Wafer transport box recognition method and system

Abstract

The application relates to the field of semiconductor technology, in particular to a wafer transport box identification method and system. The application outputs feedback data through a cross-spectrum feature fusion model, calculates and obtains an imaging quality evaluation function according to the feedback data; when the imaging quality evaluation function meets a specified threshold, inputs the double-spectrum image information corresponding to the imaging quality evaluation function into the cross-spectrum feature fusion model to generate a synthesized feature map; according to the cross-spectrum feature fusion model, the comprehensive confidence is calculated and obtained, and the evaluation standard of the comprehensive confidence is established; when the comprehensive confidence meets the evaluation standard, the synthesized feature map corresponding to the comprehensive confidence is obtained, and the actual model of the candidate wafer transport box is parsed according to the synthesized feature map. In the wafer delivery link, the wafer transport box model identification precision and identification efficiency are further improved.

Description

Technical Field

[0001] This application relates to the field of semiconductor technology, and in particular to a wafer transport box identification method and system. Background Technology

[0002] After semiconductor testing is completed, the wafers need to be shipped to ensure the subsequent packaging, transportation, and storage of the wafers. During wafer shipment, strict traceability of wafer data must be ensured to avoid issues such as mixed or lost wafers. Summary of the Invention

[0003] In view of this, the purpose of this application is to propose a wafer transport box identification method and system to solve the above-mentioned technical problems.

[0004] To achieve the above objectives, this application provides a wafer transport box identification method, comprising:

[0005] Based on the target model of the wafer transport box, alternative wafer transport boxes are obtained; Two parallel invisible light sources are used to emit two beams of parallel invisible light of different spectral bands toward the marking area of ​​the candidate wafer transport box. The reflected light from two beams of parallel invisible light of different spectral bands reflected from the marked area is collected to obtain the image information of the dual-spectral image; The image information of the dual-spectral image is input into the cross-spectral feature fusion model, and the cross-spectral feature fusion model is used to output feedback data. The imaging quality evaluation function is calculated and obtained based on the feedback data. When the imaging quality evaluation function meets the specified threshold, the image information of the dual-band image corresponding to the imaging quality evaluation function is input into the cross-band feature fusion model to generate a synthetic feature map; Based on the cross-spectral feature fusion model, the comprehensive confidence level is calculated and obtained, and an evaluation standard is established for the comprehensive confidence level. When the overall confidence level meets the evaluation criteria, the synthetic feature map corresponding to the overall confidence level is obtained, and the actual model of the candidate wafer transport box is parsed from the synthetic feature map. The actual model is compared with the target model. If the actual model matches the target model, the wafer transport box replacement operation is performed. If the actual model does not match the target model, the wafer transport box replacement operation is prohibited.

[0006] Optionally, the reflected light from the two parallel invisible light beams of different spectral bands reflected from the identified area includes: An infrared image is obtained by using an infrared sensor to collect the reflected light of two parallel invisible light beams of different spectral bands reflected from the marked area based on the shadow effect; an ultraviolet image is obtained by using an ultraviolet sensor to collect the reflected light of two parallel invisible light beams of different spectral bands reflected from the marked area based on the fluorescence and / or reflection effect; the dual-spectral image includes the infrared image and the ultraviolet image.

[0007] Optionally, the cross-spectral feature fusion model is a multi-decoder neural network; The image information includes: the structural contour information of the micro-convex structure within the marked area captured by the infrared sensor, and the fluorescence information generated due to the material difference in the marked area captured by the ultraviolet sensor; The step of inputting the image information of the dual-band image into the cross-band feature fusion model to generate a synthetic feature map includes: The structural contour information of the micro-convex structure within the marked area captured by the infrared sensor, and the fluorescence information generated due to the material difference in the marked area captured by the ultraviolet sensor, are input into the multi-decoder structure neural network. The structural contour information and the fluorescence information are fused in the multi-decoder structure neural network to generate the synthetic feature map.

[0008] Optional, also includes: When the imaging quality evaluation function is lower than a specified threshold, the illumination parameters of the two parallel invisible light sources are adjusted and the imaging quality evaluation function is calculated until the imaging quality evaluation function meets the specified threshold.

[0009] Optionally, the step of calculating and obtaining the comprehensive confidence level based on the cross-spectral feature fusion model includes: The specific formula for calculating the overall confidence level Conf is as follows: ; in, , , These are the weighting coefficients. Confidence level for optical character recognition To fuse feature matching scores, For spectral consistency.

[0010] Optionally, establishing evaluation criteria for the overall confidence level includes: A comprehensive confidence evaluation interval is established to perform deterministic measurement on the generated synthetic feature map; The comprehensive confidence level evaluation interval includes: high confidence interval Conf≥0.95; medium confidence interval 0.80≤Conf<0.95; and low confidence interval Conf<0.80. When the overall confidence level is within the high confidence level range, the corresponding synthetic feature map is obtained, and the actual model of the candidate wafer transport box is parsed from the synthetic feature map.

[0011] Optionally, establishing evaluation criteria for the overall confidence level includes: When the overall confidence level is within the medium confidence interval, a secondary identification control is established; Adjust the illumination parameters of the two parallel invisible light sources, and collect the reflected light of the two parallel invisible light beams of different spectral bands reflected from the marked area again to obtain complementary image information of the dual-spectral image; Based on the cross-spectral feature fusion model, the comprehensive confidence level is calculated and obtained. If the comprehensive confidence level is within the high confidence level range, the corresponding synthetic feature map is obtained.

[0012] Optionally, establishing evaluation criteria for the overall confidence level includes: When the overall confidence level is within the low confidence range, a rejection mechanism is established to prohibit the output of the actual model and initiate manual intervention.

[0013] Optionally, the step of resolving the actual model of the candidate wafer transport box based on the synthesized feature map includes: The actual model of the candidate wafer transport box is determined from the synthesized feature map using optical character recognition (OCR) methods.

[0014] Based on the same inventive concept, this application provides a wafer transport box identification system, including: The acquisition module is used to acquire alternative wafer transport boxes according to the target model of the wafer transport box; The light source module includes two parallel invisible light sources for emitting two beams of parallel invisible light of different spectral bands to the marking area of ​​the candidate wafer transport box, respectively. The image acquisition module is used to acquire the reflected light of two beams of parallel invisible light of different spectral bands reflected from the marked area to obtain the image information of the dual-spectral image; An image processing module is used to input the image information of the dual-spectral image into a cross-spectral feature fusion model, use the cross-spectral feature fusion model to output feedback data, calculate and obtain an imaging quality evaluation function based on the feedback data; when the imaging quality evaluation function meets a specified threshold, input the image information of the dual-spectral image corresponding to the imaging quality evaluation function into the cross-spectral feature fusion model to generate a synthetic feature map; calculate and obtain a comprehensive confidence score based on the cross-spectral feature fusion model, and establish an evaluation standard for the comprehensive confidence score; The feature map parsing module is used to obtain the synthetic feature map corresponding to the comprehensive confidence level when the comprehensive confidence level meets the evaluation criteria, and to parse the actual model of the candidate wafer transport box based on the synthetic feature map. The Manufacturing Execution System (MES) module is used to compare the actual model with the target model. If the actual model matches the target model, the wafer transport box replacement operation is executed; if the actual model does not match the target model, the replacement operation is prohibited.

[0015] As described above, the wafer transport box identification method and system provided in this application optimize the identification accuracy of wafer transport boxes by using an imaging quality evaluation function that outputs image information from dual-spectral images. Complementary image information from dual-spectral images is input into a cross-spectral feature fusion model to generate a synthetic feature map. An evaluation standard for the comprehensive confidence score is established to improve the quality of the synthetic feature map and prevent external environmental interference from reducing its identification quality. The actual model number is accurately compared with the target model number for shipment confirmation of the actual model number of the wafer transport box, forming a closed-loop, full-process monitoring system for wafer transport box identification. Ultimately, this application achieves intelligent identification of wafer transport box models while addressing the problems of low identification accuracy, weak anti-interference capability, and lack of closed-loop process control in the existing wafer shipping process. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a wafer transport box identification method provided in an embodiment of this application; Figure 2 This is a schematic diagram of wafer transport box identification in a wafer transport box identification method provided in an embodiment of this application; Figure 3 This is a block diagram of a wafer transport box identification system provided in an embodiment of this application.

[0018] Figure label: 100. MES database; 200. Control center; 300. Industrial camera; 400. Parallel beam; 500. Control area; 600. Box changing mechanism; 700. Light source emitter. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0020] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0021] It should be noted that the method in this embodiment can be executed by a single device, such as a computer or server. The method can also be applied in a distributed scenario, where multiple devices cooperate to complete the process. In such a distributed scenario, one of these devices may execute only one or more steps of the method in this embodiment, and the multiple devices will interact with each other to complete the method.

[0022] After semiconductor testing, the wafer shipping process requires strict matching of the wafer transport box model with customer requirements. The existing workflow is as follows: first, the wafer enters the facility via MES; then, the process card document is read to confirm the customer's required model; next, the transport box model is visually identified from multiple angles; then, the model in the document is compared with the transport box model. If they match, the box is changed; finally, the wafer is shipped out via MES. However, this matching process has the following drawbacks: (1) Selection relies on manual labor and is prone to errors: Operators need to read the written requirements and select the specified model from a variety of similar-looking wafer transport boxes based on memory and visual inspection. This process is highly dependent on personnel experience and attention, and is prone to errors due to negligence, which may lead to the wrong selection of wafer transport box model and bring unknown risks to subsequent wafer transportation.

[0023] (2) Difficulty in identifying wafer transport box model markings: The model markings of the transport boxes specified by the customer are usually printed in white micro-embossed lettering on the white box body, with extremely low contrast. Personnel need to repeatedly adjust the angle of observation under different lighting conditions to barely identify them, which is inefficient and the identification results are unreliable, posing a high risk of misuse.

[0024] (3) Mechanical compatibility risk: There are slight differences in the pins (clamping / holding structures) used to fix wafers inside different models of wafer transport boxes. If the wrong model is used, the wafer is very likely to be misplaced or interfere with the pins when changing boxes due to mechanical incompatibility, which may cause scratches on the wafer edge or surface and result in product scrap.

[0025] (4) The existing system lacks linkage verification between the identification results and MES, which makes it impossible to achieve closed-loop control of the entire process and poses a significant production risk.

[0026] Therefore, this application provides a wafer transport box identification method for automatically and accurately identifying wafer transport box models and preventing misoperation, thereby solving the above-mentioned problems.

[0027] Reference Figure 1 A method for identifying wafer transport boxes includes the following steps: Step S100: Obtain alternative wafer transport boxes according to the target model of the wafer transport box; Step S200: Using two parallel invisible light sources, emit two beams of parallel invisible light of different spectral bands to the identification area of ​​the candidate wafer transport box respectively; Step S300: Collect the reflected light of two parallel invisible beams of different spectral bands reflected from the marked area to obtain the image information of the dual-spectral image; Step S400: Input the image information of the dual-band image into the cross-band feature fusion model, use the cross-band feature fusion model to output feedback data, and calculate and obtain the imaging quality evaluation function based on the feedback data; Step S500: When the imaging quality evaluation function meets the specified threshold, the image information of the dual-band image corresponding to the imaging quality evaluation function is input into the cross-band feature fusion model to generate a synthetic feature map. Step S600: Calculate and obtain the comprehensive confidence level based on the cross-spectral feature fusion model, and establish an evaluation standard for the comprehensive confidence level; Step S700: When the overall confidence level meets the evaluation criteria, obtain the composite feature map corresponding to the overall confidence level, and parse the actual model of the candidate wafer transport box from the composite feature map; Step S800: Compare the actual model with the target model; If the actual model matches the target model, perform a wafer transport box replacement operation; If the actual model does not match the target model, the wafer transport box replacement operation must not be initiated.

[0028] This application optimizes the identification accuracy of wafer transport box markings by using an imaging quality evaluation function that outputs image information from dual-spectral images. Complementary image information from dual-spectral images is input into a cross-spectral feature fusion model to generate a synthetic feature map. An evaluation standard for the comprehensive confidence score is established to improve the quality of the synthetic feature map and prevent external environmental interference from reducing its recognition quality. A precise comparison between the actual model and the target model is used to confirm the actual model of the wafer transport box for shipment, forming a closed-loop, end-to-end monitoring system for wafer transport box markings. Ultimately, this application achieves intelligent identification of wafer transport box models while addressing the problems of low model identification accuracy, weak anti-interference capabilities, and lack of closed-loop process control in the current wafer shipping process.

[0029] Optionally, in step S100, the MES responds to the wafer shipping instruction, obtains the customer code corresponding to the wafer to be shipped, and the target model of the wafer transport box required by the customer, and obtains alternative wafer transport boxes according to the target model of the wafer transport box required by the customer. Meanwhile, the existing wafer transport box models are mostly printed on the box body in a low-contrast form.

[0030] For example, the wafer transport box has a white body, and the model number of the wafer transport box is in white lettering with slightly raised lettering.

[0031] Optionally, in step S200, the two parallel invisible light sources include an infrared light source and an ultraviolet light source. The two parallel invisible light sources are configured to work together through an infrared laser, an ultraviolet laser, an infrared sensor, an ultraviolet sensor, an industrial camera 300, and a collimating lens to generate an image of the marked area.

[0032] Optionally, an infrared laser is configured to emit infrared light to illuminate the marked area, an ultraviolet laser is configured to emit ultraviolet light to illuminate the marked area, an infrared sensor is configured to receive the infrared light reflected from the marked area, and convert the light intensity distribution into an electrical signal, which is then processed by an industrial camera 300 to convert into an infrared image; an ultraviolet sensor is configured to receive the ultraviolet light reflected from the marked area, and convert the light intensity distribution into an electrical signal, which is then processed by an industrial camera 300 to convert into an ultraviolet image; a collimating lens is configured to be used with either the infrared laser or the ultraviolet laser to ensure that the emitted light from the infrared laser or the ultraviolet laser is a parallel beam 400.

[0033] Furthermore, a parallel beam 400 is set to project onto the marked area in a grazing illumination manner, and an industrial camera 300 is used to shield ambient light, allowing only infrared or ultraviolet light to pass through.

[0034] Optionally, the infrared laser has a spectral range of 800nm ​​to 900nm, and the ultraviolet laser has a spectral range of 300nm to 400nm. The beam plane of the parallel beam 400 forms an angle of less than 5 degrees with the surface of the wafer transport box; the industrial camera 300 is an infrared-sensitive camera or an ultraviolet camera, and a narrow-band filter is installed in front of the lens of the industrial camera 300 to achieve complete shielding of ambient light.

[0035] In some embodiments, in step S300, acquiring the reflected light of two parallel invisible lights of different spectral bands reflected by the marked area includes: acquiring the reflected light of two parallel invisible lights of different spectral bands reflected by the marked area using an infrared sensor based on the shadow effect to obtain an infrared image; acquiring the reflected light of two parallel invisible lights of different spectral bands reflected by the marked area using an ultraviolet sensor based on the fluorescence and / or reflection effect to obtain an ultraviolet image; the dual-spectral image includes an infrared image and an ultraviolet image.

[0036] In this embodiment, the infrared sensor obtains infrared images based on the shadow effect, which are used to specifically identify the raised structural features of the wafer transport box marking area. The ultraviolet sensor obtains ultraviolet images based on fluorescence and / or reflection effects, which are used to specifically identify the material difference features of the wafer transport box marking area. By integrating and extracting the identified raised structural features and material difference features, a more comprehensive and accurate feature identification of the wafer transport box surface can be achieved.

[0037] Optionally, step S300 further includes acquiring the reflected light of two parallel invisible beams of different spectral bands reflected from the identified area to obtain a test image of the dual-spectral image; inputting the test image into the cross-spectral feature fusion model for initial identification of the wafer transport box; the cross-spectral feature fusion model outputs feedback data, and the imaging quality evaluation function is calculated and obtained based on the feedback data.

[0038] If the imaging quality evaluation function is lower than the specified threshold, after readjusting the illumination parameters of the two parallel invisible light sources, the reflected light of the two parallel invisible light beams of different spectral bands reflected from the marked area is collected again to obtain and test the test image of the dual-spectral image until the imaging quality evaluation function is not lower than the specified threshold. At this time, the wafer transport box identification process is started.

[0039] Furthermore, the image information of bispectral images is complementary image information.

[0040] Optionally, in step S400, the image information of the dual-band image is input into the cross-band feature fusion model, and the wafer transport box is initialized and identified. The cross-band feature fusion model outputs feedback data, and the imaging quality evaluation function is calculated and obtained based on the feedback data.

[0041] In this embodiment, the wafer transport box is initialized and identified to improve the accuracy of the cross-spectral feature fusion model in acquiring image information of dual-spectral images.

[0042] Furthermore, this embodiment establishes an optimization mechanism based on a closed-loop control system to ensure that the image information of the output dual-spectral image reaches the optimal imaging quality evaluation function, thereby optimizing the recognition accuracy of the wafer transport box identifier. The closed-loop control system is based on feedback control logic to execute feedback control and output feedback data through a cross-spectral feature fusion model. This feedback data is then applied to two parallel invisible light sources to optimize the wafer transport box recognition process. The formula for the closed-loop control system is as follows: ; in, Let F be the angle of incidence of the light beam, and F be the focal point of the lens. Feedback data for the illumination angle, Q represents the feedback data for lens focus, and Q is the image quality evaluation function.

[0043] Here, if the imaging quality evaluation function output by the cross-spectral feature fusion model... At this point, the imaging quality evaluation function Q is below a specified threshold. Then, after adjusting the illumination parameters of the two parallel invisible light sources, local optimization is retried. Here, the illumination parameters include the illumination angle and the lens focus. Afterwards, the reflected light from the two parallel invisible light beams of different spectral bands reflected from the marked area is collected again, and the image information of the bispectral image is obtained and tested. The formula for retried local optimization is as follows: ; ; At this point, the wafer transport box identification process is initiated until the imaging quality evaluation function Q recovers to above the specified threshold or the specified number of iterations is completed.

[0044] Here, in this embodiment, the feedback control logic between feedback data, illumination angle, and lens focus includes: Define the image quality evaluation function Q, where the formula for the image quality evaluation function Q is as follows: ; Where S is the edge sharpness, i.e., the average gradient magnitude; A and B are both weighting coefficients; C is the contrast of the marker region, calculated using the following formula: ; in, The maximum grayscale value within the identified area corresponds to the highlighted portion of the text. This represents the minimum grayscale value within the identified area, corresponding to the darker part of the background. Here, and It is a single-channel grayscale value, and the value range is usually from 0 to 255.

[0045] In the feedback control logic, the illumination angle is adjusted based on the feedback data. The specific physical relationship calculations include: When the incident angle of the beam When the angle between the beam plane of the parallel beam 400 and the surface of the wafer transport box is less than 5 degrees, the light-receiving area of ​​the sidewalls of the micro-convex characters in the marking area changes, thus affecting the contrast response. The calculation formulas include: ; in, The height of the font bulge; The width of the font strokes; The material's reflectivity coefficient is determined by searching for... Maximize the optimal angle This is used to control the stepper motor and fine-tune the posture of the light source.

[0046] In the feedback control logic, the lens focus is adjusted according to the feedback data to achieve automatic focusing of the camera in the image acquisition unit; In lens focus adjustment, the lens focus position F is adjusted. Here, the formula for calculating the image sharpness function is as follows: ; in, This represents the pixel grayscale value at coordinates (x, y). This represents the gradient in the horizontal direction, i.e., the grayscale difference between adjacent pixels. This represents the gradient in the vertical direction. Image sharpness is evaluated here by calculating the sum of squared gradient magnitudes across the entire image. A hill-climbing algorithm is used for searching. Optimal focal plane position corresponding to the peak To drive a voice coil motor or stepper motor to adjust the lens.

[0047] Furthermore, in this embodiment, based on the feedback control results of the closed-loop control system, the illumination angle or lens focus of the two parallel invisible light sources is controlled, or the infrared and ultraviolet bands corresponding to the infrared and ultraviolet sensors are automatically switched, so that the imaging quality evaluation function reaches the optimal imaging quality evaluation function, thereby ensuring the recognition accuracy of the marked area.

[0048] Thus, the closed-loop control system established in this embodiment initializes and identifies the wafer transport box. Combined with the cross-spectral feature fusion model training mechanism, it enables the two parallel invisible light sources to have adaptive adjustment capabilities based on the feedback data of the image information of the dual-spectral images. This makes the adjustment of the light sources intelligent and adaptive, and further enables the two parallel invisible light sources to adapt to transport boxes of different batches and materials, thereby reducing maintenance and upgrade costs.

[0049] In some embodiments, in step S400, the cross-spectral feature fusion model is a multi-decoder structure neural network; The dual-band image information includes the structural contour information of the micro-convex structure within the marked area captured by the infrared sensor, and the fluorescence information generated due to the material difference in the marked area captured by the ultraviolet sensor. Image information from the dual-band image is input into a cross-band feature fusion model to generate a synthetic feature map, including: The structural contour information of the micro-convex structure within the marked area captured by the infrared sensor, and the fluorescence information generated due to the material difference in the marked area captured by the ultraviolet sensor, are input into the multi-decoder structure neural network. The structural contour information and fluorescence information are fused in the multi-decoder structure neural network to generate a synthetic feature map.

[0050] This embodiment features a parallel invisible light source, resulting in extremely high accuracy in image information recognition. It employs infrared-ultraviolet dual-spectral imaging, overcoming the challenge of insufficient contrast under visible light from a physical perspective. The cross-spectral feature fusion model integrates complementary features of shadows and materials, exhibiting strong suppression capabilities against interference from ambient light changes, surface stains, and font wear.

[0051] Optionally, in a parallel invisible light source, a parallel beam 400 is emitted and projected onto the marked area in a grazing illumination manner to obtain the structural contour information of the micro-convex structure in the marked area, as well as the fluorescence information of a specific spectral band generated due to the material difference in the marked area. The cross-spectral feature fusion model is deeply bound to the optical physical process, thereby training and learning the two response modes of structural contour information and fluorescence information.

[0052] Specifically, the two response modes in this embodiment differ from diffuse reflection under visible light, and specifically include the following two cooperative responses based on different physical mechanisms: (1) Infrared spectral response characteristics include: the geometric structure contour of the marked area is collected based on the shadow effect, and 800nm ​​to 900nm near-infrared parallel light is used for grazing illumination.

[0053] The physical mechanism of infrared spectral response is that infrared light has a stable reflectivity on the surface of commonly used wafer transport boxes such as white polypropylene or polycarbonate, and is not easily affected by ambient stray light.

[0054] The infrared spectral response characteristics are as follows: the beam plane of the parallel beam 400 forms an angle of less than 5 degrees with the surface of the wafer transport box; the background area on the surface of the wafer transport box produces specular reflection; the infrared light is not captured by the industrial camera 300 and appears as a dark field; while the sidewalls of the micro-convex characters are perpendicular to the beam direction, producing strong diffuse reflection, forming a bright geometric structure outline on the infrared sensor.

[0055] (2) Ultraviolet spectral response characteristics include: collecting fluorescence information with material differences based on fluorescence and / or reflection effects, and irradiating with 300nm to 400nm ultraviolet parallel light in a grazing manner.

[0056] The physical mechanism of the ultraviolet spectral response is that wafer transport boxes are typically injection molded from chemically resistant engineering plastics. Even if the box body and the lettering are the same color (both are white), there are often slight differences in the formulation between the box body substrate and the material used for printing / injection molding in order to meet the requirements of antistatic properties (addition of carbon nanotubes or antistatic agents) or structural strength.

[0057] The response characteristics of the ultraviolet spectrum are that, under ultraviolet light excitation, some additives (such as whitening agents) absorb ultraviolet light and emit visible light with a longer wavelength in the fluorescence effect. If the formulations of the font and the box body are different, they will exhibit completely different fluorescence brightness. In terms of reflection differences, if neither has fluorescent properties, the difference in refractive index of ultraviolet light on different material surfaces is greater than that of visible light, which can also create a significant difference in grayscale contrast.

[0058] Here, the ultraviolet image is essentially a mapping of the "differences" in the material composition within the identified area, used to address the problem of metamerism that may occur within the identified area.

[0059] Among them, a dual-modal collaborative mechanism is established in the multi-decoder structure neural network to fuse structural contour information and fluorescence information to generate a synthetic feature map.

[0060] Here, the infrared channel is configured to transmit infrared light. The infrared channel is sensitive to physical contours and has strong resistance to interference from minor surface contamination. The ultraviolet channel is configured to transmit ultraviolet light. The ultraviolet channel is sensitive to material composition. When the lettering within the marking area experiences reduced elevation due to wear (resulting in reduced infrared channel effectiveness), the ultraviolet channel can still identify the markings based on material differences.

[0061] Therefore, by fusing the infrared and ultraviolet spectral response features through a cross-spectral feature fusion model, dual feature verification of "physical morphology + material properties" can be achieved to ensure that the marked area maintains a high degree of reliability in identification even under extreme conditions such as wear, oil contamination, and fluctuations in ambient light.

[0062] In some implementations, in step S500, when the imaging quality evaluation function is lower than a specified threshold, the illumination parameters of the two parallel invisible light sources are adjusted and the imaging quality evaluation function is calculated until the imaging quality evaluation function meets the specified threshold.

[0063] Optionally, in the wafer transport box identification process, the reflected light of two beams of parallel invisible light of different spectral bands reflected from the marked area is collected to obtain complementary image information of the dual-spectral image.

[0064] This is because if infrared or ultraviolet sensors are used alone for optical detection in complex industrial environments, there will be a problem of "perception blind spots".

[0065] Specifically, the causes of the "perception blind spot" problem include: (1) When the infrared sensor is deployed alone to collect image information of the marked area based on the shadow effect, although the grazing light can obtain the structural outline of the micro-convex structure in the marked area, when the font in the marked area is worn down and the height difference is extremely small, or when the wafer transport box material absorbs infrared light too strongly, resulting in uneven diffuse reflection and reduced shadow contrast, the structural outline of the micro-convex structure in the marked area may fail to be extracted.

[0066] (2) When a separate ultraviolet sensor is deployed to collect image information of the marked area based on the fluorescence effect, although the font in the marked area can emit fluorescence according to the difference in the wafer transport box material, if the font in the marked area has similar fluorescence characteristics in the ultraviolet spectrum due to ink, or the injection molding material and the bottom material of the wafer transport box (or both have no fluorescence), or the surface is covered with dirt, causing the fluorescence to be blocked, the collected image information will be black or blurry.

[0067] Therefore, it is necessary to collect the reflected light of two parallel invisible beams of different spectral bands reflected from the identified area to obtain complementary image information of the bispectral image.

[0068] Here, complementary image information needs to be orthogonally supplemented at both the physical and feature levels, specifically including: (1) At the feature level, infrared image features can provide high-frequency information on the structural contour of the micro-convex structure in the marked area. The light and dark boundary line of the micro-convex structure in the marked area can be obtained through the shadow effect, thereby outlining the clear font edge, contour and concave-convex depth.

[0069] Meanwhile, infrared light is not sensitive to surface dirt in the marking area. Dust or slight water stains usually do not cause changes in the micro-convex structure in the marking area, and the information of the structural outline of the micro-convex structure in the marking area collected in the infrared image features still exists.

[0070] Ultraviolet image features can provide response information of the wafer transport box material, thereby reflecting the brightness difference (fluorescence or reflection difference) between the font area and the background area due to different chemical or physical structures, similar to material fingerprints.

[0071] (2) At the physical level, ultraviolet light is not sensitive to mechanical wear of the font in the marking area. Even if the font is worn flat (infrared light recognition fails at this time), as long as the residual material composition is different, ultraviolet light can still excite fluorescence, so that the font in the marking area emits fluorescence, which has better spatial robustness and surface robustness.

[0072] Therefore, by orthogonally supplementing complementary image information at both the physical and feature levels, and then reflecting the orthogonally supplemented complementary image information on the synthesized feature map of the cross-spectral feature fusion model to generate a synthesized feature map, the problem of "perception blind spot" that exists when using infrared or ultraviolet sensors alone for optical detection in complex industrial environments can be further eliminated.

[0073] In some embodiments, in step S600, the comprehensive confidence level Conf is calculated and obtained according to the cross-spectral feature fusion model, including: The specific formula for calculating the overall confidence level (Conf) is as follows: ; in, , , These are the weighting coefficients. Confidence level for optical character recognition To fuse feature matching scores, For spectral consistency.

[0074] In this embodiment, a comprehensive confidence level (Conf) is established to evaluate the quality of the generated synthetic feature map.

[0075] Here, in this embodiment, the components are positioned as follows in the formula for calculating the overall confidence level Conf: (1) , , For the weighting coefficients, satisfying The overall confidence level (Conf) is calculated by weighted fusion of three components and serves as the basis for subsequent feedback control and error prevention decisions.

[0076] For example, this system suggests the following values: .

[0077] (2) This reflects the degree of determinism of the feature map parsing module for the character sequence. For a recognition result of length n, the average of the maximum probabilities of each character calculated using the softmax formula is taken. The specific formula is as follows: ; in, This indicates that in optical character recognition, the i-th character is recognized as the result. The probability is calculated using the softmax formula.

[0078] (3) This component represents the similarity between the fused bispectral features and the standard template library. Cosine similarity is mapped to the [0,1] interval using the sigmoid function, as shown in the following formula: ; in, The feature vector extracted from the bispectral image; This is the standard template feature vector corresponding to the wafer transport box model; The temperature coefficient is used to control the discrimination index and can be determined statistically through the validation set. The bispectral image is a fusion image of the infrared and ultraviolet spectral bands.

[0079] (4) To ensure spectral consistency, this component is cross-validated using the consistency of independent identification results from the infrared and ultraviolet spectral bands, as defined below: ; in, At this time, the identification results of the two spectral bands are completely consistent. This occurs when the results are inconsistent but the character regions overlap. This represents the intersection-over-union ratio (IoU) of the character recognition regions in the infrared and ultraviolet images. When the recognition results of the two spectral bands conflict and the localization regions do not overlap, it indicates an anomaly in imaging or registration, and a lower consistency score is given.

[0080] Here, when calculating the overall confidence level Conf, the confidence level is determined using optical character recognition. Primarily based on feature matching degree As a supplement, based on spectral consistency As a cross-validation item.

[0081] In this embodiment, a quality evaluation function is established and the illumination parameters are repeatedly adjusted using the quality evaluation function to adjust the illumination parameters to the optimal value and obtain the bispectral image corresponding to the optimal value, so as to better improve the acquisition accuracy of the bispectral image.

[0082] In some embodiments, in step S700, the evaluation criteria for establishing the comprehensive confidence level Conf include: Establish a comprehensive confidence evaluation interval to perform deterministic measurement on the generated synthetic feature map; The overall confidence level range includes: High confidence interval: Conf ≥ 0.95; The medium confidence interval is 0.80 ≤ Conf < 0.95; Low confidence interval: Conf < 0.80; When the overall confidence level (Conf) is within the high confidence range, the corresponding synthetic feature map is obtained, and the actual model of the candidate wafer transport box is parsed from the synthetic feature map.

[0083] In this embodiment, by establishing an evaluation standard for the comprehensive confidence level Conf, a hierarchical control strategy for wafer transport box identification is implemented, thereby achieving the goal of using the optimal illumination parameters for identification of the identification area of ​​the wafer transport box, and improving the adaptability of wafer transport box identification optimization.

[0084] Here, the hierarchical control strategy includes: calculating and obtaining the comprehensive confidence level Conf based on the cross-spectral feature fusion model, establishing a comprehensive confidence level evaluation interval based on the value of the comprehensive confidence level Conf, and different comprehensive confidence level evaluation intervals correspond to different system response rules.

[0085] In some embodiments, step S700 establishes an evaluation criterion for the overall confidence level, including: When the overall confidence level (Conf) is within the medium confidence interval, a secondary identification control is established. Adjust the illumination parameters of the two parallel invisible light sources, and collect the reflected light of the two parallel invisible light beams of different spectral bands reflected from the marked area again to obtain the complementary image information of the dual-spectral image. Based on the cross-spectral feature fusion model, the comprehensive confidence level Conf is calculated and obtained. If the comprehensive confidence level Conf is within the high confidence interval, the corresponding synthetic feature map is obtained.

[0086] In this embodiment, a secondary recognition control is established to continuously adjust the accuracy of the synthesized feature map through feedback control, thereby improving the generation quality of the synthesized feature map.

[0087] Optionally, if the overall confidence level (Conf) is still below 0.95 after secondary identification, the alarm process will be initiated.

[0088] Furthermore, by recording the changes in the overall confidence level (Conf) after each parameter fine-tuning, a mapping relationship between the illumination parameters and the overall confidence level (Conf) is formed. Based on the mapping relationship, the optimal illumination parameters are obtained. Subsequently, the optimal illumination parameters can be directly called for wafer transport boxes of the same model to further improve the identification efficiency of the wafer boxes.

[0089] In some embodiments, step S700 establishes an evaluation criterion for the overall confidence level, including: When the overall confidence level (Conf) is in the low confidence range, a rejection mechanism is established to prohibit the output of the actual model and initiate manual intervention.

[0090] In this embodiment, by establishing a rejection mechanism, the reliability of the wafer transport box model identification result can be guaranteed, and the adaptive adjustment of the imaging quality evaluation function, illumination angle and lens focus in the closed-loop control system can be realized. Thus, the comprehensive confidence level Conf can be used to realize the deterministic measurement of the output result by the cross-spectral feature fusion model.

[0091] Optionally, when Conf < 0.80, the cross-spectral feature fusion model does not output the specific model of the wafer transport box, but only returns a "recognition failed" status. This rejection mechanism is used to ensure that the downstream box-changing control module only performs actions when the recognition result is reliable, which conforms to the industrial error-proofing design principle.

[0092] In some embodiments, in step S700, when the overall confidence level meets the evaluation criteria, the corresponding synthetic feature map is obtained, and the actual model of the candidate wafer transport box is parsed from the synthetic feature map, including: The actual model of the candidate wafer transport box is extracted from the synthetic feature map using optical character recognition (OCR) methods.

[0093] In this embodiment, an optical character recognition method is used to accurately convert the text information in the generated synthetic feature map into recognizable and editable text data, so as to facilitate subsequent comparison operations and improve the automation and recognition efficiency of wafer transport box identification.

[0094] Optionally, the optical character recognition method in this embodiment specifically includes: Step S710, the image preprocessing stage, includes image registration, image noise suppression, and image contrast enhancement, which are used to optimize the quality of the bispectral image and facilitate image fusion enhancement and feature extraction of the bispectral image. Step S720, Image Fusion Enhancement Stage, includes feature decoupling and feature fusion reconstruction, used to generate synthetic feature maps; Step S730, the feature extraction stage, includes feature extraction based on optical character recognition methods, global fusion feature extraction, and multi-scale feature aggregation, which are used to parse the actual model of the candidate wafer transport box based on the synthesized feature map.

[0095] In this embodiment, the cross-spectral feature fusion model can achieve full-process mandatory error prevention. Through the closed-loop logic design of "identification-comparison-execution-posting", it isolates human operation from key decision-making. Only when the model matches is the replacement box and system posting allowed, thus eliminating the possibility of human error flowing to the next link from the mechanism.

[0096] Optionally, in the image preprocessing stage of step S710, image registration, image noise suppression, and image contrast enhancement are included to optimize the quality of the bispectral image and facilitate image fusion enhancement and feature extraction of the bispectral image.

[0097] Step S711: Image registration; Due to physical installation discrepancies between the infrared and ultraviolet sensors, pixel-level alignment of the infrared and ultraviolet images is first performed. Specific pixel-level alignment methods include: A checkerboard calibration board or a custom dot matrix calibration block is used to pre-calibrate the homography transformation matrix H of the infrared and ultraviolet sensors. During operation, the ultraviolet image is used... Using infrared images as a benchmark Spatial alignment is achieved through perspective transformation of the infrared image. Edges are clipped to eliminate invalid areas.

[0098] Here, the formula for perspective transformation is as follows: ; Step S712: Image noise suppression; The infrared channel uses bilateral filtering to suppress sensor thermal noise while preserving the edges of the micro-convex structure; the ultraviolet channel fluorescence imaging is prone to shot noise, which is processed using nonlocal mean filtering.

[0099] Step S713: Image contrast enhancement; Specifically, since there are shadow contours in infrared images and fluorescent regions in ultraviolet images, this application performs image processing on the different characteristics of shadow contours and fluorescent regions under grazing illumination to enhance image contrast.

[0100] In the infrared image, adaptive histogram equalization was used to enhance the grayscale difference between the convex characters and the box background. Fluorescent areas in the ultraviolet image typically appear bright; a top-hat transform was used to extract bright features and suppress background diffuse reflection.

[0101] Optionally, in the image fusion enhancement stage of step S720, feature decoupling and feature fusion reconstruction are included to generate a synthetic feature map; Step S721, Feature decoupling; The cross-spectral feature fusion model inputs infrared and ultraviolet images of the identified region captured by infrared and ultraviolet sensors and rigorously registered. The fusion AI model employs a dual-encoder structure with shared weights. After extracting infrared and ultraviolet feature maps respectively, a decoupling module separates features orthogonal to "geometric shape" and "material properties." Infrared image features focus on geometric contours, convexity height information, and shadow edges, while ultraviolet image features focus on material properties and differences in surface fluorescence response.

[0102] Thus, this application uses infrared and ultraviolet images to establish a dual-spectral image with complementary image information of the identified region, which can significantly improve the recognition accuracy.

[0103] Step S722: Feature fusion and reconstruction; Specifically, in the cross-spectral feature fusion model, after processing the infrared and ultraviolet images using feature decoupling, the infrared and ultraviolet images are sent to different decoders. The cross-spectral feature fusion model extracts the infrared image features (highlighting edges) and the ultraviolet image features (highlighting regions) respectively. Finally, the infrared image features and ultraviolet image features are combined through a fusion layer to generate an ultra-clear synthetic feature map that contains both clear boundaries and clear material contrast, thereby significantly improving the recognition accuracy.

[0104] In this process, features that are decoupled and orthogonal to geometric shape and material properties are cascaded into channels and input into a multi-decoder structure, which includes a main decoder and an auxiliary decoder.

[0105] The main decoder is used to generate ultra-clear synthetic feature maps, highlighting character outlines and material contrast.

[0106] The auxiliary decoder is used to output a confidence mask, marking regions with high reliability in the fused image for the feature map parsing module to focus on.

[0107] The bispectral image preserves the shadow contours in the infrared image while superimposing the fluorescence region enhancement effect in the ultraviolet image, making low-contrast characters appear clearly distinguishable in the bispectral image.

[0108] Furthermore, in this embodiment, relying on the optical physics rendering engine, parameters such as material optical properties, protrusion height of the marked area, parallel light wavelength and illumination angle are input to accurately simulate and generate synthetic feature maps of the corresponding spectral bands, which are used for training and optimization of cross-spectral feature fusion models. Here, the rendering steps of the optical physical rendering engine include: (1) Geometric modeling: construct a microstructure mesh for the marking area to accurately represent the height of the font protrusion and the shape of the sidewall.

[0109] For example, the font protrusion height is 0.2mm to 0.5mm.

[0110] (2) Optical simulation: In the infrared channel, based on ray tracing, the formation of shadows and the distribution of diffuse reflection under parallel light grazing conditions are simulated, and grayscale images are output. In the ultraviolet channel, based on the fluorescence excitation model, the difference in luminescence intensity of different materials under ultraviolet irradiation is simulated, and fluorescence images are output.

[0111] (3) Degradation simulation, superimposed with real working condition interference, and randomly added surface dirt, wear-induced attenuation of protrusion height, environmental stray light noise, etc.

[0112] (4) Image synthesis generates paired bi-spectral synthesized feature maps and automatically labels the corresponding model tags for cross-spectral feature fusion model pre-training or data augmentation.

[0113] Optionally, step S730, the feature extraction stage, includes feature extraction based on optical character recognition methods, global fusion feature extraction, and multi-scale feature aggregation, used to parse the actual model of the candidate wafer transport box from the synthesized feature map.

[0114] Step S731: Feature extraction based on optical character recognition method; Character region features are extracted from bispectral images using optical character recognition features. Specific extraction methods include: A differentiable binarization network is used to detect character regions and extract sequence features end-to-end. Convolutional feature vectors are extracted for each character position within the character region and fed into an attention mechanism decoder. The output is a probability distribution based on the character category, and character region features are extracted based on the probability distribution results.

[0115] Step S732: Global fusion feature extraction; Used for fusing feature matching degree Feature vector extraction, specifically including: The dual-spectral image is input into a pre-trained lightweight backbone network. Before the global pooling layer, a spatial attention mechanism is introduced to make the model focus on the labeled region and output a 512-dimensional normalized feature vector vextract. Cosine similarity is calculated with the standard feature vector in the template library, and global fusion features are extracted based on the similarity calculation structure.

[0116] Step S733: Multi-scale feature aggregation; To address the issue of varying character sizes in shipping box models, a feature pyramid structure is employed to extract low-level detail features (capturing micro-convex textures) and high-level semantic features (capturing the overall character structure), which are then weighted and fused to form the final feature representation.

[0117] Therefore, this application performs image fusion enhancement processing and feature extraction on the acquired dual-band images, and uses the feature map parsing module to parse out the actual wafer transport box model.

[0118] Optionally, in step S800, the actual model number is compared with the target model number. If the actual model number matches the target model number, the wafer transport box replacement operation is performed; if the actual model number does not match the target model number, the wafer transport box replacement operation is prohibited, including: If the actual model matches the target model, the comparison result will be stored in the MES database 100. Whether to generate a log record indicating whether the actual model matches the target model; If a posting operation for wafer shipment is generated and executed through MES.

[0119] The wafer transport box identification method in this embodiment improves the identification efficiency of wafer transport box labels, making the identification of wafer transport box labels traceable. The automatic identification speed and box changing speed of wafer transport box labels are much faster than manual identification. At the same time, the operation logs of consistent and inconsistent identification are synchronized to the MES database in real time, realizing the digitalization and full traceability of the production process.

[0120] Specifically, the actual model is compared with the target model. If the actual model and the target model are the same, it is determined to be consistent, and an instruction to allow use is generated to execute the wafer transport box replacement operation. If the actual model and the target model are inconsistent, it is determined to be inconsistent, and the replacement operation is prohibited. The comparison result is stored in the MES database 100. It is determined whether a log record of the actual model and the target model being consistent is generated. If it is generated, the wafer shipment posting operation is executed through MES. If it is not generated, MES prohibits outbound shipment.

[0121] The comparison results are stored in MES database 100, and the updated log data in MES database 100 is retrieved using LotNo (Lot Number, batch number) as the index. Simultaneously, if the actual model does not match the target model, an error alarm command is generated, the box replacement control module automatically shuts down and is prevented from starting, requiring a physical key to restart, a buzzer alarm is triggered, and the current operating station is locked.

[0122] In addition, this application is equipped with a safety light curtain or protective cover, and the light source is turned on only during detection to avoid potential risks of eye damage.

[0123] Reference Figure 3 As shown, based on the same inventive concept, this embodiment also provides a system based on a wafer transport box identification method, including the following modules: The acquisition module is used to acquire alternative wafer transport boxes based on the target model of the wafer transport box; The light source module includes two parallel invisible light sources, which are used to emit two beams of parallel invisible light of different spectral bands to the marking area of ​​the candidate wafer transport box respectively; The image acquisition module is used to acquire the reflected light of two parallel invisible light beams of different spectral bands reflected from the marked area, and obtain the image information of the dual-spectral image; The image processing module is used to input image information from dual-spectral images into a cross-spectral feature fusion model, output feedback data from the cross-spectral feature fusion model, calculate and obtain an imaging quality evaluation function based on the feedback data, and input the image information of the corresponding dual-spectral image into the cross-spectral feature fusion model to generate a synthetic feature map when the imaging quality evaluation function meets a specified threshold. Based on the cross-spectral feature fusion model, the module calculates and obtains the comprehensive confidence score and establishes an evaluation standard for the comprehensive confidence score. The feature map parsing module is used to obtain the synthetic feature map corresponding to the comprehensive confidence level when the comprehensive confidence level meets the evaluation criteria, and to parse the actual model of the candidate wafer transport box from the synthetic feature map. The Manufacturing Execution System (MES) module is used to compare the actual model with the target model. If the actual model matches the target model, the wafer transport box replacement operation is performed; if the actual model does not match the target model, the replacement operation is prohibited.

[0124] In this embodiment, a system based on a wafer transport box identification method is established to transform the manual identification of wafer transport box markings into an automated wafer transport box marking identification process. This avoids problems such as easy identification errors, difficulty in marking identification, mechanical compatibility risks, and lack of linkage between control systems that occur during manual identification, thereby further improving the accuracy and efficiency of wafer transport box marking identification.

[0125] Reference Figure 2 In this embodiment, control software is integrated into the control center 200 to control the acquisition module to acquire the target model of the wafer transport box and screen out candidate wafer transport boxes. Then, parallel beams 400 based on infrared and / or ultraviolet light are emitted to the marked area by the light source emitter 700 (infrared laser and / or ultraviolet laser) in the light source module, respectively. Figure 2 The parallel beam 400 generates a control area 500. Through the image acquisition module in the industrial camera 300, a parallel invisible light source is configured to acquire the reflected light of two parallel invisible lights of different spectral bands reflected from the marked area, thereby obtaining dual-spectral image information.

[0126] Here, the industrial camera 300 integrates an infrared sensor and / or an ultraviolet sensor. The test module of the control center 200 initializes the identification of the wafer transport box and outputs a comprehensive confidence score Conf to optimize the identification process of the wafer transport box. In the image processing module and the feature map analysis module, a cross-spectral feature fusion model is used to perform image acquisition on the identification area of ​​the candidate wafer transport box, generate a synthetic feature map, and parse out the actual model of the candidate wafer transport box. The manufacturing execution system module includes a MES database 100. The box changing mechanism 600 is linked with the MES database 100 to compare the actual model with the target model. If the actual model matches the target model, the wafer transport box changing operation is performed; if the actual model does not match the target model, the box changing operation is prohibited.

[0127] It should be noted that the above description describes some embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0128] The embodiments of the various products and devices in this application are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For ease of description, the above devices are described in functional modules. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware.

[0129] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application (including the claims) is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the above embodiments of this application, which are not provided in detail for the sake of brevity.

[0130] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0131] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description.

[0132] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. A method for identifying wafer transport boxes, characterized in that, include: Based on the target model of the wafer transport box, alternative wafer transport boxes are obtained; Two parallel invisible light sources are used to emit two beams of parallel invisible light of different spectral bands toward the marking area of ​​the candidate wafer transport box. The reflected light from two beams of parallel invisible light of different spectral bands reflected from the marked area is collected to obtain the image information of the dual-spectral image; The image information of the dual-spectral image is input into the cross-spectral feature fusion model, and the cross-spectral feature fusion model is used to output feedback data. The imaging quality evaluation function is calculated and obtained based on the feedback data. When the imaging quality evaluation function meets the specified threshold, the image information of the dual-band image corresponding to the imaging quality evaluation function is input into the cross-band feature fusion model to generate a synthetic feature map; Based on the cross-spectral feature fusion model, the comprehensive confidence level is calculated and obtained, and an evaluation standard is established for the comprehensive confidence level. When the overall confidence level meets the evaluation criteria, the synthetic feature map corresponding to the overall confidence level is obtained, and the actual model of the candidate wafer transport box is parsed from the synthetic feature map. The actual model is compared with the target model. If the actual model matches the target model, the wafer transport box replacement operation is performed. If the actual model does not match the target model, the wafer transport box replacement operation must not be initiated.

2. The wafer transport box identification method according to claim 1, characterized in that, The reflected light from the two parallel invisible beams of different spectral bands reflected from the identified area includes: An infrared image is obtained by using an infrared sensor to collect the reflected light of two parallel invisible light beams of different spectral bands reflected from the marked area based on the shadow effect; an ultraviolet image is obtained by using an ultraviolet sensor to collect the reflected light of two parallel invisible light beams of different spectral bands reflected from the marked area based on the fluorescence and / or reflection effect; the dual-spectral image includes the infrared image and the ultraviolet image.

3. The wafer transport box identification method according to claim 2, characterized in that, The cross-spectral feature fusion model is a multi-decoder neural network; The image information includes: the structural contour information of the micro-convex structure within the marked area captured by the infrared sensor, and the fluorescence information generated due to the material difference in the marked area captured by the ultraviolet sensor; The step of inputting the image information of the dual-band image into the cross-band feature fusion model to generate a synthetic feature map includes: The structural contour information of the micro-convex structure within the marked area captured by the infrared sensor, and the fluorescence information generated due to the material difference in the marked area captured by the ultraviolet sensor, are input into the multi-decoder structure neural network. The structural contour information and the fluorescence information are fused in the multi-decoder structure neural network to generate the synthetic feature map.

4. The wafer transport box identification method according to claim 1, characterized in that, Also includes: When the imaging quality evaluation function is lower than a specified threshold, the illumination parameters of the two parallel invisible light sources are adjusted and the imaging quality evaluation function is calculated until the imaging quality evaluation function meets the specified threshold.

5. The wafer transport box identification method according to claim 1, characterized in that, The step of calculating and obtaining the comprehensive confidence level based on the cross-spectral feature fusion model includes: The specific formula for calculating the overall confidence level Conf is as follows: ； in, , , These are the weighting coefficients. Confidence level for optical character recognition To fuse feature matching scores, For spectral consistency.

6. The wafer transport box identification method according to claim 5, characterized in that, The establishment of evaluation criteria for the comprehensive confidence level includes: A comprehensive confidence evaluation interval is established to perform deterministic measurement on the generated synthetic feature map; The comprehensive confidence level evaluation interval includes: high confidence interval Conf≥0.95; medium confidence interval 0.80≤Conf<0.95; and low confidence interval Conf<0.

80. When the overall confidence level is within the high confidence level range, the corresponding synthetic feature map is obtained, and the actual model of the candidate wafer transport box is parsed from the synthetic feature map.

7. The wafer transport box identification method according to claim 6, characterized in that, The establishment of evaluation criteria for the comprehensive confidence level includes: When the overall confidence level is within the medium confidence interval, a secondary identification control is established; Adjust the illumination parameters of the two parallel invisible light sources, and collect the reflected light of two beams of parallel invisible light of different spectral bands reflected from the marked area again to obtain complementary image information of the dual-spectral image; Based on the cross-spectral feature fusion model, the comprehensive confidence level is calculated and obtained. If the comprehensive confidence level is within the high confidence level range, the corresponding synthetic feature map is obtained.

8. The wafer transport box identification method according to claim 6, characterized in that, The establishment of evaluation criteria for the comprehensive confidence level includes: When the overall confidence level is within the low confidence range, a rejection mechanism is established to prohibit the output of the actual model and initiate manual intervention.

9. The wafer transport box identification method according to claim 1, characterized in that, The step of resolving the actual model of the candidate wafer transport box based on the synthesized feature map includes: The actual model of the candidate wafer transport box is determined from the synthesized feature map using optical character recognition (OCR) methods.

10. A wafer transport box identification system, characterized in that, include: The acquisition module is used to acquire alternative wafer transport boxes according to the target model of the wafer transport box; The light source module includes two parallel invisible light sources for emitting two beams of parallel invisible light of different spectral bands to the marking area of ​​the candidate wafer transport box, respectively. The image acquisition module is used to acquire the reflected light of two beams of parallel invisible light of different spectral bands reflected from the marked area to obtain the image information of the dual-spectral image; An image processing module is used to input the image information of the dual-spectral image into a cross-spectral feature fusion model, use the cross-spectral feature fusion model to output feedback data, calculate and obtain an imaging quality evaluation function based on the feedback data; when the imaging quality evaluation function meets a specified threshold, input the image information of the dual-spectral image corresponding to the imaging quality evaluation function into the cross-spectral feature fusion model to generate a synthetic feature map; calculate and obtain a comprehensive confidence score based on the cross-spectral feature fusion model, and establish an evaluation standard for the comprehensive confidence score. The feature map parsing module is used to obtain the synthetic feature map corresponding to the comprehensive confidence level when the comprehensive confidence level meets the evaluation criteria, and to parse the actual model of the candidate wafer transport box based on the synthetic feature map. The manufacturing execution system module is used to compare the actual model with the target model. If the actual model matches the target model, the wafer transport box replacement operation is performed. If the actual model does not match the target model, the box replacement operation must not be initiated.

Images