generated by the capillary pressure gradient field to induce the noble metal nanoprobes to naturally deposit at the root of the hair during the liquid retraction process; constructing local enhanced hot spots in the nano gaps through the electromagnetic coupling effect of the local surfaces between the nanoprobes; using laser to excite the local enhanced hot spots to generate local plasmon resonance, forming an enhanced electromagnetic field to enhance the Raman scattering intensity of the pesticide residue molecules, and obtaining the enhanced Raman signal stream.

[0010] Preferably, the process of obtaining the pesticide residue signal matrix includes: controlling a laser to output excitation light with two slightly offset wavelengths and irradiating the area of ​​the monk fruit to be tested; for each wavelength of excitation light, using a polarization beam splitter to decompose the excited enhanced Raman signal stream into mutually orthogonal parallel polarization components and vertical polarization components; using an area array sensor to capture the two-dimensional light intensity data of the parallel polarization components and vertical polarization components, and combining them with a three-dimensional pixel coordinate matrix to perform spatial feature mapping, generating a corresponding orthogonal polarization image matrix; extracting the depolarization degree feature value of each pixel in the orthogonal polarization image matrix and calculating the corresponding polarization compensation weight coefficient; performing pixel-level weighted difference operation on the orthogonal polarization image matrix using the polarization compensation coefficient to cancel the background interference caused by scattering, extracting the net Raman feature signal of the pesticide residue components, and generating the pesticide residue signal matrix.

[0011] Preferably, the step of outputting the pesticide residue differential image stream includes: controlling the operating temperature of the laser to a first preset temperature value, outputting a laser of a first wavelength and illuminating the area to be tested, synchronously capturing the Raman and fluorescence mixed signal fed back from the area to be tested using an area array detector, and mapping the signal intensity value to the corresponding pixel space according to the three-dimensional pixel coordinate matrix to reconstruct and generate a first pesticide residue signal matrix; adjusting the operating temperature of the laser to a second preset temperature value, switching to a second wavelength and outputting a laser of a second wavelength to illuminating the area to be tested, and repeating the above capture and mapping process while keeping the acquisition parameters of the detection system unchanged until the spatial overlap between the currently acquired pixel feature data and the first pesticide residue signal matrix meets a preset convergence condition, thereby ending the above repetition process and generating a second pesticide residue signal matrix reflecting the same spatial location information; performing a pixel-by-pixel intensity difference operation on the first pesticide residue signal matrix and the second pesticide residue signal matrix using an image subtraction algorithm, canceling static fluorescence background interference through difference, and extracting the pesticide residue Raman feature signal that shifts with wavelength, and outputting the pesticide residue differential image stream reflecting the spatial distribution information of pesticide residue components.

[0012] Preferably, the signal intensity value includes: the gray value reflecting the original photosensitive response of each spatial pixel in the area to be measured, the stimulated Raman intensity corresponding to the vibrational spectrum of the characteristic molecules of pesticide residue components, and the continuous broadband fluorescence radiation intensity superimposed on the background of the vibrational spectrum of the characteristic molecules of pesticide residue components.

[0013] Preferably, the step of establishing the attenuation correction model of the Raman signal includes: acquiring real-time surface topography data of the area to be tested, calculating the local curvature parameters and surface normal vectors and mapping them to the three-dimensional pixel coordinate matrix, and registering them with the pixel-by-pixel spatial position of the pesticide residue feature differential image stream; calculating the incident energy attenuation factor based on the angle between the surface normal vector and the incident beam, and calculating the loss weight coefficient of the light receiving angle of the detection system in combination with the local curvature parameters, and establishing a gain compensation mapping function relationship for the Raman signal intensity with the energy projection attenuation factor and the loss weight coefficient as independent variables to construct the attenuation correction model; and using the attenuation correction model to perform intensity compensation and numerical repair on the pesticide residue differential image stream to restore the true distribution intensity of pesticide residue components, so as to output the corrected quantitative detection result of pesticide residues.

[0014] Preferably, the step of performing pixel compensation on the pesticide residue differential image to complete pesticide residue concentration imaging includes: using an attenuation correction model to perform pixel-by-pixel intensity gain compensation on the pesticide residue differential image to restore the true scattering intensity at each coordinate point, and combining it with a preset concentration conversion coefficient to complete quantitative spatial imaging of the pesticide residue concentration of monk fruit.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention creatively integrates three major technologies: three-dimensional spatial positioning and obstacle avoidance, capillary-driven nanoprobe local enhancement, and frequency-shifting polarization differential detection. By guiding the excitation beam precisely through the gaps in the capillary fibers and utilizing the nanoprobe to construct enhanced hotspots in the recessed areas, the target signal is amplified at the physical level. Simultaneously, by utilizing the characteristic that the Raman peak of pesticide residue molecules remains essentially unchanged in fluorescence background as the excitation wavelength shifts, and combining this with the depolarization characteristics of the capillary fibers, polarization differential processing is performed to effectively remove fluorescence and scattering background interference at the spectral level. This dual strategy of spatially enhancing the signal stream and spectrally removing interference significantly improves the signal-to-noise ratio and sensitivity of the detection, enabling direct imaging of trace pesticide residues on the surface of complex biological samples.

[0016] 2. This invention integrates a high-precision distance sensor to acquire real-time three-dimensional curvature gradient and surface normal data of the detection area, and uses this data to establish a Raman signal attenuation correction model. This model can compensate for pixel intensity deviations caused by incident energy attenuation and differences in signal collection efficiency point by point. After correction, the output differential image of pesticide residues can accurately reflect the spatial distribution of pesticide residue concentrations, providing a solid data foundation for accurate risk assessment.

[0017] 3. This invention utilizes the molecular fingerprinting capability of Raman spectroscopy, further confirming the characteristic peaks of the signal flow through frequency shifting differential analysis, effectively eliminating interference from non-target substances. For special agricultural products with irregular, fuzzy surfaces, such as monk fruit, this invention, based on non-contact processing, actively adapts to sample morphology through guided optical path and probe adaptive enrichment, and performs intelligent signal processing with background suppression and attenuation correction. This significantly improves the detection accuracy and applicability for complex samples, providing an innovative solution for rapid, in-situ visual detection of surface contaminants in agricultural products. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the workflow of the Raman spectroscopy-based imaging method for suppressing the fluorescence background of pesticide residues in monk fruit according to the present invention. Figure 2 This is a schematic diagram of the attenuation correction and quantitative imaging structure of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figures 1 to 2 The present invention relates to a method for suppressing the fluorescence background of pesticide residues in monk fruit based on Raman spectroscopy, and the technical solution is as follows: A method for suppressing the background fluorescence of pesticide residues in monk fruit based on Raman spectroscopy includes: The distribution data of hair tissue in the test area is obtained by using a distance sensor, and the spatial filtering characteristics of the confocal optical path are combined to drive the light beam through the gap in the test area and lock the three-dimensional pixel coordinate matrix. Based on the three-dimensional pixel coordinate matrix, the local curvature distribution features are extracted and mapped to the capillary pressure gradient field. The capillary action is used to promote the enrichment of noble metal nanoprobes, constructing a local enhanced hot spot. The Raman scattering signal is enhanced by the local electromagnetic field, and the enhanced Raman signal stream of pesticide residue components is output. Using the frequency-shifting excitation method, the laser temperature is controlled to output lasers of different wavelengths to irradiate the area under test. Based on the depolarization physical characteristics of the fuzzy fiber, polarization phase detection is performed on the enhanced Raman signal stream for different wavelengths of laser. The light intensity data under orthogonal polarization states is captured by the area array sensor. The polarization difference method is executed to remove scattering interference, obtain the pesticide residue signal matrix, and perform differential processing by the image subtraction algorithm to output the pesticide residue differential image stream. By using a distance sensor to extract the curvature gradient of the area to be measured, and combining it with the pesticide residue differential image stream, an attenuation correction model for the Raman signal is established. Pixel compensation is then performed on the pesticide residue differential image stream to complete pesticide residue concentration imaging.

[0021] Example 1: This embodiment provides an application scenario for pesticide residue detection in fresh organic monk fruit under conditions of high-density fuzz covering the surface. Because the surface of organic monk fruit is densely covered with a fuzzy structure with strong fluorescence interference and high scattering characteristics, traditional optical detection methods can often only obtain messy signals from the tips of the fuzz, making it difficult to reach the actual surface of the fruit peel where pesticide residues accumulate.

[0022] Furthermore, the process of locking the three-dimensional pixel coordinate matrix includes: using a distance sensor to acquire the distribution data of the hairy tissue in the area to be tested and identifying the connecting gaps through which the light beam can pass; driving the light beam through the connecting gaps and using the spatial pinhole in the optical path to perform spatial filtering, filtering out astigmatism and focusing on the surface of the fruit peel; reading the scanning state parameters at the time of focusing in real time, and mapping the scanning state parameters to three-dimensional physical space coordinates based on the geometric projection model, thereby locking the three-dimensional pixel coordinate matrix.

[0023] Specifically, the step of locking the three-dimensional pixel coordinate matrix includes: when probing the surface of the monk fruit, the system activates a high-precision laser displacement ranging sensor, which emits an infrared ranging beam and receives its echo signal to acquire the topographic point cloud data of the monk fruit surface. Because the surface of the monk fruit is densely covered with hairy tissue, the sensor receives reflection data at different height levels; at this time, the control circuit extracts signal intensity abrupt change points, identifying the low scattering rate connected regions between the hairy fibers as the physical gaps that allow the excitation beam to directly penetrate and reach the surface of the fruit peel.

[0024] The system then drives the biaxial galvanometer to adjust the incident vector of the excitation laser, ensuring the laser beam is accurately aligned with the predetermined connecting gap. A spatial pinhole is placed in the optical path at a position conjugate to the focal plane of the objective lens, forming a confocal detection structure. When the beam passes through the fibrous gap and focuses on the fruit peel surface, only the effectively reflected light from the focal plane passes through the pinhole, while the defocused astigmatism generated by the fibrous fibers is physically shielded by the pinhole. The axial height of the objective lens is finely adjusted via a stepper drive mechanism. When the light intensity received by the detector reaches a preset threshold, it is determined that the beam has been precisely focused on the fruit peel surface. At the instant focusing is completed, the system simultaneously acquires the real-time status parameters of the scanning mechanism. These real-time status parameters include the deflection angle of the biaxial galvanometer at the current sampling point and the axial displacement value of the objective lens drive motor. Based on the spatial mapping relationship between the biaxial deflection and axial displacement, a homogeneous coordinate transformation matrix is ​​used to synchronously map these real-time status parameters to the same three-dimensional physical coordinate system. In this process, a beam direction vector is constructed using a dual-axis deflection angle. Based on the system's equivalent focal length, this beam direction vector is projected and transformed to determine the focal point as the two-dimensional coordinate components of the horizontal x-axis and y-axis. The axial displacement of the objective lens is then mapped to the height coordinate component of the vertical z-axis. Through cascaded operations of homogeneous coordinate transformation matrices, these components are calculated as the absolute position coordinates (x, y, z) of the focal point in three-dimensional physical space. Subsequently, spatial registration is performed with the pixel index of the area array sensor to lock and generate the three-dimensional pixel coordinate matrix.

[0025] The preset threshold is determined using pre-stored calibration data of the reflectivity characteristics of the monk fruit peel surface. Specifically, this calibration data is obtained offline by scanning standardized monk fruit samples free of pesticide residues at multiple points to acquire their reflected light intensity envelope within the depth of focus range. The system sets a dynamic threshold range based on the average peak value of the envelope. This range is set such that the amplitude of the real-time received photoelectric signal must exceed three times the system's background noise and reach 80% to 110% of the peak intensity in the calibration data.

[0026] The distance sensor employs a 905 nm wavelength modulated laser, utilizing the time-of-flight principle for distance measurement. The ranging error is controlled within ±3 mm, and the working distance covers 50 mm to 500 mm. The processing logic is as follows: surface conditions are determined by analyzing the waveform characteristics of the laser echo signal. When the time spread of the echo signal exceeds a preset pulse width threshold, it is identified as a minor surface protrusion; when the echo signal intensity is less than twice the average background noise, it is identified as a surface void region. In the spatial coordinate mapping process, the horizontal coordinate value is determined by the product of the detection depth and the horizontal detection angle tangent; the vertical coordinate value is determined by the product of the detection depth and the vertical detection angle tangent; and the depth coordinate value is directly determined by the depth offset measured by the sensor. The equivalent optical parameter is set to 50 mm, the angle detection accuracy reaches 0.01 degrees, and the depth direction discrimination reaches 0.1 mm.

[0027] This invention utilizes high-precision laser displacement ranging and signal intensity abrupt change recognition to accurately detect the physical distribution of the fuzzy tissue on the surface of monk fruit, and automatically identifies the tiny gaps through which the light beam can penetrate. This physically avoids the strong scattering interference of high-density fuzz on the excitation light, ensuring that energy reaches the fruit peel surface directly. Secondly, the introduction of a confocal detection structure and spatial pinhole filtering technology effectively shields out-of-focus stray light from the non-focal plane, significantly improving the signal-to-noise ratio and solving the problem of background noise masking characteristic spectra under complex morphologies. Finally, homogeneous coordinate transformation converts the real-time state parameters of the scanning mechanism into three-dimensional physical space coordinates, achieving precise registration between optical sampling points and spatial geometric positions.

[0028] Furthermore, the three-dimensional pixel coordinate matrix is ​​extracted for local surface fitting, and the curvature characteristic parameters of each pixel in the actual physical space are calculated. According to the Yang-Laplace equation, the local curvature parameters are quantified into capillary pressure values, generating a capillary pressure gradient field pointing towards the depression area at the root of the hair. Utilizing the physical tendency generated by the capillary pressure gradient field, noble metal nanoprobes are naturally deposited at the root of the hair during liquid retraction. Through the electromagnetic coupling effect of the local surface between nanoprobes, local enhanced hot spots are constructed within the nano gaps. Laser excitation of the local enhanced hot spots generates local plasmon resonance, forming an enhanced electromagnetic field to enhance the Raman scattering intensity of pesticide residue molecules, thereby obtaining the enhanced Raman signal flow.

[0029] Specifically, the steps for obtaining the enhanced Raman signal stream include: First, the system calls the moving least squares algorithm to perform local surface fitting on the local neighborhood pixels in the three-dimensional pixel coordinate matrix. By establishing a bivariate quadratic polynomial regression model, the discrete spatial coordinates are transformed into a continuous mathematical surface function. Then, the partial derivatives of this function are solved to construct the Hessian matrix, and the eigenvalues ​​of the Hessian matrix are extracted to accurately calculate the first and second principal curvature values ​​of the sampling points in the actual physical space. Because the hairy roots on the surface of the monk fruit exhibit significant microscopic depressions at the junction with the peel, the curvature characteristic parameters at this location will be significantly different from those of the flat peel area.

[0030] Based on the Yang-Laplace principle of physics, there is a quantitative proportional relationship between local curvature and the pressure difference between the inside and outside of the liquid surface. The system multiplies the sum of the two calculated principal curvature values ​​with the preset solvent surface tension coefficient, quantifying the geometric curvature parameter value into a physical capillary pressure value. A solution containing noble metal nanoparticles is pre-dropped into the area to be tested. Due to the extremely small radius of curvature at the root of the capillary, a significant negative pressure is generated in this area. This pressure difference driven by curvature difference generates a capillary pressure gradient field pointing towards the root of the capillary in the microscopic space. Utilizing the physical tendency of the liquid during natural evaporation, the noble metal nanoprobes dispersed in the solution are spontaneously aggregated and naturally deposited towards the high curvature region at the root of the capillary; as the liquid solvent recedes, the noble metal nanoprobes deposited at the root of the capillary form a dense stacked structure. When the physical gap between adjacent nanoparticles shrinks to within ten nanometers, a strong electromagnetic coupling effect is generated between the particles, and a high-density localized enhanced hotspot is constructed within these nanoscale gaps.

[0031] The calculated curvature feature parameters of each pixel in the actual physical space are summed, and this sum is multiplied by the surface tension coefficient. Based on the Yang-Laplace principle, the pressure difference between the inside and outside of the local liquid surface has a definite proportional relationship with the sum of the principal curvature values ​​at that point and the product of the surface tension coefficient. Therefore, the curvature parameters in the geometric dimension are quantified into capillary pressure values ​​in the physical dimension. Because the sum of the principal curvatures at the root of the capillary is significantly greater than that of the surrounding flat area, a large pressure difference and a significant negative pressure are generated in this local area, constructing a capillary pressure gradient field from the flat area to the root of the capillary in the microscopic space. Utilizing the physical tendency of liquids during natural evaporation, the noble metal nanoprobes dispersed in the solution are spontaneously aggregated and naturally deposited towards this high pressure gradient region, ultimately constructing a localized enhanced hotspot at the root of the capillary.

[0032] In establishing the capillary pressure gradient field, the system calculates the pressure difference based on the physical properties of the fluid interface. The value of this pressure difference is equal to the product of the surface tension coefficient of the liquid and the sum of the values ​​of the two principal curvature characteristic parameters mentioned above.

[0033] Considering the influence of ambient temperature on physical parameters, a dynamic compensation mechanism is introduced into the system. Under the standard environment of 25 degrees Celsius, the surface tension coefficient is set to 0.072 Newtons per meter, and is linearly corrected according to the relationship that the value decreases by 0.00015 Newtons per meter for every degree Celsius increase, so as to ensure the stability of the pressure gradient field under different temperature control conditions.

[0034] To obtain accurate principal curvature of the interface, the system performs moving least squares surface fitting on the collected surface topography data. The specific algorithm configuration is as follows: a local region with a radius of three millimeters around each sampling point is selected, containing no fewer than 50 valid sampling points. A local surface is constructed using a bivariate quadratic polynomial to calculate the geometric curvature of the surface. A distance weighting mechanism is introduced; the specific logic of the weighting function is that the weight value decreases exponentially with the square of the distance between the sampling point and the center point, meaning that the closer the sampling data is to the point to be calculated, the higher its contribution to the surface fitting result.

[0035] In specific biochemical detection applications, the system employs gold nanoparticles with a particle size between 20 and 50 nanometers as the noble metal nanoprobe, with an initial concentration of 10 to the power of 12 particles per milliliter of the test liquid. Under a controlled environment of 25 degrees Celsius and 45% relative humidity, the time constant for liquid retraction is 180 seconds. This time constant ensures that solute particles have sufficient time to accumulate in the predetermined region under the drive of the pressure gradient, achieving highly sensitive signal enhancement.

[0036] The capillary pressure gradient was used to drive the noble metal nanoprobe to migrate and deposit in the hairy root depression area. The enrichment effect increased the volume fraction of nanoparticles in the region by at least 50 times compared with the initial concentration of the solution.

[0037] The localized enhanced hotspot achieves an order-of-magnitude increase in Raman signal. A laser of a specific wavelength emits excitation light, precisely illuminating the localized enhanced hotspot. Under the influence of the incident photomagnetic field, free electrons on the surface of the noble metal nanoparticles undergo stimulated oscillation, generating localized surface plasmon resonance. This resonance effect induces an extremely strong local electromagnetic field within the nano-gap, significantly enhancing the stimulated Raman scattering cross-section of pesticide residue molecules adsorbed near the hotspot. The increased stimulated Raman scattering cross-section increases the probability of energy exchange between pesticide residue molecules and excitation photons, resulting in an order-of-magnitude increase in the flux of characteristic scattered photons generated by pesticide residue molecules under the same incident light power. The detection system efficiently collects these high-intensity characteristic scattered photons induced by the cross-section enhancement through an optical acquisition link, and uses a high-sensitivity detector to convert the optical signal into an electrical signal, outputting an enhanced Raman signal stream reflecting the concentration of pesticide residue components. This process ensures that even at extremely low pesticide residue concentrations, a clear characteristic spectrum, free from system noise, can still be obtained.

[0038] The process of obtaining the preset solvent surface tension coefficient includes: measuring the surface tension of the nanocolloid solution at different temperatures and mass fractions using the pendant drop method in advance, and establishing a multi-dimensional physical property calibration library; during detection, the system searches the calibration library based on the real-time temperature data acquired by the sensor, and calculates the precise tension coefficient at the current instant using bilinear interpolation logic. Subsequently, the system multiplies the tension coefficient value by the sum of the calculated curvature feature parameters of the pixel in the actual physical space, thereby converting the geometric morphology features into a physical quantity reflecting the capillary pressure gradient.

[0039] This invention deeply integrates three-dimensional morphological features with microfluidic principles, utilizing local principal curvature parameters to construct a capillary pressure gradient field, enabling the directional spontaneous enrichment of noble metal nanoprobes in the recessed area of ​​the hairy root of *Siraitia grosvenorii*. This process effectively solves the problem of uneven distribution of nanomaterials on complex biological surfaces, and can create enhanced hotspots in areas containing pesticide residues. Secondly, by utilizing the plasmon resonance effect, the stimulated Raman scattering cross-section of the target molecules is significantly expanded. This results in an order-of-magnitude increase in the scattered photon flux, enhancing the ability to capture weak signals and obtaining characteristic spectra with high signal-to-noise ratios even in the presence of pesticide residues, effectively eliminating interference from system noise. This method improves detection sensitivity and reliability from a physical mechanism perspective, providing a solid technical guarantee for the accurate quantitative analysis of pesticide residue components in complex biological contexts.

[0040] Further, the process of obtaining the pesticide residue signal matrix includes: controlling the laser to output excitation light with two slightly offset wavelengths and illuminating the area of ​​the monk fruit to be tested; for each wavelength of excitation light, using a polarization beam splitter to decompose the excited enhanced Raman signal stream into mutually orthogonal parallel polarization components and vertical polarization components; using an area array sensor to capture the two-dimensional light intensity data of the parallel polarization components and vertical polarization components, and combining them with a three-dimensional pixel coordinate matrix to perform spatial feature mapping, generating a corresponding orthogonal polarization image matrix; extracting the depolarization degree feature value of each pixel in the orthogonal polarization image matrix and calculating the corresponding polarization compensation weight coefficient; performing pixel-level weighted difference operation on the orthogonal polarization image matrix using the polarization compensation coefficient to cancel the background interference caused by scattering, extracting the net Raman feature signal of the pesticide residue components, and generating the pesticide residue signal matrix.

[0041] Specifically, the process of generating the pesticide residue signal matrix includes: the system precisely controls the operating current of the thermistor and the cooling chip of the semiconductor laser to change the operating temperature of the laser resonant cavity, outputting two excitation beams with slight wavelength shifts. These two excitation beams are sequentially irradiated onto the test area of ​​the monk fruit. Since the Raman spectral lines shift synchronously with the excitation wavelength, while background fluorescence and ambient astigmatism remain stable within a small wavelength range, this step lays the physical foundation for subsequent interference removal through differential operations. For each wavelength of excitation light, the system guides the excited enhanced Raman signal stream to a polarization beam splitter. Utilizing the birefringence of this optical element, the mixed signal is decomposed into two mutually orthogonal beams: one is a parallel polarization component with the same polarization direction as the incident laser, and the other is a perpendicular polarization component with an orthogonal direction. Since the random scattering of light by the hairs and peel tissue on the surface of the monk fruit causes significant depolarization, and the characteristic Raman scattering of pesticide residue molecules has specific polarization characteristics, this phase-separated detection can effectively extract the characteristic differences of the signal in the polarization dimension.

[0042] In this embodiment, the minute offset is limited to 0.3 nm to 0.5 nm. The system controls the temperature compensation of the laser resonant cavity by ±2 degrees Celsius by precisely adjusting the resistance feedback of the thermistor inside the semiconductor laser. This allows the laser to cyclically switch and output a first excitation light and a second excitation light with a wavelength difference constant between 0.3 nm and 0.5 nm. The minute offset ensures a significant shift in the Raman scattering characteristic peak of pesticide residue molecules, while the background fluorescence maintains a stable energy distribution within this wavelength transition range.

[0043] A high-sensitivity CCD sensor is used to simultaneously capture two-dimensional light intensity data of the parallel polarization component and the vertical polarization component. The two-dimensional light intensity distribution data consists of multiple pixel units arranged in rows and columns, and each pixel unit has a corresponding and unique physical pixel index.

[0044] Using a bilinear interpolation algorithm, the pixel indices in the two-dimensional light intensity distribution data are matched with the three-dimensional pixel coordinate matrix. Through this mapping process, the system reconstructs the two-dimensional light intensity distribution data into a spatial attribute matrix with real physical space dimensions, achieving pixel-by-pixel alignment of the parallel polarization components and the vertical polarization components in three-dimensional physical space.

[0045] The system executes a weight coefficient optimization algorithm based on the least squares criterion, specifically including: A sliding window is defined centered on the current pixel, and local grayscale data of the parallel polarization component and the vertical polarization component within the window are extracted. The least squares method is used to minimize the sum of squared residuals after differencing. The algorithm treats the parallel polarization grayscale value as the dependent variable and the vertical polarization grayscale value as the independent variable. According to the principle of least squares, the polarization compensation weight coefficient is derived by the following ratio: the sum of the products of the parallel polarization grayscale values ​​and the vertical polarization grayscale values ​​within the window, divided by the sum of the squares of the vertical polarization grayscale values. The system applies the polarization compensation weight coefficient to the corresponding pixel unit in the two-dimensional light intensity distribution data, specifically by subtracting the product of the vertical polarization component pixel grayscale value and the corresponding position weight coefficient from the pixel grayscale value of the parallel polarization component. Through this pixel-level weighted difference operation, background interference is eliminated and the net Raman feature signal is extracted. Combined with the spatial coordinates of the net Raman feature signal in the three-dimensional pixel coordinate matrix, the pesticide residue signal matrix is ​​generated.

[0046] When preprocessing the detection data, the scattering characteristics of the target surface are quantified by calculating the degree of depolarization. The calculation logic for the degree of depolarization is as follows: First, obtain the intensity values ​​of the parallel polarization component and the perpendicular polarization component; calculate the ratio of the absolute value of the difference between the two to the sum of the two; then subtract the ratio from the value of 1; the final percentage value is the degree of depolarization. The value of the degree of depolarization ranges from 0 to 1.

[0047] Experimental studies show that different landform features exhibit significant quantitative differences under polarization fields: the depolarization degree of surface fuzziness ranges from 70% to 95%; the depolarization degree of smooth fruit peel areas ranges from 10% to 30%; while pesticide residue molecules, due to their specific molecular arrangement, typically have a depolarization degree below 5%. To eliminate noise interference from the fuzzy background, the system uses a 5-pixel multiplied 5-pixel sliding window for least-squares fitting and introduces a polarization compensation weight coefficient. This polarization compensation weight coefficient is calculated by subtracting the depolarization degree of the pixel from a value of 1; the magnitude of the polarization compensation weight coefficient is negatively correlated with the depolarization degree, thus suppressing the contribution of highly depolarized regions to the fitting results.

[0048] The mathematical model used in the fitting process is set as follows: the intensity value of the parallel polarization component is equal to the product of the undetermined slope coefficient and the intensity value of the vertical polarization component, plus an intercept constant. The system solves for the optimal parameters by minimizing a loss function, which is defined as the sum of the products of the weight coefficient of each pixel and the square of the difference between the measured value and the model prediction value of the parallel polarization component within a specified 5x5 window. Through this weighted fitting mechanism, the influence of the fuzzy region with extremely high depolarization degree on the fitting result is significantly suppressed because its weight is close to zero, thus achieving high-precision extraction of pesticide residue signals.

[0049] This invention employs a dual decoupling of dual-wavelength frequency-shifting excitation and polarization-phase detection, enabling simultaneous suppression of interference signals in both the spectral and polarization dimensions. Utilizing the physical property that the fluorescence background remains stable as the Raman characteristic peaks of pesticide residues shift with wavelength, combined with polarization state difference identification, it solves the detection challenge of coexisting strong fluorescence background and diffuse scattering interference from fuzzy tissue. Secondly, by introducing pixel-level depolarization degree eigenvalues ​​and polarization compensation weighting coefficients, it achieves normalization processing of non-specific scattering signals from complex surfaces. It accurately removes polarization interference noise caused by fuzzy tissue, extracting high-purity net Raman characteristic signals, significantly improving the signal-to-noise ratio and the accuracy of substance identification. Finally, it deeply integrates multi-dimensional photoelectric information with a three-dimensional spatial coordinate matrix, ensuring strict consistency of data at each sampling point in physical space, greatly enhancing the system's sensitivity for pesticide residue detection in complex biological sample backgrounds, and providing reliable data support for accurate quantitative analysis of pesticide residues.

[0050] Further, the steps for outputting the pesticide residue differential image stream include: controlling the operating temperature of the laser to a first preset temperature value, outputting a laser of a first wavelength and illuminating the area to be tested, simultaneously capturing the Raman and fluorescence mixed signal fed back from the area to be tested using an area array detector, and mapping the signal intensity value to the corresponding pixel space according to the three-dimensional pixel coordinate matrix to reconstruct and generate a first pesticide residue signal matrix; adjusting the operating temperature of the laser to a second preset temperature value, switching to a second wavelength and outputting a laser of a second wavelength to illuminating the area to be tested, and repeating the above capture and mapping process while keeping the acquisition parameters of the detection system unchanged until the spatial overlap between the currently acquired pixel feature data and the first pesticide residue signal matrix meets a preset convergence condition, at which point the above repetition process ends, generating a second pesticide residue signal matrix reflecting the same spatial location information; performing a pixel-by-pixel intensity difference operation on the first pesticide residue signal matrix and the second pesticide residue signal matrix using an image subtraction algorithm, canceling static fluorescence background interference through difference, and extracting the pesticide residue Raman feature signal that shifts with wavelength, and outputting the pesticide residue differential image stream reflecting the spatial distribution information of pesticide residue components.

[0051] During the process of acquiring pixel features that match the spatial coordinates of the first pesticide residue signal matrix, the system executes the following spatial registration logic to ensure the rigor of the data stream and the reproducibility of the hardware execution: The system first performs spatial registration preprocessing on the pesticide residue signal matrices obtained from two samplings, and evaluates the spatial alignment quality of the two matrices through normalized cross-correlation calculation. Specifically, the system performs frequency domain transformation on the non-zero regions representing pesticide residue characteristics in the first and second pesticide residue signal matrices. By calculating the cross-power spectrum of the two frequency domain matrices, and then reverting to the spatial domain via inverse transformation, a spatial correlation function spectrum is obtained. The system automatically searches for the maximum value point in the spatial correlation function spectrum. The coordinate offset of this maximum value point corresponds to the optimal translation vector between the two sets of matrices in the horizontal and vertical directions. The system determines the scanning process by judging whether the translation vector meets the preset convergence criteria: when the peak intensity of the spatial correlation function spectrum exceeds the preset 0.95 threshold, and the magnitude of the corresponding translation vector (i.e., the square root of the sum of the squares of the horizontal and vertical displacements) is less than or equal to 0.5 pixel steps, the system determines that the two matrices have reached the convergence criterion for spatial coordinate registration and then stops the repeated scanning process.

[0052] If the above convergence conditions are not met, the system will automatically activate the feedback control loop: based on the calculated displacement deviation, a spatial compensation command is generated and applied to the galvanometer drive module in real time, driving the galvanometer to deflect by a small angular increment to correct the physical scanning position. To prevent computational dead loops caused by extreme interference in the hardware environment, the maximum number of iterations in this fine-tuning compensation process is preset to three. If the convergence threshold is still not reached after three iterations of fine-tuning, the system will automatically extract the two sets of pixel data with the highest cross-correlation for weighted fusion processing and output an abnormal status record, ensuring the robustness of pesticide residue signal extraction while ensuring the determinism of the detection process.

[0053] The preset convergence conditions specifically include a normalized cross-correlation coefficient threshold, a spatial displacement vector magnitude threshold, and a maximum number of iterations. The selection criteria are as follows: First, regarding the selection of a normalized cross-correlation coefficient greater than 0.95: In practical scenarios for pesticide residue detection, due to the inherent photoelectric and thermal noise of area array sensors and diffuse reflection interference from weak pesticide residue fluorescence signals, even if the two sampling matrices are physically aligned, their pixel grayscale distribution cannot reach the ideal cross-correlation value of 1.0. Setting 0.95 as the threshold is a balance point determined based on sensor signal-to-noise ratio test results. This can filter out random fluctuations caused by noise while ensuring that the extracted pixel features have extremely high spatial consistency, thereby guaranteeing the accuracy of differential extraction.

[0054] Second, regarding the selection criteria for a translation vector magnitude less than or equal to 0.5 pixels: This parameter is the core guarantee for achieving sub-pixel-level configuration. According to the Nyquist sampling theorem and image interpolation theory, when the spatial alignment deviation of the two sets of matrices is reduced to within half a pixel, the gray quantification error caused by spatial discrete sampling will be minimized, sufficient to support subsequent high-precision quantitative analysis of pesticide residue components. If the displacement tolerance is set too large (e.g., greater than one pixel), artifacts will appear at the edges after differentiation, interfering with the identification of true pesticide residue characteristics.

[0055] Third, the selection criteria for a maximum of three iterations: This is based on a comprehensive consideration of the response time of the laser galvanometer drive module and the dynamic stability of the system. Extensive engineering experiments have verified that, under normal testing conditions, spatial deviations caused by mechanical micro-vibrations or environmental temperature drift are typically within two pixel periods, and convergence can be achieved through one or two closed-loop compensations. Setting the upper limit to three iterations aims to balance detection efficiency and registration accuracy. This not only corrects most physical displacement deviations but also effectively prevents the system from falling into an invalid loop due to failure to converge under extreme environmental interference (such as strong vibrations), ensuring the determinism of the overall testing process.

[0056] Specifically, the process of outputting the pesticide residue differential image stream includes: First, the system precisely locks the laser's operating temperature to a first preset temperature value using a closed-loop temperature control module composed of a built-in thermoelectric cooler and a high-precision thermistor. In this state, the laser outputs continuous excitation light with a center wavelength of the first wavelength (e.g., 785.0 nm) and illuminates the area of ​​the monk fruit to be tested. A high-sensitivity area array detector synchronously captures the feedback signal generated by the excitation. This signal appears as Raman characteristic peaks of pesticide residue components superimposed on a broad-spectrum background fluorescence. Based on the three-dimensional pixel coordinate matrix locked in the aforementioned steps, the system maps the light intensity values ​​sensed by the detector to the corresponding spatial pixel coordinates in real time. Through spatial interpolation and resampling techniques, it reconstructs and generates a first pesticide residue signal matrix reflecting the excitation of the first wavelength. Subsequently, the system automatically adjusts the operating current of the thermoelectric cooler to smoothly switch the laser resonant cavity temperature to a second preset temperature value, outputting a second wavelength excitation light with a slight frequency shift (e.g., shifted to 785.5 nm). While keeping the exposure time, gain factor, and other acquisition parameters of the area array detector completely unchanged, the above light signal acquisition process is repeated. To ensure strict spatial correspondence between the two captured data sets, the system uses an image registration algorithm to monitor the position feedback of the scanning mechanism in real time until it acquires pixel feature data that completely overlaps with the first pesticide residue signal matrix in physical coordinates. After mapping, the system generates a second pesticide residue signal matrix reflecting the same spatial location information.

[0057] An image subtraction algorithm is invoked to perform pixel-by-pixel intensity difference processing on the first and second pesticide residue signal matrices. Since the fluorescence background generated by the peel and internal components of the monk fruit is a continuous broadband spectrum, its spectral intensity and shape remain static under the aforementioned excitation wavelength shift of 0.3 nm to 0.5 nm; while the Raman characteristic signals of pesticide residue components shift at the same frequency with the change in excitation wavelength. By performing pixel-by-pixel subtraction, the static fluorescence background interference is physically canceled out, while the wavelength-shifting Raman characteristic peaks are extracted, forming a differential spectral signal with positive and negative polarity characteristics. The system summarizes the net signal intensity after subtraction of all pixels and performs absolute value processing, outputting a pesticide residue differential image stream reflecting the spatial distribution information of pesticide residue components on the surface of the monk fruit.

[0058] The specific implementation process of the image subtraction algorithm is as follows: First, an affine transformation algorithm is called to spatially align the first and second pesticide residue signal matrices. The system extracts the feature texture of the monk fruit surface as a reference anchor point, and through translation, rotation, and scaling transformations, ensures that the spatial coordinate deviation of corresponding pixels in the two matrices is lower than a preset sub-pixel threshold. Second, the real-time laser power ratio at the two sampling times is obtained, and the pixel intensity of the first or second pesticide residue signal matrix is ​​linearly scaled. By adjusting the background intensity of non-signal areas to the same reference energy level, it is ensured that subsequent differential operations can physically cancel the static fluorescence background.

[0059] The third step involves iterating through the registered matrices and subtracting the pixel intensity value at the same position in the second matrix from the pixel intensity value at position (x,y) in the first matrix. At this point, because the Raman peak of pesticide residues shifts with wavelength, the difference result exhibits a characteristic waveform with alternating positive and negative polarities at the Raman frequency shift, while the fluorescence background difference is a stable baseline that tends towards zero.

[0060] The fourth step involves performing full-wave rectification (i.e., absolute value processing) on ​​the differential bipolar signal to integrate the effective energy generated by the Raman shift. Subsequently, the system accumulates and integrates the net intensity of each pixel within the characteristic frequency band and maps it to grayscale values, ultimately generating a differential image stream that reflects the spatial distribution characteristics of pesticide residues.

[0061] The preset sub-pixel threshold is determined based on the matching relationship between the point spread function of the optical imaging system and the sensor pixel size. The specific acquisition process is as follows: In the pre-calibration stage, the system acquires two consecutive frames of standard sample images and performs cross-correlation calculations to obtain the system's inherent random vibration displacement standard deviation; 1.5 times this standard deviation is set as the error judgment threshold. The specific value range of this threshold is set between 0.05 pixels and 0.15 pixels. When the coordinate residual after affine transformation is lower than this range, it is determined that the two matrices have completed spatial alignment and meet the conditions for pixel-by-pixel differential. If the residual is higher than this range, the system automatically triggers a secondary compensation algorithm or alarm to avoid interference from pseudo-differential signals caused by spatial mismatch.

[0062] This invention achieves precise frequency shift of the excitation wavelength by adjusting the laser temperature, and, combined with image subtraction, fully utilizes the physical differences in wavelength response between the fluorescence background and the Raman signal. Through pixel-by-pixel differential processing, the strong static fluorescence background interference from the monk fruit itself is effectively canceled without damaging the sample, significantly improving the signal-to-noise ratio and detection sensitivity of the pesticide residue characteristic spectrum. Secondly, through the deep integration of the three-dimensional pixel coordinate matrix and image registration algorithm, strict overlap of the pesticide residue signal matrix in spatial physical coordinates under different excitation wavelengths is ensured, effectively avoiding differential ghosting caused by surface morphology fluctuations and guaranteeing higher spatial purity of the extracted pesticide residue signal. Finally, a differential image stream reflecting the spatial distribution of pesticide residue components is output, achieving a technological leap from traditional fixed-point spectral measurement to large-area spatial imaging. This not only visually displays the distribution of pesticide residues on the surface of the monk fruit, but also provides a high-quality and standardized data source for subsequent precise quantitative analysis and concentration compensation.

[0063] Furthermore, the signal intensity value includes: the gray value reflecting the original photosensitive response of each spatial pixel in the test area, the stimulated Raman intensity corresponding to the vibrational spectrum of the characteristic molecules of pesticide residue components, and the continuous broadband fluorescence radiation intensity superimposed on the background of the vibrational spectrum of the characteristic molecules of pesticide residue components.

[0064] Specifically, the signal intensity value acquisition and analysis steps are as follows: The system uses a high-sensitivity area array detector (such as a charge-coupled device) to capture the comprehensive optical signal fed back from the surface of the monk fruit. When a photon strikes the photosensitive pixel of the detector, photogenerated carriers are generated through the photoelectric effect and converted into a voltage signal by the readout circuit. Subsequently, the voltage signal is converted into a digital quantity with a specific positioning depth (such as a 16-bit binary depth) via a built-in analog-to-digital converter, i.e., the output is a grayscale value reflecting the original photosensitive response of each spatial pixel in the test area. This grayscale value is the basic raw data for all subsequent signal processing, recording the total photon energy received by the detector. The acquired comprehensive grayscale value includes the stimulated Raman intensity at a specific wavelength position. This stimulated Raman intensity value corresponds to the characteristic vibrational energy level of the molecular bonds inside the pesticide residue component (such as carbamate pesticides). When a monochromatic excitation laser irradiates the pesticide residue molecule, due to inelastic scattering, the incident photon exchanges energy with the molecular vibrational energy level, resulting in a frequency shift of the photon energy with molecular characteristics, i.e., a characteristic Raman shift. The value of the characteristic Raman frequency shift is determined by the vibrational mode of the chemical bonds inside the pesticide residue molecule; the system achieves qualitative analysis of the pesticide residue components by identifying the position of the characteristic Raman frequency shift in the spectrum.

[0065] The system separates and maps optical signals within specific frequency shift ranges to corresponding pixel coordinates using a spectrometer. The intensity values ​​reflect the distribution density of pesticide residue molecules within the test area. Because the natural pigments and organic matter in monk fruit produce strong autofluorescence upon stimulation, the stimulated Raman intensity in the detector-captured signal is often superimposed on the continuous broadband fluorescence radiation intensity. Physically, the fluorescence spectrum presents as a background baseline with an extremely wide envelope and slow wavelength variation. The system records the gray values ​​of the non-characteristic response regions on both sides of the pesticide residue characteristic spectral lines and uses a baseline estimation algorithm to reconstruct the fluorescence radiation intensity values ​​superimposed on the pesticide residue component signals. Finally, the system's response signal model logically decouples these three components. By identifying the total gray value of the pixels and removing the background intensity generated by the continuous broadband fluorescence radiation, the net stimulated Raman intensity is accurately extracted.

[0066] Further, the steps for establishing the Raman signal attenuation correction model include: acquiring real-time surface topography data of the area to be tested, calculating the local curvature parameters and surface normal vectors and mapping them to the three-dimensional pixel coordinate matrix, and registering them with the pixel-by-pixel spatial position of the pesticide residue feature differential image stream; calculating the incident energy attenuation factor based on the angle between the surface normal vector and the incident beam, and calculating the loss weight coefficient of the detection system's receiving angle in combination with the local curvature parameters; establishing a gain compensation mapping function relationship for Raman signal intensity using the energy projection attenuation factor and the loss weight coefficient as independent variables to construct the attenuation correction model; and using the attenuation correction model to perform intensity compensation and numerical repair on the pesticide residue differential image stream to restore the true distribution intensity of pesticide residue components, so as to output the corrected quantitative detection result of pesticide residues.

[0067] Specifically, the steps for establishing a Raman signal attenuation correction model and completing spatial imaging of pesticide residue concentration include: First, real-time surface topography data of the area to be tested is acquired. Then, by performing numerical differentiation on the three-dimensional pixel coordinate matrix locked in the preceding steps, the local curvature parameter value and surface normal vector corresponding to each pixel are calculated.

[0068] The aforementioned geometric feature parameters are spatially registered pixel-by-pixel with the pesticide residue differential image stream to ensure that each spectral intensity data has a corresponding three-dimensional geometric attribute. Based on optical geometry principles, the system calculates the incident energy attenuation factor for each sampling point. This attenuation factor is obtained by solving the cosine of the angle between the incident laser beam vector and the surface normal vector at that point, and is used to correct for fluctuations in the effective excitation power per unit area caused by the surface undulations of the monk fruit. Simultaneously, combined with the local curvature parameters, the loss weighting coefficient of the detection system's receiving angle is calculated. Changes in surface curvature alter the spatial distribution envelope of the scattered light from pesticide residue molecules, causing some scattered light to exceed the effective numerical aperture receiving range of the optical lens. The system quantifies the light loss caused by topographic factors by calculating the overlap ratio between the solid angle of the scattered light distribution under the current curvature and the lens acquisition angle.

[0069] The loss weighting coefficient reflects the proportion of scattered light that cannot be completely collected by the optical lens due to surface curvature, quantifying the physical loss of the signal in the detection link. The system establishes a gain compensation mapping function for Raman signal intensity using the calculated incident energy attenuation factor and the light collection loss weighting coefficient as independent variables. This function, as the core of the correction model calculation, can dynamically calculate the compensation factor required to restore the true scattered intensity based on the three-dimensional spatial coordinates of each pixel; the geometric feature parameters include the local curvature parameter value and the surface normal vector corresponding to each pixel.

[0070] The calculation process for the compensation factor includes: multiplying the incident energy attenuation factor and the light loss weighting coefficient to obtain the optical transmission efficiency coefficient of the pixel due to its geometric shape. The transmission efficiency coefficient represents the proportion of the light signal captured by the detection system under the current complex physical shape to the total signal under ideal flat conditions. To restore the true scattering intensity, the system takes the reciprocal of the transmission efficiency coefficient as the compensation factor according to the gain compensation mapping function; that is, dividing the value of 1 by the transmission efficiency coefficient to obtain the reciprocal as the compensation factor. The transmission efficiency coefficient is used as the denominator, and the denominator value cannot be 0. For example, if a pixel suffers a 50% loss in overall signal due to tilt and curvature, the compensation factor calculated by the system according to the above logic is two.

[0071] This invention quantifies the impact of complex surface undulations on optical detection at the physical level by deeply coupling microscopic geometric parameters with spectral signal intensity at the pixel level. By calculating the incident energy attenuation factor using the surface normal vector, it effectively corrects fluctuations in effective excitation power caused by incident angle deviations, ensuring physical consistency of excitation intensity within the test area. Secondly, by combining local curvature calculations to determine the loss weight of the detection system's receiving angle, it accurately compensates for energy loss from scattered light caused by the curvature of biological surfaces. This quantification repairs physical losses in the detection chain, significantly improving the system's signal recovery capability when processing non-uniform samples. Finally, the model achieves accurate conversion from the original observed signal to the true intensity of pesticide residue components through a dynamic gain compensation function. This significantly improves the quantitative accuracy of pesticide residue detection and the reliability of spatial imaging, completely resolving measurement errors caused by the geometric heterogeneity of biological samples, and providing core technical support for the standardized quantitative analysis of pesticide residues.

[0072] Furthermore, the step of performing pixel compensation on the pesticide residue differential image to complete pesticide residue concentration imaging includes: using the attenuation correction model to perform pixel-by-pixel intensity gain compensation on the pesticide residue differential image to restore the true scattering intensity at each coordinate point, and combining it with a preset concentration conversion coefficient to complete quantitative spatial imaging of the pesticide residue concentration of monk fruit.

[0073] Specifically, the process of completing the pesticide residue concentration imaging includes: the system first calls the pre-constructed attenuation correction model to perform online repair for each spatial pixel in the pesticide residue differential image stream. During processing, the optical loss caused by the tilt and curvature changes of the monk fruit surface is offset by multiplying the pixel grayscale value in the original differential image with the compensation factor. The image data after gain compensation eliminates the ghosting and intensity deviation caused by the geometric shape of the tested sample, restoring the true scattered Raman signal intensity of each coordinate point in physical space. This scattered Raman signal intensity value excludes the interference of the external environment and macroscopic morphology, reflecting the contribution of pesticide residue molecules to the inelastic scattering of excitation light per unit volume. This step ensures that even in the complex concave areas or edge tilted areas of the monk fruit, the detection results can maintain physical uniformity and accuracy. The system introduces a concentration conversion coefficient obtained in advance through standard concentration sample calibration. This concentration conversion coefficient establishes a linear proportional mapping relationship between the Raman characteristic scattering intensity and the molar concentration of the substance. The system multiplies the restored true scattered Raman signal intensity by the concentration conversion coefficient, converting the optical intensity information into a pesticide residue concentration value. Subsequently, the system rearranges the calculated pesticide residue concentration values ​​for each pixel according to their spatial indices in the three-dimensional pixel coordinate matrix. During the visualization rendering stage, the system performs pixel-level color mapping using a color lookup table. Specifically, it uses the pesticide residue concentration value at each point as an index variable to retrieve and match the corresponding red, green, and blue primary color components in the color lookup table, generating and outputting a quantitative spatial distribution image of pesticide residue concentration on the surface of the monk fruit. This imaging result can intuitively display the enriched distribution area of ​​pesticide residues on the surface of the monk fruit and supports retrieving precise pesticide residue concentration data corresponding to any coordinate point in the image through an interactive interface.

[0074] This invention proposes a Raman spectroscopy-based imaging method for suppressing the fluorescence background of pesticide residues in monk fruit. By performing pixel-by-pixel intensity gain compensation, it completely eliminates optical losses and intensity deviations caused by the complex surface undulations, tilt, and curvature variations of monk fruit. An attenuation correction model is used to restore the true scattering intensity, ensuring high physical homogeneity and measurement accuracy even in concave or edge areas of the fruit, eliminating detection errors caused by the geometric heterogeneity of biological samples. Furthermore, by combining a standard-calibrated concentration conversion coefficient, it achieves precise quantitative conversion from abstract optical intensity to specific molar concentration. This technical approach not only improves the scientific rigor of the detection but also ensures the repeatability and standardization of the results. Finally, a spatial distribution cloud map of pesticide residues is output through three-dimensional spatial index reconstruction and a color lookup table. This imaging method not only clearly reveals the enrichment patterns of pesticide residues on the fruit surface but also supports concentration data interaction at any coordinate point, providing strong technical support for accurate, rapid, and non-destructive monitoring of trace pesticide residues.

[0075] Example 2: This embodiment, based on the system architecture described in Embodiment 1, further describes the specific implementation of the preset concentration conversion coefficient and gain compensation mapping function. For parts not described in detail in this embodiment, please refer to the relevant content in Embodiment 1.

[0076] The process of obtaining the preset concentration conversion coefficient establishes a quantitative correlation between the intensity of the corrected Raman characteristic peak and the mass concentration of pesticide residues per unit area. The specific implementation steps are as follows: First, a series of pesticide standard solutions with known concentration gradients are prepared, covering the legal limits for pesticide residue detection in monk fruit. Multiple groups of clean monk fruit samples with uniformly distributed surface hairs are selected as reference samples, and equal amounts of standard solutions of different concentrations are uniformly applied to the sample surface using a quantitative spraying process. The distance sensor described in this invention is used to lock the three-dimensional spatial coordinates, and after local enhancement and polarization difference processing, the net Raman scattering intensity of specific pesticide residue components (such as the fingerprint peak of carbamate pesticides at a specific wavenumber) is extracted.

[0077] Considering the complex matrix effect of the monk fruit peel and the non-uniformity of its high-curvature surface, the system employs a surface matrix matching correction method to extract intensity sample points from different curvature ranges within a three-dimensional coordinate matrix. Using a weighted least squares fitting algorithm, a quantitative mapping curve is constructed with the pesticide residue concentration of each sample point as the independent variable and the corresponding restored true scattering intensity as the dependent variable. This mapping curve follows a linear proportional relationship in the low-concentration range, and its slope is defined as the linear concentration conversion coefficient. In the high-concentration or signal saturation range, the system uses a second-order polynomial regression model for correction. The final generated concentration conversion coefficients are stored in the system memory in the form of a lookup table or parameter matrix. During the actual imaging stage, the grayscale value of the optical intensity is converted into pesticide residue concentration in standard physical units by real-time retrieval of the regression parameters corresponding to the current pixel.

[0078] The attenuation correction model achieves pixel-level restoration of the observed signal by establishing a gain compensation mapping function based on the inverse operation of a multi-factor product. The logical foundation of the gain compensation mapping function is as follows: First, calculate the incident energy attenuation factor: this factor is the absolute value of the cosine of the angle between the object surface normal vector and the system optical axis. It is used to correct for energy dilution caused by changes in the projected area of ​​the beam on the inclined surface.

[0079] Next, the light collection angle loss coefficient is calculated: this loss coefficient is determined by comparing the ratio of the solid angle of the scattered light distribution to the solid angle of the lens acquisition.

[0080] The solid angle of the scattered light distribution is equal to twice pi and the product of "the value minus the cosine of the half-angle of the scattering". The solid angle captured by the lens is determined by the system's numerical aperture, which is equal to the product of twice pi and the square root of the square of the numerical aperture minus the numerical aperture. In this embodiment, the system's numerical aperture is set to 0.5.

[0081] The overall transmission efficiency and compensation gain were then determined: the overall transmission efficiency of the system is the product of the incident energy attenuation factor and the light collection angle loss coefficient. To ensure the stability of signal processing, the value of this transmission efficiency was strictly limited to between 0.05 and 1.

[0082] The gain compensation function is set to the reciprocal of the overall transfer efficiency. When the calculated overall transfer efficiency is not less than 0.05, The system utilizes the calculation logic of the aforementioned gain compensation function, taking the reciprocal of the overall transmission efficiency as the gain value and multiplying it by the original pixel brightness to compensate for energy loss caused by optical path attenuation. The system then applies this gain value to compensate for pixel brightness. For pixels located in boundary regions with extremely low transmission efficiency (i.e., an overall transmission efficiency below 0.05), these pixels are identified as extremely weak signal areas. In this case, the system stops applying the reciprocal compensation logic and instead uses a seven-pixel multiplied seven-pixel neighborhood mean interpolation algorithm for smoothing. The system then uses the mean of the surrounding effectively compensated pixels to reconstruct the brightness of this pixel, preventing excessive amplification of background electrical noise. By setting an efficiency lower limit of 0.05, the system effectively suppresses excessive amplification of photoelectric noise in dark areas while compensating for weak signals, avoiding artifacts in high-tilt regions and ensuring the signal-to-noise ratio for pesticide residue feature extraction.

[0083] The execution process of the gain compensation mapping function is as follows: the incident energy attenuation factor is multiplied pixel by pixel by the light loss weighting coefficient to obtain a loss value representing the overall transmission efficiency of the photoelectric detection link, which is used as the transmission efficiency coefficient. The effective value of this coefficient is between zero and one. To restore the true distribution intensity of pesticide residue components, the system takes the reciprocal of the transmission efficiency loss coefficient as the pixel enhancement compensation factor. To prevent numerical explosion and noise amplification caused by excessively low loss values ​​at sample edges or extreme concave areas (i.e., positions where the normal angle approaches 90 degrees or the curvature is extremely large), the model presets a reliability gain threshold. When the calculated transmission efficiency coefficient is lower than 10% of the preset proportional threshold, the system automatically switches to the neighborhood mean interpolation algorithm to perform numerical repair on the pixel, instead of directly performing a factor magnification. Finally, by multiplying the original grayscale value in the pesticide residue differential image stream with this dynamic compensation factor, the energy consistency correction of all pixels is achieved, ensuring that the intensity of the imaging result is only controlled by the concentration of pesticide molecules.

[0084] The logic for obtaining the preset proportional threshold is based on the signal-to-noise ratio (SNR) calibration boundary of the photoelectric detection system. In practice, the system first collects the detector's inherent noise floor standard deviation in a completely dark environment without excitation light, using this as the underlying physical benchmark to measure the degree of signal quality degradation. Subsequently, the system establishes a reference reference signal by measuring the response intensity of a standard reference sample at full power. Combining this with the linear response law of the photoelectric detector at different incident energy levels, the system calculates the percentage loss corresponding to the critical level where signal transmission loss causes the real-time signal intensity to drop to ten times the noise floor standard deviation. This percentage is defined as the preset proportional threshold. In real-world applications, this threshold is typically set between 10% and 20%. When the transmission efficiency loss value calculated by the system exceeds the preset ratio threshold, it is determined that the signal quality of the current pixel has entered the nonlinear noise sensitive area. In order to avoid blindly amplifying the signal and introducing significant random speckle noise, the system automatically switches to the neighborhood mean interpolation algorithm and uses the spatial correlation of healthy pixels within a preset radius around the faulty pixel to perform numerical reconstruction and repair. If the loss value is lower than the threshold, the linear gain compensation of the original data is maintained. While ensuring the spatial continuity of the pesticide residue distribution image, the original spectral features of the high signal-to-noise ratio region are preserved with high quality.

[0085] The process of obtaining the preset ratio threshold includes the following specific steps: First, the system continuously acquires 1,000 frames of raw data from the detector under completely dark conditions with all light sources turned off, without any laser excitation. The system then performs statistical analysis on the raw grayscale values ​​of all pixels to calculate the standard deviation of the background noise of the entire detector array.

[0086] Second, clean monk fruit free of pesticide residues was prepared as a reference sample. A complete 3D scan, local enhancement, and polarization difference processing were performed under full-power excitation, and the resulting pesticide residue difference image matrix was recorded. Pixel grayscale values ​​of flat surface regions (specifically, regions with a curvature gradient less than 0.1 mm) were extracted from this matrix, and the average grayscale value of these flat regions was calculated as the reference signal intensity.

[0087] Third, based on the linear response characteristics of photodetectors, a critical signal-to-noise ratio of ten is defined. That is, when the real-time signal strength drops to ten times the standard deviation of the noise floor, the signal is considered to have entered the nonlinear noise sensitive region. At this point, the critical value of transmission loss is calculated, specifically by dividing ten times the standard deviation of the noise floor by the reference signal strength.

[0088] Fourth, the loss threshold calculated above is converted into a percentage value, which yields the preset ratio threshold. Taking a commonly used 12-bit deep optocoupler sensor as an example, if its noise floor standard deviation is about five gray levels, and the reference signal strength of the reference flat area is about five hundred gray levels, then the calculation result is 10%, which is consistent with the preset ratio threshold range described in this embodiment.

[0089] The specific implementation logic of the neighborhood mean interpolation algorithm is as follows: For pixels deemed unreliable (i.e., pixels whose transmission efficiency coefficient is lower than the aforementioned loss threshold), the system first establishes a square neighborhood window centered on the pixel with a radius of three pixels. Within the neighborhood window, pixels with a transmission efficiency coefficient greater than or equal to the loss threshold are selected and defined as healthy pixels.

[0090] If the number of healthy pixels within the neighborhood window is no less than 50% of the total number of pixels in the neighborhood (i.e., in a 7x7 neighborhood containing 49 pixels, at least 25 are healthy pixels), then the spatially weighted average of all healthy pixels is calculated as the repair value. This weighting coefficient is derived using the Gaussian weighting method: the weight value decreases exponentially with the square of the geometric distance between the neighboring pixel and the central pixel to be repaired. The repaired pixel value is the sum of the products of all healthy pixels in the neighborhood and their corresponding weight values, divided by the total weight value.

[0091] If the percentage of healthy pixels in the neighborhood is less than 50% after the initial search, the system will automatically expand the search range, increasing the radius to five pixels and repeating the above judgment. If the percentage of healthy pixels still fails to meet the standard after expanding the radius, the system will mark the pixel as a defective pixel, display it in the final imaging result with a neutral color or transparency, and mark the corresponding spatial coordinates in the data output record, clearly indicating to the user that the quantitative detection results of the above imaging location are unreliable.

[0092] Compared to existing pesticide residue detection technologies for monk fruit, this invention achieves the following quantitative technical improvements through the synergistic effects of three-dimensional spatial positioning, capillary pressure gradient field enhancement, polarization difference interference suppression, and optical attenuation correction: The detection sensitivity of this invention is increased by more than 500 times compared to traditional surface-enhanced Raman spectroscopy. Experimental data shows that its detectable concentration limit reaches 0.01 micrograms per square centimeter, effectively capturing extremely small amounts of pesticide residue signals. The system's spatial imaging resolution is accurate to 200 micrometers. Compared to the point-by-point measurement mode of traditional point-scanning spectrometers, the area imaging technology used in this invention expands the coverage area of ​​a single measurement by fifty times, significantly improving the scanning capability for irregular areas on the fruit peel surface. The complete detection process for a single sample, including three-dimensional positioning, signal enhancement, polarization difference processing, attenuation correction, and image reconstruction, takes only about three minutes. Compared to existing methods that typically require two to three hours for central laboratory testing, this invention improves detection efficiency by more than forty times, providing a technical advantage for rapid on-site screening. For the determination of pesticide residue concentration, the relative error (i.e., relative standard deviation) is strictly controlled within ±8%. This accuracy fully complies with the national food safety standards for maximum pesticide residue limits in food, meeting the stringent requirements for the accuracy of agricultural product quality and safety testing.

[0093] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for suppressing the background fluorescence of pesticide residues in monk fruit based on Raman spectroscopy, characterized in that, include: The distribution data of hair tissue in the test area is obtained by using a distance sensor, and the spatial filtering characteristics of the confocal optical path are combined to drive the light beam through the gap in the test area and lock the three-dimensional pixel coordinate matrix. Based on the three-dimensional pixel coordinate matrix, the local curvature distribution features are extracted and mapped to the capillary pressure gradient field. The capillary action is used to promote the enrichment of noble metal nanoprobes, construct local enhanced hot spots, generate local electromagnetic field enhanced Raman scattering signals, and output enhanced Raman signal stream of pesticide residue components. Using the frequency-shifting excitation method, the laser temperature is controlled to output lasers of different wavelengths to irradiate the area under test. Based on the depolarization physical characteristics of the fuzzy fiber, polarization phase detection is performed on the enhanced Raman signal stream for different wavelengths of laser. The light intensity data under orthogonal polarization states is captured by the area array sensor. The polarization difference method is executed to remove scattering interference, obtain the pesticide residue signal matrix, and perform differential processing by the image subtraction algorithm to output the pesticide residue differential image stream. By using a distance sensor to extract the curvature gradient of the area to be measured, and combining it with the differential image stream of pesticide residues, an attenuation correction model for the Raman signal is established. Pixel compensation is then performed on the differential image stream of pesticide residues to complete the imaging of pesticide residue concentration.

2. The method for suppressing the fluorescence background of pesticide residues in monk fruit based on Raman spectroscopy according to claim 1, characterized in that, The process of locking the three-dimensional pixel coordinate matrix includes: using a distance sensor to acquire the distribution data of the hairy tissue in the area to be measured and identifying the connecting gaps through which the light beam can pass; driving the light beam through the connecting gaps and using the spatial pinhole in the optical path to perform spatial filtering, filtering out astigmatism and focusing on the surface of the fruit peel; reading the scanning state parameters at the time of focusing in real time, and mapping the scanning state parameters to three-dimensional physical space coordinates based on the geometric projection model, thereby locking the three-dimensional pixel coordinate matrix.

3. The method for suppressing the fluorescence background of pesticide residues in monk fruit based on Raman spectroscopy according to claim 1, characterized in that, The process of outputting the Raman signal stream of pesticide residue components includes: extracting the three-dimensional pixel coordinate matrix for local surface fitting, calculating the curvature characteristic parameters of each pixel in the actual physical space, quantizing the local curvature parameters into capillary pressure values ​​according to the Yang-Laplace equation, and generating a capillary pressure gradient field pointing towards the depression area of ​​the root of the hair; utilizing the physical tendency generated by the capillary pressure gradient field to induce the noble metal nanoprobes to naturally deposit at the root of the hair during the liquid retraction process; constructing local enhanced hot spots in the nano gaps through the electromagnetic coupling effect of the local surface between the nanoprobes; using laser to excite the local enhanced hot spots to generate local plasmon resonance, forming an enhanced electromagnetic field to enhance the Raman scattering intensity of pesticide residue molecules, and obtaining the enhanced Raman signal stream.

4. The method for suppressing background fluorescence of pesticide residues in monk fruit based on Raman spectroscopy as described in claim 1. Its characteristics are The process of obtaining the pesticide residue signal matrix includes: controlling a laser to output excitation light with two slightly offset wavelengths and irradiating the area of ​​the monk fruit to be tested; for each wavelength of excitation light, using a polarization beam splitter to decompose the excited enhanced Raman signal stream into mutually orthogonal parallel polarization components and vertical polarization components; using an area array sensor to capture the two-dimensional light intensity data of the parallel polarization components and vertical polarization components, and combining them with a three-dimensional pixel coordinate matrix to perform spatial feature mapping, generating a corresponding orthogonal polarization image matrix; extracting the depolarization degree feature value of each pixel in the orthogonal polarization image matrix and calculating the corresponding polarization compensation weight coefficient; performing pixel-level weighted difference operation on the orthogonal polarization image matrix using the polarization compensation coefficient to cancel the background interference caused by scattering, extracting the net Raman feature signal of the pesticide residue components, and generating the pesticide residue signal matrix.

5. The method for suppressing the fluorescence background of pesticide residues in monk fruit based on Raman spectroscopy according to claim 1, characterized in that, The steps for outputting a pesticide residue differential image stream include: controlling the laser's operating temperature to a first preset temperature value, outputting a laser of a first wavelength and illuminating the area to be tested, synchronously capturing the Raman and fluorescence mixed signal fed back from the area to be tested using an area array detector, and mapping the signal intensity value to the corresponding pixel space according to the three-dimensional pixel coordinate matrix to reconstruct and generate a first pesticide residue signal matrix; adjusting the laser's operating temperature to a second preset temperature value, switching to a second wavelength and outputting a laser of a second wavelength to illuminating the area to be tested, and repeating the above capture and mapping process while keeping the acquisition parameters of the detection system unchanged until the spatial overlap between the currently acquired pixel feature data and the first pesticide residue signal matrix meets a preset convergence condition, ending the above repetition process and generating a second pesticide residue signal matrix reflecting the same spatial location information; performing a pixel-by-pixel intensity difference operation on the first pesticide residue signal matrix and the second pesticide residue signal matrix using an image subtraction algorithm, canceling static fluorescence background interference through difference, and extracting the pesticide residue Raman feature signal that shifts with wavelength, and outputting the pesticide residue differential image stream reflecting the spatial distribution information of pesticide residue components.

6. The method for suppressing the fluorescence background of pesticide residues in monk fruit based on Raman spectroscopy according to claim 5, characterized in that, The signal intensity values ​​include: grayscale values ​​reflecting the original photosensitive response of each spatial pixel in the test area, stimulated Raman intensity corresponding to the vibrational spectral lines of characteristic molecules of pesticide residues, and continuous broadband fluorescence radiation intensity superimposed on the background of the vibrational spectral lines of characteristic molecules of pesticide residues.

7. The method for suppressing the fluorescence background of pesticide residues in monk fruit based on Raman spectroscopy according to claim 1, characterized in that, The steps for establishing an attenuation correction model for Raman signals include: acquiring real-time surface topography data of the area to be measured, calculating local curvature parameters and surface normal vectors and mapping them to the three-dimensional pixel coordinate matrix, and registering them with the pixel-by-pixel spatial position of the pesticide residue feature differential image stream; calculating the incident energy attenuation factor based on the angle between the surface normal vector and the incident beam, and calculating the loss weight coefficient of the light receiving angle of the detection system in combination with the local curvature parameters; establishing a gain compensation mapping function relationship for Raman signal intensity using the energy projection attenuation factor and the loss weight coefficient as independent variables to construct the attenuation correction model; and using the attenuation correction model to perform intensity compensation and numerical repair on the pesticide residue differential image stream to restore the true distribution intensity of pesticide residue components, so as to output the corrected quantitative detection results of pesticide residues.

8. The method for suppressing the fluorescence background of pesticide residues in monk fruit based on Raman spectroscopy according to claim 1, characterized in that, The steps for performing pixel compensation on the pesticide residue differential image to complete pesticide residue concentration imaging include: using an attenuation correction model to perform pixel-by-pixel intensity gain compensation on the pesticide residue differential image to restore the true scattering intensity at each coordinate point, and combining it with a preset concentration conversion coefficient to complete quantitative spatial imaging of the pesticide residue concentration of monk fruit.

Images