An infant diaper stain detection system based on image recognition

Abstract

The present invention relates to the technical field of diaper stain detection, and particularly to a baby diaper stain detection system based on image recognition. The system includes S1, equipment setup; S2, establishing a thickness compensation coefficient matrix; S3, real-time correction of ultraviolet signals; S4, frequency domain band-stop filtering; S5, visible light assisted verification. The present invention samples the ultraviolet image of a clean diaper by setting a UV-CCD camera, and considering the penetration of stains in the diaper, a thickness compensation coefficient matrix is added to the ultraviolet image to improve the detection accuracy. In addition, due to the reflection of SAP molecules on ultraviolet rays, a frequency domain band-stop filtering step is added to filter out the interference caused by SAP molecules, so that prominent stains can be identified in the ultraviolet image. At the same time, to increase the credibility of stain detection, visible light image assisted verification is added, and through multi-modal fusion judgment, the accuracy of stain recognition is increased.

Description

Technical Field

[0001] The present invention relates to the technical field of diaper stain detection, and in particular to a method for detecting stains on baby diapers based on image recognition. Background Art

[0002] The existing image stain detection method is to set an industrial camera or a depth camera on the production line, establish a target detection module and a deep learning model for the corresponding stains, establish a database and learn according to the input stain samples, so that the target detection module can identify the stain samples during the detection process. However, when this method is applied to the production process of diapers, there is a mismatch phenomenon. Since the background color of diapers is usually white and has a certain thickness, when some inconspicuous stains drop on the surface of the diaper, the color reflected after the stains are absorbed is not obvious. Therefore, the industrial camera's discrimination of this stain is inaccurate and it is easy to skip this stain area. It is considered to add ultraviolet light irradiation on the surface of the diaper, so that under the irradiation of ultraviolet light, due to the spectral characteristics, the stain presents high-brightness fluorescence in the image, thereby identifying the stain position.

[0003] However, all existing diapers on the market will fill SAP molecules (superabsorbent resin particles) in the core layer to improve the water absorption capacity of the diapers. The reflectivity of this SAP molecule to ultraviolet light is as high as 85%, which easily masks the appearance of stains in the image. How to detect stains on diapers with SAP molecules under ultraviolet light has become a problem. Summary of the Invention

[0004] Other features and advantages of the present invention will be described in the following specification, and will become partially obvious from the specification, or can be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in the specification and other specification drawings.

[0005] The purpose of the present invention is to overcome the above deficiencies and provide a method for detecting stains on baby diapers based on image recognition. By installing a pulsed ultraviolet light source on the stain detection production line, the corresponding stains on the surface of the diaper can emit high-brightness fluorescence under the action of ultraviolet light. First, set a UV-CCD camera to sample the ultraviolet image of the clean diaper, and considering the permeability of the stain in the diaper, add a thickness compensation coefficient matrix to the ultraviolet image to improve the detection accuracy. In addition, due to the reflection of the SAP molecule to ultraviolet light, add a frequency domain band-stop filtering step to filter out the interference caused by the SAP molecule, so that prominent stains can be identified in the ultraviolet image. At the same time, to increase the credibility of stain detection, add visible light image auxiliary verification, and through multi-modal fusion judgment, increase the accuracy of stain recognition.

[0006] The present invention provides a method for detecting stains on baby diapers based on image recognition, including:

[0007] S1. Equipment setup: Set up a vision inspection platform on the conveyor belt for stain detection. The vision inspection platform includes a pulsed ultraviolet light source, a beam splitter prism, a UV-CCD camera, a CMOS camera, and a thickness measurement roller linked to the conveyor belt. The computer controls the pulsed ultraviolet light source with a low inclination angle to turn on. When the conveyor belt starts, n non-stained diapers pass through the detection area, and the UV-CCD camera collects ultraviolet images.

[0008] S2. Establish a thickness compensation coefficient matrix: Preprocess the ultraviolet images to establish an ultraviolet intensity distribution map. , and the thickness measurement roller synchronously collects thickness data. , and establish a thickness compensation coefficient matrix. ;

[0009] S3. Real-time correction of ultraviolet signals: Perform pixel-by-pixel compensation on each pixel in the currently detected ultraviolet image.

[0010] S4. Frequency domain band-stop filtering: Convert the compensated image to the frequency domain to generate an amplitude spectrum and a phase spectrum . Locate the SAP interference peak in the amplitude spectrum, apply a Gaussian notch filter, and perform IFFT on the filtered amplitude spectrum and the original phase spectrum to obtain a denoised image , and this denoised image is a reference image without pollution.

[0011] S5. Visible light assisted verification: When a suspected stain is detected in the ultraviolet image, extract the visible light image through the CMOS camera at the same position, extract the LBP texture features from the visible light image, calculate the weighted confidence of the ultraviolet and visible light detection results to confirm whether it is a stain, otherwise initiate a secondary scan.

[0012] In some embodiments, in step S2, the specific formula for the thickness compensation coefficient matrix is:

[0013]

[0014] where is the ultraviolet intensity distribution map, is the average thickness.

[0015] In some embodiments, in step S3, the specific compensation formula for pixel compensation is:

[0016]

[0017] where is the compensated pixel, is a pixel of the ultraviolet image, is the thickness compensation for the pixel point, d(t) is the real-time thickness, T(t) is the current temperature, T0 is the reference temperature, α is the temperature coefficient, and the temperature coefficient = 0.02 / °C.

[0018] In some embodiments, in step S4, the SAP interference peak is at a radial frequency of 0.15 - 0.3, and a Gaussian notch filter is applied to the amplitude spectrum:

[0019]

[0020] Among them, , is the coordinate of the center point in the frequency domain obtained by FFT transformation, , is the coordinate of the SAP interference signal in the frequency domain, is the attenuation rate of the Gaussian function.

[0021] In some embodiments, in step S5, the specific process of the ultraviolet image and visible light image assisted verification is as follows:

[0022] S51. When the ultraviolet image detects a suspected stain, extract its minimum bounding rectangle ROI, and locate the corresponding region of the visible light image through the spatial mapping matrix;

[0023] S52. Extract the texture features of the ROI region of the visible light image, and calculate the LBP-TOP features of the ROI region:

[0024]

[0025]

[0026] Among them, P is the number of neighboring pixels, R is the neighborhood radius, is the gray value of the central pixel, is the gray value of the p-th neighboring pixel, is the sign function, which has only two results, 0 and 1, is assigned;

[0027] After calculation, a 256-dimensional feature histogram is generated, the contrast and energy features of the multi-directional gray-level co-occurrence matrix are extracted, and the texture features are obtained;

[0028] S53. Color space conversion, convert the RGB image to the HSV space, and calculate the mean and standard deviation of the saturation S and brightness V of the ROI region;

[0029]

[0030] Wherein, N is the total number of pixels in the ROI region, is the saturation value of the i-th pixel, is the lightness value of the i-th pixel, , are the average values of saturation and lightness in the ROI region, reflecting the concentration and brightness level of the overall color; represents the saturation standard deviation, represents the lightness standard deviation, and abnormal color patches are detected based on the numerical difference;

[0031] S54, Dual-modal decision fusion, input ultraviolet confidence , visible light feature score , and the judgment result of the final stain is output through weighting. The confidence weighting rule is:

[0032]

[0033] is the ultraviolet confidence, obtained through the normalization process of the ultraviolet image, is the output probability of the SVM classifier for visible light features, is the score after confidence weighting, and are the optimal weights determined by the ROC curve, = 0.7, = 0.3.

[0034] In some embodiments, in step S1, the pulsed ultraviolet light source is set at a low inclination on the side of the stain detection conveyor belt, and the incident angle on the surface of the diaper is 15° ± 2°. The incident light is divided into two paths by a beam splitting prism. One path is narrowband ultraviolet light of 395 nm, and a UV-CCD camera is set above the ultraviolet irradiation area. A 400 nm long-pass filter is added to the lens of the UV-CCD camera. The other path is visible light, and a CMOS camera is set above the visible light irradiation area. The UV-CCD camera and the CMOS camera are started asynchronously for acquisition, so that the two cameras collect images of the same area during the operation of the stain detection conveyor belt.

[0035] In some embodiments, a fail-safe mechanism is established. First, the background drift is monitored, and a reference image is automatically collected every 30 minutes. If > the set threshold, an alarm is triggered to remind the production personnel that there is too large an error in the background image used as the pollution-free reference; a compensation failure fallback mechanism is also set. When the surface temperature of the diaper > 50°C or the thickness fluctuation measured by the thickness measurement roller > 20%, an alarm is triggered to remind the production personnel that the current production temperature is too high or the thickness difference of the diaper is too large.

[0036] By adopting the above technical solutions, the beneficial effects of the present invention are:

[0037] In the present invention, a pulsed ultraviolet light source is installed on the stain detection production line, so that the corresponding stains on the surface of the diaper can emit high-brightness fluorescence under the action of ultraviolet light. First, a UV-CCD camera is set to sample the ultraviolet image of the clean diaper. Considering the permeability of the stains in the diaper, a thickness compensation coefficient matrix is added to the ultraviolet image to improve the detection accuracy. In addition, due to the reflection of the SAP molecules on the ultraviolet light, a frequency-domain band-stop filtering step is added to filter out the interference caused by the SAP molecules, so that prominent stains can be identified in the ultraviolet image. At the same time, to increase the credibility of stain detection, visible light image-assisted verification is added, and through multi-modal fusion determination, the accuracy of stain recognition is increased.

[0038] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure.

[0039] Undoubtedly, such objects of the present invention and other objects will become more apparent after the details of the preferred embodiments described in the following with multiple drawings and illustrations.

[0040] To make the above and other objects, features, and advantages of the present invention more obvious and understandable, one or more preferred embodiments are specifically given below, and in conjunction with the accompanying drawings shown, the detailed description is as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention.

[0042] In the drawings, the same components are denoted by the same reference numerals, and the drawings are schematic and not necessarily drawn to actual scale.

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only one or several embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on such drawings without creative efforts.

[0044] Figure 1 It is a schematic diagram of the overall process of the stain detection system in some embodiments of the present invention;

[0045] Figure 2 It is a schematic diagram of the dual-image assisted verification process in some embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below in conjunction with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

[0047] In addition, in the description of the present invention, it should be understood that the orientation or positional relationship indicated by terms such as "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "axial", "radial", "circumferential", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention.

[0048] In the present invention, unless otherwise clearly specified and defined, terms such as "installation", "connection", "connection", "fixation", etc. should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or integrated; it can be directly connected, or indirectly connected through an intermediate medium, and can be the communication between two elements or the interaction relationship between two elements. However, indicating a direct connection means that there is no connection relationship constructed through a transition structure between the two connected main bodies, and they are only connected through the connection structure to form a whole. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0049] In the present invention, unless otherwise clearly specified and defined, the first feature being "on" or "under" the second feature can be that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0050] Refer to Figure 1 - Figure 2 , Figure 1 is a schematic diagram of the overall process of the stain detection system in some embodiments of the present invention; Figure 2 is a schematic diagram of the dual-image assisted verification process in some embodiments of the present invention.

[0051] According to some embodiments of the present invention, the present invention provides a method for detecting stains on baby diapers based on image recognition, which is characterized by including:

[0052] S1. Equipment setup: Build a vision detection platform on the stain detection conveyor belt. The vision detection platform includes a pulsed ultraviolet light source, a beam splitter prism, a UV-CCD camera, a CMOS camera, and a thickness measurement roller linked to the conveyor belt. The computer controls the pulsed ultraviolet light source with a low inclination angle to turn on. When the conveyor belt starts, n stain-free diapers pass through the detection area, and the UV-CCD camera collects ultraviolet images.

[0053] The pulsed ultraviolet light source is set with a low inclination angle on the side of the stain detection conveyor belt, and the incident angle on the surface of the diaper is 15° ± 2°. The incident light is divided into two paths by the beam splitter prism. One path is narrowband ultraviolet light of 395 nm. A UV-CCD camera is set above the ultraviolet irradiation area, and a 400 nm long-pass filter is added to the lens of the UV-CCD camera. The other path is visible light. A CMOS camera is set above the visible light irradiation area. The UV-CCD camera and the CMOS camera start asynchronous acquisition, so that the two cameras collect images of the same area during the operation of the stain detection conveyor belt.

[0054] The ultraviolet light is irradiated on the surface of the diaper at a very low angle. Using the specular reflection characteristics of SAP molecules, most of the reflected light is directed to the non-imaging area, reducing the direct reflection of 85% of SAP into the camera. At the same time, a 400 nm long-pass filter is added to the lens of the UV-CCD camera, allowing only fluorescent stains with wavelengths of 420 - 650 nm to pass through, completely blocking the 365 nm light reflected by SAP molecules, effectively suppressing the reflected light of SAP molecules to ultraviolet light, and making the stain images collected by the UV-CCD camera accurate.

[0055] S2. Establish a thickness compensation coefficient matrix: Preprocess the ultraviolet images to establish an ultraviolet intensity distribution map , and the thickness measurement roller synchronously collects thickness data , establish a thickness compensation coefficient matrix , and compensate for the problem of unclear fluorescence reflection of stains caused by depth through the thickness compensation coefficient matrix;

[0056] The thickness compensation coefficient matrix The specific formula is:

[0057]

[0058] Among them, is the ultraviolet intensity distribution map, is the average thickness.

[0059] S3. Real-time correction of ultraviolet signal: Perform pixel-by-pixel compensation on each pixel in the currently detected ultraviolet image to form a complete image;

[0060] The specific compensation formula for pixel compensation is:

[0061]

[0062] where, is the compensated pixel, is the pixel of the ultraviolet image, is the thickness compensation for the pixel point, d(t) is the real-time thickness, T(t) is the current temperature, T0 is the reference temperature, α is the temperature coefficient, and the temperature coefficient = 0.02 / °C.

[0063] S4. Frequency-domain band-stop filtering: Convert the compensated image to the frequency domain to generate the amplitude spectrum and the phase spectrum , locate the SAP interference peak in the amplitude spectrum, apply a Gaussian notch filter, and perform IFFT on the filtered amplitude spectrum and the original phase spectrum to obtain the denoised image , and this denoised image is the pollution-free reference image;

[0064] The SAP interference peak is at the radial frequency of 0.15 - 0.3. Apply the Gaussian notch filter to the amplitude spectrum:

[0065]

[0066] where, , is the coordinate of the center point of the frequency domain obtained by FFT transformation, , is the coordinate of the SAP interference signal in the frequency domain, is the attenuation rate of the Gaussian function;

[0067] Illustrate with examples , The method for determining the parameters. First, perform a two-dimensional FFT on the ultraviolet image of the clean sample of SAP molecules to generate the amplitude spectrum, observe the bright rings or bright lines in the amplitude spectrum, and record the center coordinates of the bright areas . If the SAP particles vibrate at 120 Hz, the corresponding spatial frequency is:

[0068]

[0069] Through the can be determined.

[0070] S5. Visible light assisted verification: When a suspected stain is detected in the ultraviolet image, the visible light image is extracted through a CMOS camera at the same position. The LBP texture features are extracted from the visible light image, and the weighted confidence of the ultraviolet and visible light detection results is calculated to confirm whether it is a stain. Otherwise, a secondary scan is initiated;

[0071] The specific process of the ultraviolet image and visible light image assisted verification is as follows:

[0072] S51. When a suspected stain is detected in the ultraviolet image, its minimum bounding rectangle ROI is extracted, and the corresponding area of the visible light image is located through the spatial mapping matrix;

[0073] The UV-CCD camera and the CMOS camera are calibrated for the external parameters of the two cameras using a checkerboard calibration board to establish a pixel-level mapping relationship. The calibration process is automatically performed each time the device is powered on:

[0074]

[0075] Among them, is the upper left corner coordinate of the rectangle, is the visible light ROI coordinate, is the visible light ROI coordinate, is the visible light ROI coordinate, is the visible light ROI coordinate, is the visible light ROI coordinate, is the visible light ROI coordinate, are the scale and rotation coefficients calculated by the calibration board, is the translation compensation amount, used to correct the binocular parallax of the two cameras. At the same time, to avoid mapping errors, the visible light ROI is expanded by ±3 pixels; is the translation compensation amount, used to correct the binocular parallax of the two cameras. At the same time, to avoid mapping errors, the visible light ROI is expanded by ±3 pixels;

[0076] S52. Extract the texture features of the ROI area of the visible light image, and calculate the LBP-TOP features of the ROI area:

[0077]

[0078]

[0079] Among them, P is the number of neighborhood pixels, R is the neighborhood radius, is the gray value of the central pixel, is the gray value of the p-th neighborhood pixel, is the sign function, with only two results: 0 and 1, is assignment;

[0080] After calculation, a 256-dimensional feature histogram is generated, and the contrast and energy features of the multi-directional gray-level co-occurrence matrix are extracted to obtain the texture features;

[0081] S53. Color space conversion, convert the RGB image to the HSV space, and calculate the mean and standard deviation of the saturation S and value V in the ROI region;

[0082]

[0083] where N is the total number of pixels in the ROI region, is the saturation value of the i-th pixel, is the value of the i-th pixel, 、 and are the average values of saturation and value in the ROI region, reflecting the concentration and brightness level of the overall color; represents the saturation standard deviation, represents the value standard deviation, and detect abnormal color patches according to the numerical difference;

[0084] Give an example of the scenario of detecting abnormal color patches. For example, the lubricating oil used to lubricate the equipment on the production line accidentally drips on the surface of the diaper during the production process. Although it is transparent and colorless, it will reduce the saturation and increase the value fluctuation;

[0085] First, convert the ROI image of visible light from RGB to HSV, and calculate the statistic:

[0086] , while for the normal area , the saturation drops by 21%; , while for the normal area , the standard deviation doubles. According to the judgment rule, if then it is determined that there is a stain, indicating that there is a stain in the current image. It should be understood that the judgment rule can be adjusted adaptively according to the actual production line and the specific type of diaper, and determined according to the production situation, not a fixed value. This is only an example;

[0087] S54. Bimodal decision fusion, input the ultraviolet confidence , the visible light feature score , and output the final judgment result of the stain through weighted. The confidence weighting rule is:

[0088]

[0089] is the ultraviolet confidence, obtained through the normalization process of the ultraviolet image, is the output probability of the SVM classifier of the visible light feature, is the score after confidence weighting, and are the optimal weights determined by the ROC curve, =0.7, =0.3.

[0090] Preferably, a conflict arbitration mechanism is also set for the ultraviolet results and visible light results, and the established conflict judgment mechanism is shown in Table 1:

[0091] Table 1

[0092] Ultraviolet result Visible light result Action Significance Positive Positive Confirm stain and trigger sorting Bimodal consistency, high confidence Positive Negative Initiate secondary scan Ultraviolet may have false alarms and secondary scan verification is required Negative Positive Record log and conduct manual review Visible light detects missed ultraviolet inspections and algorithm iteration is required

[0093] Preferably, a fail-safe mechanism is also established. First, the background drift is monitored, and a reference image is automatically collected every 30 minutes. If > the set threshold value, an alarm is triggered to remind the production personnel that there is too large an error in the background image used as the pollution-free reference, and the specific calculation formula for updating the reference image is:

[0094]

[0095] During the process of automatically collecting the reference image every 30 minutes, the current image and the background image are automatically fused, and the weight ratio of the current image to the background image is 1:9. is the pixel value of the current image, is the pixel value of the background image, is the pixel value of the newly generated background image;

[0096] A compensation failure fallback mechanism is also set. When the surface temperature of the diaper > 50 °C or the thickness fluctuation measured by the thickness measurement roller > 20%, an alarm is triggered to remind the production personnel that the current production temperature is too high or the thickness difference of the diaper is too large.

[0097] It should be understood that the embodiments disclosed in the present invention are not limited to the specific processing steps or materials disclosed herein, but should extend to equivalent alternatives of such features understood by those of ordinary skill in the relevant art. It should also be understood that the terms used herein are for the purpose of describing specific embodiments only and do not mean to limit.

[0098] The "embodiments" mentioned in the specification mean that the specific features or characteristics described in connection with the embodiments are included in at least one embodiment of the present invention. Therefore, the phrase "embodiments" that appears throughout the specification does not necessarily refer to the same embodiment.

[0099] In addition, the described features or characteristics can be combined in any other suitable way into one or more embodiments. In the above description, some specific details, such as thickness, quantity, etc., are provided to provide a comprehensive understanding of the embodiments of the present invention. However, those skilled in the relevant art will understand that the present invention can be implemented without one or more of the above specific details or can also be implemented using other methods, components, materials, etc.

Claims

1. A method for detecting stains on baby diapers based on image recognition, characterized in that, Including: S1. Equipment setup: Build a vision detection platform on the soiled diaper conveyor belt. The vision detection platform includes a pulsed ultraviolet light source, a beam splitter prism, a UV-CCD camera, a CMOS camera, and a thickness measurement roller linked to the conveyor belt. The computer controls the activation of the pulsed ultraviolet light source with a low inclination angle. When the conveyor belt starts, n non-soiled diapers pass through the detection area, and the UV-CCD camera collects ultraviolet images. S2. Establish a thickness compensation coefficient matrix: Preprocess the ultraviolet image to establish an ultraviolet intensity distribution map , and the thickness measurement roller synchronously collects thickness data , and establish a thickness compensation coefficient matrix ; Thickness compensation coefficient matrix The specific formula is as follows: Among them, is the ultraviolet intensity distribution map, is the average thickness; S3. Real-time correction of ultraviolet signals: Perform pixel-by-pixel compensation on each pixel in the currently detected ultraviolet image. The specific compensation formula for pixel compensation is: Among them, is the compensated pixel, is the ultraviolet image pixel, is the thickness compensation for the pixel point, d(t) is the real-time thickness, T(t) is the current temperature, T0 is the reference temperature, α is the temperature coefficient, and the temperature coefficient = 0.02 / °C; S4, Frequency-domain band-stop filtering: Convert the compensated image to the frequency domain to generate the amplitude spectrum and the phase spectrum , Locate the SAP interference peak in the amplitude spectrum, apply a Gaussian notch filter, and perform IFFT on the filtered amplitude spectrum and the original phase spectrum to obtain the denoised image , The denoised image is the pollution-free reference image; S5. Visible light assisted verification: When a suspected stain is detected in the ultraviolet image, extract the visible light image at the same position through the CMOS camera, extract the LBP texture features from the visible light image, calculate the weighted confidence of the ultraviolet and visible light detection results, and confirm whether it is a stain. Otherwise, initiate a secondary scan.

2. The method for detecting stains on baby diapers based on image recognition according to claim 1, wherein In step S4, the SAP interference peak is at a radial frequency of 0.15 - 0.

3. Apply a Gaussian notch filter to the amplitude spectrum: Among them, , are the coordinates of the center point in the frequency domain obtained by FFT transformation, , are the coordinates of the SAP interference signal in the frequency domain, is the attenuation rate of the Gaussian function.

3. The method for detecting stains on baby diapers based on image recognition according to claim 1, wherein, In step S5, the specific process of ultraviolet image and visible light image assisted verification is as follows: S51. When a suspected stain is detected in the ultraviolet image, extract its minimum bounding rectangle ROI, and locate the corresponding area of the visible light image through the spatial mapping matrix; S52. Extract texture features from the ROI area of the visible light image and calculate the LBP-TOP features of the ROI area: Where P is the number of neighborhood pixels, R is the neighborhood radius, is the gray value of the central pixel, is the gray value of the p-th neighborhood pixel, is the sign function, which has only two results, 0 and 1, is assigned; After calculation, generate a 256-dimensional feature histogram, extract the contrast and energy features of the multi-directional gray-level co-occurrence matrix, and obtain the texture features; S53. Color space conversion, convert the RGB image to the HSV space, and calculate the mean and standard deviation of the saturation S and brightness V of the ROI area; where N is the total number of pixels in the ROI area, is the saturation value of the i-th pixel, is the lightness value of the i-th pixel, , are the average saturation and lightness of the ROI area, reflecting the concentration and brightness level of the overall color; represents the standard deviation of saturation, represents the standard deviation of lightness, and detects abnormal color patches according to the numerical difference; S54. Bimodal decision fusion, input ultraviolet confidence , visible light feature score , and the judgment result of the final stain is output through weighting. The confidence weighting rule is as follows: is the ultraviolet confidence level, obtained through the normalization process of the ultraviolet image, is the output probability of the SVM classifier for visible light features, is the score after confidence level weighting, and is the optimal weight determined through the ROC curve, = 0.7, = 0.

3.

4. The method for detecting stains on baby diapers based on image recognition according to claim 1, wherein, In step S1, the pulsed ultraviolet light source with a low inclination angle is set on the side of the soiled diaper conveyor belt, and the incident angle on the surface of the diaper is 15° ± 2°. The incident light is divided into two paths by the beam splitter prism. One path is narrowband ultraviolet light of 395 nm. A UV-CCD camera is set above the ultraviolet irradiation area, and a 400 nm long-pass filter is installed on the lens of the UV-CCD camera. The other path is visible light. A CMOS camera is set above the visible light irradiation area. The UV-CCD camera and the CMOS camera start asynchronous acquisition, so that the two cameras collect images of the same area during the operation of the soiled diaper conveyor belt.

5. The method for detecting stains on baby diapers based on image recognition according to claim 1, characterized in that A fail-safe mechanism is established. First, background drift is monitored, and a reference image is automatically collected every 30 minutes. If > the set threshold, an alarm is triggered to remind the production personnel that there is too large an error in the background image used as the pollution-free reference. is the pixel value of the background image, is the pixel value of the newly generated background image; a compensation failure fallback mechanism is also set. When the surface temperature of the diaper > 50 °C or the thickness fluctuation measured by the thickness measurement roller > 20%, the production personnel are reminded that the current production temperature is too high or the thickness difference of the diaper is too large.

Images