An SMT patch quality image recognition method and system for chip production

Abstract

This invention relates to the field of image recognition technology, and provides a method and system for SMT (Surface Mount Technology) chip manufacturing quality image recognition. The method includes: adjusting image acquisition parameters according to component type information to acquire detection image information of the component to be inspected; extracting defect image feature information related to defects; determining defect discrimination criteria information according to component type information; making a preliminary judgment on the defects of the component to be inspected based on the defect image feature information and the defect discrimination criteria information, and generating preliminary judgment result information; when the preliminary judgment result information indicates uncertainty, acquiring supplementary detection information and performing multi-source fusion processing and state probability reassessment on the supplementary detection information to reduce judgment uncertainty; and making a final judgment on the defects based on the result of reducing judgment uncertainty, thereby obtaining defect judgment information and quality judgment information. This invention has the effect of improving overall production efficiency and product delivery cycle.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and specifically to a method and system for SMT (Surface Mount Technology) chip manufacturing quality image recognition. Background Technology

[0002] In modern industrial production, especially in the manufacturing of electronic components such as chips, the quality of surface mount technology (SMT) assembly is crucial. Traditional quality inspection methods often struggle to keep pace with the increasing complexity of components and the demand for higher production efficiency. These methods typically rely on manual visual inspection or the use of automated optical inspection (AOI) systems with fixed rules. However, these methods face significant limitations when dealing with minute defects, diverse material properties, and the need for uninterrupted, rapid, and continuous production. In the SMT assembly process, failure to promptly and accurately detect minute defects such as cold solder joints and component misalignment can lead to extensive rework in subsequent production stages, severely impacting overall product quality and manufacturing efficiency.

[0003] To meet the growing market demand for high-performance chips, a chip production line recently introduced a new chip model. This chip's packaging material utilizes a composite coating containing micron-sized metal particles to optimize heat dissipation. Under conventional industrial inspection light sources, the surface reflectance spectrum of this new material exhibits subtle differences compared to the traditional epoxy resin packaging materials used previously. However, existing image recognition systems typically employ a fixed set of camera exposure time control parameters, failing to adjust them in a timely manner. Consequently, when acquiring images of PCB boards containing the new chip, brightness information in certain component areas or surrounding pad areas deviates, resulting in loss of detail in highlighted areas or insufficient image contrast.

[0004] Traditional SMT (Surface Mount Technology) component quality inspection methods often rely on preset image processing rules, such as binarization based on fixed brightness thresholds and edge strength determination by calculating pixel gradients. The robustness of these fixed rules is severely challenged when local brightness unevenness occurs in the image. For example, a slightly underexposed solder joint area may have a brightness value below the preset binarization threshold, leading to a false positive by the system; conversely, a slightly overexposed component edge may have insufficient contrast with the background to achieve the gradient strength required for edge detection calculations, resulting in missed component misalignment. This sensitivity to subtle brightness changes significantly increases the false positive rate of traditional methods, increasing unnecessary rework and overlooking genuine defects.

[0005] This decline in recognition capability is particularly pronounced when detecting minute defects. For example, the system cannot accurately judge the extremely subtle irregularities at the edges of solder joints or the slight offset of components on pads by less than one pixel—areas that fall between acceptable and obvious defects. As a result, many problems that should have been identified and resolved at the beginning of the SMT production process are mistakenly allowed to proceed, accumulating until they surface in the more valuable subsequent packaging and testing stages. This not only leads to a large amount of rework and significantly increases production costs but also severely impacts overall production efficiency and product delivery cycles.

[0006] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0007] This application discloses an image recognition method and system for SMT chip manufacturing quality, aiming to solve the limitations of traditional SMT chip quality inspection methods in handling minute defects, diverse material properties, and uninterrupted, rapid, and continuous production, as well as the problems of existing image recognition systems such as brightness information deviation, poor robustness of fixed rules, high misjudgment rate, and decreased ability to identify minute defects under new packaging materials.

[0008] The technical solution of this application is as follows: In a first aspect, this application discloses an image recognition method for SMT (Surface Mount Technology) chip manufacturing quality, comprising the following steps: By applying various illumination conditions to the component under test, the surface optical characterization information of the component under test is obtained, and the component type information of the component under test is determined based on the surface optical characterization information; Based on the component type information, the image acquisition parameters are adjusted before acquiring the detection image information of the component to be detected; The detected image information is processed to extract defect image feature information related to the defect; Based on the component type information, determine the defect identification criteria information; Based on defect image feature information and defect discrimination standard information, a preliminary judgment is made on the defects of the component to be detected and a preliminary judgment result is generated. When the preliminary judgment results indicate that there is uncertainty in the judgment, supplementary detection information is obtained and multi-source fusion processing and state probability reassessment are performed on the supplementary detection information to reduce the judgment uncertainty. Based on the result of reducing the judgment uncertainty, the defect is finally judged to obtain defect judgment information and quality judgment information.

[0009] Through this technical solution, this application can dynamically adjust the image acquisition parameters and defect discrimination criteria according to the component type, effectively address the detection challenges of different materials and components, and significantly reduce the uncertainty of judgment through multi-source fusion and state probability re-evaluation mechanism, thereby improving the accuracy and reliability of SMT placement quality image recognition, and solving the shortcomings of traditional methods in dealing with new packaging materials and minor defects.

[0010] Secondly, this application also discloses an SMT (Surface Mount Technology) chip manufacturing quality image recognition system for performing SMT chip manufacturing quality image recognition, comprising: The component information acquisition module is used to acquire the surface optical characterization information of the component under test by applying various illumination conditions, and to determine the component type information of the component under test based on the surface optical characterization information; The image acquisition execution module is used to acquire detection image information of the component to be detected after adjusting the image acquisition parameters according to the component type information; The defect feature extraction module is used to process the detected image information and extract defect image feature information related to the defect; The discrimination criterion determination module is used to determine the defect discrimination criterion information based on the component type information; The preliminary judgment execution module is used to make a preliminary judgment on the defects of the component to be detected based on the defect image feature information and defect discrimination standard information, and generate preliminary judgment result information. The quality judgment execution module is used to obtain supplementary detection information and perform multi-source fusion processing and state probability reassessment on the supplementary detection information when the preliminary judgment result indicates that there is uncertainty in the judgment. This reduces the uncertainty in the judgment, and makes a final judgment on the defect based on the result of reducing the uncertainty in the judgment, thus obtaining defect judgment information and quality judgment information.

[0011] Through this technical solution, this application can effectively implement the SMT placement quality image recognition method through modular design. The modules work together to dynamically adjust the detection strategy according to the component type. Through multi-source fusion and state probability re-evaluation mechanism, the accuracy and reliability of defect identification are significantly improved, providing an efficient and intelligent quality control means for chip production.

[0012] Beneficial Effects: This application discloses an image recognition method for SMT (Surface Mount Technology) chip manufacturing quality. By applying various illumination conditions to the component under inspection, it acquires its surface optical characterization information and determines the component type based on this information. This innovative step effectively solves the problems in existing technologies, such as differences in surface reflectance spectra caused by novel packaging materials and the inability of fixed camera exposure parameters to adapt to diverse materials. It avoids the loss of detail or insufficient contrast in bright areas of the image. Subsequently, the image acquisition parameters and defect discrimination criteria are dynamically adjusted according to the component type information, overcoming the problems of poor robustness and high false positive rate of fixed rules in traditional methods when local brightness is uneven. This significantly improves the adaptability and accuracy of the detection. More importantly, when there is uncertainty in the initial judgment result, this method can effectively reduce the judgment uncertainty by acquiring supplementary detection information and performing multi-source fusion processing and state probability reassessment. This solves the problem that traditional methods cannot accurately judge the ambiguous area between acceptable and obvious defects when identifying minor defects. Ultimately, the final decision was made based on the reduced uncertainty in the judgment, which prevented many problems that should have been discovered and resolved at the beginning of the SMT production process from being mistakenly released. This significantly reduced the rework rate, saved production costs, and improved overall production efficiency and product delivery cycle. Attached Figure Description

[0013] Figure 1 This is a flowchart of a method for SMT (Surface Mount Technology) chip manufacturing quality image recognition in one embodiment of the present invention; Figure 2 This is a flowchart of an SMT (Surface Mount Technology) chip manufacturing quality image recognition method according to another embodiment of the present invention. Figure 3 This is a system block diagram of an SMT (Surface Mount Technology) chip manufacturing quality image recognition system according to another embodiment of the present invention. Explanation of reference numerals in the attached figures: 1. SMT placement quality image recognition system for chip manufacturing; 11. Component information acquisition module; 12. Image acquisition execution module; 13. Defect feature extraction module; 14. Judgment standard determination module; 15. Preliminary judgment execution module; 16. Quality judgment execution module. Detailed Implementation

[0014] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0015] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0016] This application proposes an image recognition method for SMT (Surface Mount Technology) chip manufacturing quality, combining... Figure 1 As shown, it includes: S1. By applying various illumination conditions to the component to be tested, the surface optical characterization information of the component to be tested is obtained, and the component type information of the component to be tested is determined based on the surface optical characterization information. S2, after adjusting the image acquisition parameters according to the component type information, acquire the detection image information of the component to be detected; S3, process the detected image information and extract defect image feature information related to the defect; S4. Determine the defect identification criteria information based on the component type information; S5, based on defect image feature information and defect discrimination standard information, makes a preliminary judgment on the defects of the component to be detected and generates preliminary judgment result information; S6. When the preliminary judgment result indicates that there is uncertainty in the judgment, supplementary detection information is obtained and multi-source fusion processing and state probability reassessment are performed on the supplementary detection information to reduce the uncertainty in the judgment. Based on the result of reducing the uncertainty in the judgment, the defect is finally judged to obtain defect judgment information and quality judgment information.

[0017] To make the technical solution of this application easier and clearer to understand, the key terms, data meanings and implementation environment involved are explained below.

[0018] "Components to be inspected" refers to electronic components that require quality inspection during the SMT (Surface Mount Technology) assembly process, including but not limited to chips, resistors, capacitors, and other surface-mount devices. "Surface optical characterization information" refers to data acquired optically to describe the surface optical properties of the components to be inspected. This information may include reflectivity distribution, transmission characteristics, scattering characteristics, and the resulting physical properties such as material composition, surface roughness, or uniformity of the encapsulation coating. "Component type information" is parameter information used to identify the specific category of the components to be inspected. This may include component model, package type, material characteristics, or batch attributes. This information guides the selection of subsequent image acquisition strategies and defect discrimination rules. "Image acquisition parameters" refer to imaging settings that affect the quality of the inspection image, including exposure time, gain, white balance, focal length, or depth of field settings. "Inspection image information" refers to the visual data of the components to be inspected acquired by the image acquisition device under the set image acquisition parameters. "Defect image feature information" refers to visual features extracted from the inspection image information that are related to potential quality defects. This may include edge features, texture features, color features, or brightness distribution features. "Defect discrimination criteria information" refers to the rule parameters or discrimination model used to determine whether a component under inspection has a defect. This can be configured according to different component types and defect types. "Preliminary judgment result information" refers to the initial detection conclusion obtained based on the inspection image information and defect discrimination criteria information. "Judgment uncertainty" refers to the situation where the preliminary judgment result cannot clearly distinguish the component state with a preset confidence level. "Supplementary detection information" refers to auxiliary information obtained through additional detection methods when the judgment is uncertain. "Multi-source fusion processing" refers to the process of comprehensively analyzing and integrating multiple detection information. "State probability reassessment" refers to recalculating and updating the probability of a component being in different states based on multi-source fusion.

[0019] The implementation environment of this application typically includes an automated optical inspection system that integrates various controllable illumination devices, high-resolution image acquisition equipment, image processing units, and inspection decision control units to achieve online or offline quality inspection of SMT surface mount components.

[0020] This application proposes an image recognition method for SMT (Surface Mount Technology) chip manufacturing quality, which improves the accuracy of identifying various types of defects under complex production conditions through a multi-stage, adaptive inspection process. Specifically, firstly, various illumination conditions are applied to the component to be inspected to obtain its surface optical characterization information. These conditions can include ring illumination, coaxial illumination, backlighting, or polarized illumination, etc., acquiring image information of the component under reflection, transmission, or scattering states through different illumination methods. By analyzing the surface optical characterization information, optical response features reflecting the component's packaging material, surface structure, and coating characteristics can be extracted, thereby determining the component type information. For example, when the packaging material of the component contains metal particles of a specific size or a composite material structure, it will exhibit reflection or scattering characteristics different from conventional components under specific wavelengths or polarization conditions, and the system can distinguish the component type based on these characteristics.

[0021] After determining the component type information, the image acquisition parameters are adjusted accordingly, and the detection image information of the component to be inspected is acquired under the adjusted parameter conditions. Different component types differ in surface reflectivity, texture complexity, and pad structure. By specifically adjusting image acquisition parameters such as exposure time, gain, aperture, or focal length, the brightness, contrast, and sharpness of the inspection image can be made suitable for defect identification, thereby reducing the interference caused by overexposure, underexposure, or loss of detail to subsequent analysis.

[0022] Subsequently, the detected image information is processed to extract defect image feature information related to the defects. This processing may include image denoising, contrast enhancement, region segmentation, and feature extraction. Edge detection algorithms are used to obtain component outlines and solder joint boundaries, texture analysis algorithms are used to extract surface texture variation features, and brightness or color distribution analysis is used to identify abnormal areas. Different types of defects have different manifestations at the image feature level. For example, a cold solder joint may appear as blurred solder joint edges or uneven brightness distribution, and component misalignment may appear as a spatial deviation between the component outline and the pad outline.

[0023] Based on this, the corresponding defect discrimination criteria are determined according to the component type information. Since components with different packaging forms, different solder joint structures, and different process conditions have different tolerance ranges for defects, the system can dynamically load or generate corresponding defect discrimination criteria based on the component type for targeted judgment of defect image feature information.

[0024] Then, based on the defect image feature information and defect discrimination criteria information, a preliminary defect judgment is performed on the component to be inspected, generating preliminary judgment result information. The preliminary judgment result information may include defect type identifier, defect location, and corresponding confidence level parameters. When the preliminary judgment result information indicates that the detection conclusion is in the low confidence range, that is, there is uncertainty in the judgment, the system does not directly output the final conclusion, but triggers further detection procedures.

[0025] In cases of uncertainty, supplementary detection information is acquired, and multi-source fusion processing and state probability reassessment are performed based on this information. Supplementary detection information may include image information from different viewing angles, internal structure detection information, or other physical detection information. By fusing the supplementary detection information with the aforementioned surface optical characterization information and detection image information, the probability distribution of the component under different states can be reassessed at a higher information dimension. The state probability reassessment process reduces the uncertainty caused by feature overlap or environmental disturbances in the initial judgment stage, and a final judgment is made on the defect state of the component to be inspected based on the reassessed state probability information, thereby obtaining defect judgment information and quality judgment information.

[0026] Through the aforementioned multi-stage, adaptive detection and judgment mechanism, the method of this application can effectively reduce the false positive rate and false negative rate in complex production environments and mixed scenarios of multiple component types, thereby improving the stability and accuracy of SMT placement quality inspection.

[0027] Optional, combined Figure 2 As shown, when the preliminary judgment result indicates the existence of judgment uncertainty, the steps of obtaining supplementary detection information and performing multi-source fusion processing and state probability reassessment on the supplementary detection information include: A1, Analyze and determine the source of uncertainty, and identify the category of uncertainty source information as at least one of internal material fluctuations, environmental disturbance artifacts, or overlap of minor defects with normal characteristics; A2. Based on the uncertainty source category information, select the target probe operation to reduce the uncertainty of judgment from the preset probe operation library; the preset probe operation library is a pre-established probe operation collection library, which at least stores the type of each probe operation, the applicable uncertainty source category for triggering, and the corresponding execution parameters. A3, Perform target probe operation to obtain supplementary detection information; A4, the supplementary detection information, surface optical characterization information, and detection image information are fused together to obtain fused information; A5, based on fused information, re-evaluates the state probability information of the current region of the element to be detected; A6. Make a final judgment on the defect based on the re-evaluated state probability information. During the final judgment, if the state probability information of one of the defect states is higher than that of the non-defect state, it is judged as a real defect. If the state probability information of the non-defect state is higher than that of the defect state, it is judged as normal. If the state probability information of multiple states is the same, enter the iterative verification process to further reduce the uncertainty of the judgment.

[0028] The analysis and judgment of the source of uncertainty refers to the system's further joint analysis of the preliminary judgment results, original detection image information, and surface optical characterization information after obtaining the preliminary judgment results, in order to identify the root cause of insufficient confidence in the current judgment. Specifically, the system can comprehensively evaluate the distribution characteristics, stability, and matching degree with component type information of defect image feature information based on a preset set of rules, statistical analysis models, or trained machine learning models. This determines whether the uncertainty stems from changes in the component's intrinsic properties, external imaging environment interference, or a high similarity between defect features and normal features. The uncertainty source category information is used to categorize the above causes, which may include, but is not limited to, internal material fluctuations, environmental disturbance artifacts, or overlap between minor defects and normal features. By clearly defining the uncertainty source category information, the system can avoid blindly increasing detection complexity under uncertain conditions, and instead provide a basis for the targeted selection of subsequent detection strategies.

[0029] Among them, internal material fluctuations refer to the surface optical characterization information of the same component that deviates from that of conventional samples under optical imaging conditions due to factors such as differences in component material composition, non-uniformity of packaging structure, or fluctuations in surface treatment process; environmental disturbance artifacts refer to non-defect image anomalies introduced by external factors such as uneven illumination, local reflection, shadow occlusion, dust adhesion, or background stray light; and overlap between micro-defects and normal features refers to the defect scale being close to the resolution limit of the imaging system, or its visual appearance being highly similar to the texture and edge effect of normal materials, making it difficult to reliably distinguish them in the initial judgment stage.

[0030] The preset probe operation library is a set of detection operations pre-established and maintained by the system. It stores various selectable probe operation types and their corresponding execution parameter configurations. Probe operations can include illumination methods in different spectral bands, illumination conditions with different polarization directions, high-magnification or high numerical aperture imaging methods, confocal microscopy, X-ray imaging, thermal imaging, or electrical property testing, etc. Each probe operation in the preset probe operation library is associated with at least one uncertainty source category and records its applicable conditions and corresponding parameter information, such as illumination wavelength, illumination angle, light intensity, exposure time, or magnification. Based on the identified uncertainty source category information, the system selects the target probe operation with the highest matching degree from the preset probe operation library to improve the effectiveness of subsequent supplementary detection information in eliminating judgment uncertainties.

[0031] Performing target probe operations refers to the system driving the relevant detection hardware to perform a second detection of the target area of ​​the component to be inspected, based on the selected probe operation type and its corresponding parameter configuration. This process may include adjusting the illumination mode, switching the imaging device, or activating additional detection units to obtain supplementary detection information that differs from the original detection image information in terms of information dimension or resolution level. Supplementary detection information can manifest as image information obtained under different spectral conditions, higher spatial resolution microscopic morphology information, component internal structural distribution information, or physical property data related to defects.

[0032] Multi-source fusion processing refers to the joint processing of supplementary detection information with previously acquired surface optical characterization information and detection image information to form fused information for decision-making. Multi-source fusion processing can be performed at the data layer, feature layer, or decision layer. By registering, weighting, or verifying consistency among different information sources, the fused information can outperform any single information source in terms of noise resistance, stability, and discriminative ability, thereby enhancing the ability to characterize the true state of the detected element.

[0033] Re-evaluating state probability information based on fused information refers to the system using the fused information to update the probability distribution of the detected element in different states in the current area. State probability information can represent the probability value of an element being in a normal state or multiple defective states. The re-evaluation process is used to correct the biases introduced by insufficient information or feature overlap in the initial judgment stage, making the probability distribution closer to the actual physical state.

[0034] The final judgment process is based on the re-evaluated state probability information. When the probability of a certain defective state is significantly higher than that of other states, the system determines that state as the final judgment result; when the probability of the normal state is dominant, it is judged as a defect-free state; when the probabilities of multiple states are still at similar levels and the preset confidence level condition is not met, the system determines that the current judgment still has uncertainty and enters the iterative verification process, which involves selecting probe operations again and obtaining new supplementary detection information until a clear judgment result is reached or the preset termination condition is met.

[0035] In some preferred embodiments, for example, when performing quality inspection on solder joints of BGA packaged chips, the initial assessment results may indicate a slight shadow feature in a certain solder joint area. This makes it impossible for the system to distinguish whether the shadow is caused by normal solder joint geometric edge effects or by minor solder joint defects. In this case, the system analyzes and determines the source of uncertainty, identifying whether the uncertainty source category may be environmental disturbance artifacts or minor defects overlapping with normal features. Based on this uncertainty source category information, the system selects a matching target probe operation from a preset probe operation library, such as high-angle ring illumination and high-magnification confocal microscopy.

[0036] The system first performs high-angle ring illumination to acquire supplementary detection image information, reducing the impact of shadows and reflections on the imaging results. This supplementary detection image information is then fused with the original surface optical characterization information and the detection image information using multi-source fusion processing. Subsequently, the system re-evaluates the state probability information of the solder joint area based on the fused information. If the re-evaluation result still does not reach the preset judgment confidence level, the system further performs high-magnification confocal microscopy imaging to acquire high-resolution information on the microstructure of the solder joint, and performs multi-source fusion and state probability re-evaluation again. Through the above-mentioned layered and progressive probe selection and fusion judgment mechanism, the system can gradually eliminate judgment uncertainty and ultimately accurately determine the state of the solder joint, thereby effectively avoiding misjudgment and missed judgment.

[0037] Optional, the iterative verification process includes: Acquire surface temperature information, local charge distribution information, and microstructure information of the current region of the component under test; Based on surface temperature information, local charge distribution information, and micromorphological information, the instantaneous or cumulative effects of the preceding probe operation on the surface or internal structure of the element under test are evaluated, and the impact evaluation results are obtained. Based on the type and intensity information of the preceding probe operation, predict the optical artifact information introduced by the preceding probe operation; Based on the impact assessment results and optical artifact information, the parameters of subsequent probe operations are adjusted, and the adjusted subsequent probe operations are executed to obtain supplementary detection information.

[0038] Specifically, in the iterative verification process, the surface temperature, local charge distribution, and microstructure information of the current region of the component under test are first acquired. Surface temperature information can be obtained using an infrared thermal imager or a contact temperature sensor to monitor localized heating that may be caused by probe operation. Local charge distribution information can be obtained using an electrometer or a scanning Kelvin probe microscope to detect charge accumulation or dissipation that may result from probe contact with the component or the interaction of an electric field. Microstructure information can be obtained using high-resolution imaging equipment such as atomic force microscopy (AFM) or scanning electron microscopy (SEM) to observe whether probe operation has caused microstructural changes such as surface scratches, indentations, or material migration. This information comprehensively reflects the changes in the physical state of the component under test before and after probe operation.

[0039] Furthermore, based on the acquired surface temperature, local charge distribution, and microstructure information, the instantaneous or cumulative effects of preceding probe operations on the surface or internal structure of the device under test are evaluated, thereby obtaining information on the impact assessment results. For example, a significant increase in surface temperature may indicate excessive probe energy or prolonged operation; abnormal local charge distribution may indicate a risk of electrostatic damage; and changes in microstructure may signify physical damage. These assessments help quantify the potential impact of preceding operations on the device.

[0040] Simultaneously, based on the type and intensity information of preceding probe operations, the optical artifacts that may be introduced by the preceding probe operations can be predicted. For example, certain probe operations (such as laser scanning and high-intensity illumination) may produce artifacts such as scattering, reflection, or diffraction in the detection image. These artifacts are not defects in the component itself, but may be misjudged as defects. Through pre-established artifact models or empirical data, the characteristics of these artifacts can be predicted according to the probe type (such as light source type and wavelength), intensity (such as light intensity and voltage), and duration.

[0041] Finally, based on the impact assessment results and optical artifact information, the parameters of subsequent probe operations are adjusted, and the adjusted subsequent probe operations are executed to obtain supplementary detection information. For example, if the impact assessment results indicate a potential damage risk, the intensity or duration of subsequent probe operations can be reduced; if specific optical artifacts are predicted, the illumination angle, wavelength, or filtering settings of subsequent image acquisition can be adjusted to reduce artifact interference. Through this adaptive adjustment, the supplementary detection information obtained subsequently can be ensured to be more accurate and reliable, avoiding negative impacts caused by the probe operation itself.

[0042] Optional, the iterative verification process includes: The material type information of the component to be tested, the sequence of probe operations executed and the corresponding parameter information, and the real-time environmental perception data information are obtained to construct the uncertainty state description information of the current area of ​​the component to be tested; For the description information of the uncertainty state, the information gain index expected to be brought about by each probe operation under the current uncertainty state is calculated from the preset probe operation library; the calculation of the information gain index takes into account the instantaneous or cumulative effects caused by the preceding probe operation and the introduced optical artifact information. Based on the information gain metric, the probe operation with the largest information gain and the total execution cost metric meeting the preset threshold is selected; when multiple probe operations have the same information gain metric, the probe operation with the lowest damage risk metric is selected first. Perform the selected probe operation to obtain supplementary detection information; The supplementary detection information is fused with historical detection information to obtain the fused information. Based on the fused information, the state probability information of various states is re-evaluated; When the uncertainty is reduced to the uncertainty threshold or the maximum number of iterations is reached, the iterative verification is terminated and the final judgment result is output.

[0043] Specifically, constructing an uncertainty description of the current region of the component under test refers to the system comprehensively acquiring multi-dimensional data, including the material type information of the component under test, such as its material, structure, and batch characteristics; the sequence of probe operations already executed and their corresponding parameter information, such as which probes were used previously, their intensity, and duration; and real-time environmental sensing data, such as temperature, humidity, and vibration. This information is integrated to form a comprehensive and dynamic context, used to accurately describe the specific manifestations and potential causes of the current uncertainty in the judgment.

[0044] The information gain metric can be understood as the degree to which the uncertainty of a judgment is expected to be reduced after performing a probe operation. This metric is calculated based on the currently constructed uncertainty state description information and evaluates the potential effect of each probe operation from a pre-defined probe operation library. Specifically, when calculating the information gain metric, the instantaneous or cumulative effects that previous probe operations may cause, such as thermal effects, mechanical stress, or charge accumulation, as well as the optical artifacts that these operations may introduce, such as changes in surface reflectivity and local color deviations. By taking these factors into account, the actual effect of probe operations can be predicted more accurately, avoiding erroneous decisions due to ignoring historical influences.

[0045] In practical applications, the selection of probe operations is based on the calculated information gain metric. The system prioritizes probe operations that maximize information gain to ensure that uncertainty is reduced with maximum efficiency in each iteration. Simultaneously, to balance detection efficiency and resource consumption, the total execution cost metric is considered to ensure that the total cost of the selected operations meets a preset threshold. When multiple probe operations have the same information gain metric, to protect the component under test, the system further prioritizes the probe operation with the lowest damage risk metric to minimize unnecessary damage to the component.

[0046] After the selected probe operation is performed, the acquired supplementary detection information is immediately fused with historical detection information. Historical detection information includes, but is not limited to, surface optical characterization information, detection image information, and supplementary detection information acquired in previous iterations. This fusion process aims to integrate all available data to form a more comprehensive and accurate view of the component's state. Based on the fused information, the system re-evaluates the state probability information of various states in the current region (e.g., different types of defects or normal states), thereby updating the confidence level regarding the true state of the component.

[0047] Ultimately, the termination condition for the iterative verification process is set as either the uncertainty being reduced to within a preset uncertainty threshold range, or the preset maximum number of iterations being reached. Once either condition is met, the iterative process terminates, and the final judgment result, including defect judgment information and quality judgment information, is output.

[0048] Optionally, the step of considering the instantaneous or cumulative effects of preceding probe operations and the introduced optical artifact information includes: Acquire information on the surface temperature distribution, local charge distribution, and instantaneous changes in the microstructure of the current region of the component under test; The surface temperature distribution information, local charge distribution information, and instantaneous change data of micromorphology are correlated with the type, intensity, and duration information of the probe operation performed to obtain the correlation analysis results. Based on the correlation analysis results, assess the instantaneous or cumulative impact caused by the preceding probe operation to obtain impact assessment results information; Based on the type and intensity information of the preceding probe operation, the corresponding optical artifact information is predicted. Based on the impact assessment results and optical artifact information, adjust the parameters of subsequent probe operations and execute the adjusted subsequent probe operations; when the impact assessment results indicate that the potential damage risk exceeds the preset threshold, terminate the iterative verification and output an alarm message.

[0049] Specifically, in the iterative verification process, to more accurately consider the impact of preceding probe operations on the component and potential optical artifacts, it is first necessary to acquire information on the surface temperature distribution, local charge distribution, and instantaneous changes in the microstructure of the current region of the component under test. Surface temperature distribution information reflects the local thermal effects that probe operations may cause, local charge distribution information reveals the impact of electrical probe operations or electrostatic accumulation on the component surface, and instantaneous changes in the microstructure directly monitor minute deformations or damage that may be caused by physical probe operations. This information is crucial for assessing the component's condition and potential damage.

[0050] Furthermore, the acquired surface temperature distribution information, local charge distribution information, and instantaneous change data of microstructure are correlated with the type, intensity, and duration information of the performed probe operations to obtain correlation analysis results. This correlation analysis aims to establish a causal relationship model between the probe operation history and changes in the physical state of the component. For example, machine learning algorithms can be used to analyze the specific impact patterns of probe operations of specific types and intensities on the surface temperature, charge, or morphology of the component at a specific duration.

[0051] Based on the correlation analysis results, the instantaneous or cumulative effects of preceding probe operations can be assessed, thus obtaining impact assessment information. This impact assessment information quantifies the physical, chemical, or electrical changes that probe operations may cause to components, such as whether they induce material fatigue, structural stress, or performance degradation. Simultaneously, based on the type and intensity information of preceding probe operations, corresponding optical artifacts can be predicted. Optical artifacts refer to non-realistic defect features that probe operations themselves (such as probe contact, localized heating, or electric field effects) may introduce into the inspection image. If these artifacts are not identified and compensated for, they will seriously interfere with the accuracy of defect judgment.

[0052] Therefore, based on the impact assessment results and optical artifact information, the parameters of subsequent probe operations can be dynamically adjusted, and the adjusted operations can be executed. For example, if the assessment results indicate a slight risk of thermal damage, the power or duration of subsequent probe operations may be reduced; if a specific optical artifact is predicted, the image acquisition parameters or image processing algorithm may be adjusted accordingly to eliminate its impact. As an important safety mechanism, when the impact assessment results indicate a potential damage risk exceeding a preset threshold, the system will immediately terminate the iterative verification and output an alarm message to prevent irreversible damage to the components.

[0053] Optionally, the iterative verification process may also include the following steps: Calculate the execution cost metrics for each probe operation; execution cost metrics include time consumption, energy consumption, and the potential micro-damage to the element being tested; Evaluate the subsequent interference indicators of each probe operation on subsequent operations; By comprehensively considering information gain, execution cost, damage risk, and subsequent interference metrics, probe priority scores are assigned to each probe operation. The probe operation with the highest probe priority score is selected for execution; when multiple probe operations have the same probe priority score, the probe operation with the lowest damage risk index is selected first. Perform the selected probe operation to obtain supplementary detection information, and then perform multi-source fusion processing on the supplementary detection information and historical detection information to update the state probability information; When the uncertainty is reduced to the uncertainty threshold or the maximum number of iterations is reached, the iterative verification is terminated and the final judgment result is output.

[0054] Specifically, in the iterative verification process, to select probe operations more precisely, it is first necessary to calculate the execution cost metric for each probe operation. This execution cost metric is a quantitative assessment of the resources required to perform a specific probe operation, which may include time consumption, energy consumption, and the degree of potential micro-damage to the component under test. For example, time consumption can refer to the time required for the probe operation to start and complete, energy consumption can refer to the electrical energy consumed during the operation, and the degree of potential micro-damage can be a quantitative assessment of the irreversible effects that may be caused to the surface or internal structure of the component.

[0055] Furthermore, it is necessary to evaluate the subsequent interference metrics of each probe operation on subsequent operations. The subsequent interference metric refers to the degree to which a current probe operation affects other probe operations or detection processes that may be performed later. For example, some probe operations may leave residues on the component surface or alter the local physical properties of the component, thereby affecting the accuracy of subsequent optical, electrical, or mechanical probes. Evaluating this metric helps avoid introducing new uncertainties or reducing the effectiveness of subsequent detections due to the current operation.

[0056] Based on this, information gain, execution cost, damage risk, and subsequent interference metrics are comprehensively considered to assign a probe priority score to each probe operation. This probe priority score is a comprehensive quantitative value used to measure the overall merits of each probe operation. Its calculation can employ weighted summation, multi-objective optimization, or other decision-making models to balance the effectiveness of information acquisition, operational economy, component safety, and impact on subsequent processes.

[0057] Therefore, the system selects the probe operation with the highest probe priority score for execution. This selection mechanism ensures that in each iteration, the most optimized and balanced probe operation under the current conditions is selected. As a preferred implementation, when multiple probe operations have the same probe priority score, the probe operation with the lowest damage risk index is selected to maximize the protection of the device under test.

[0058] After performing the selected probe operation, supplementary detection information is acquired and then fused with historical detection information through multi-source processing to update the state probability information. This process is similar to the information fusion and state probability reassessment in the above embodiments, aiming to further reduce judgment uncertainty by utilizing the new supplementary information.

[0059] Finally, when the uncertainty is reduced to the uncertainty threshold or the maximum number of iterations is reached, the iterative verification terminates and the final judgment result is output. This ensures that the iteration process can stop in a timely manner when the preset judgment accuracy or resource limit is reached, thereby providing efficient and reliable quality judgment.

[0060] Optionally, the step of assigning probe priority scores to each probe operation includes: Obtain production target information for the current production stage; Obtain batch attribute information of the component to be tested; Based on production target information and batch attribute information, adjust the weights of information gain index, execution cost index, damage risk index and subsequent interference index in the probe priority score calculation. Based on the adjusted weights, the information gain index, execution cost index, damage risk index, and subsequent interference index are weighted and summed to obtain the probe priority score for each probe operation.

[0061] Specifically, production target information refers to the main objectives set by the current chip production line. These may include pursuing high production throughput, ensuring extremely low defect escape rates, prioritizing the detection of specific types of critical defects, or maximizing component yield. These objectives directly influence the emphasis on the efficiency and accuracy of the inspection process. Batch attribute information can be understood as the inherent characteristics or relevant background information of the batch of components to be inspected. This may include the component's material sensitivity, its criticality level within the entire product, known variability that may exist during manufacturing, historical defect rate data, or the component's cost. These attributes provide important context for assessing the potential impact and benefits of probe operations. Adjusting weights refers to dynamically modifying the relative importance of information gain indicators, execution cost indicators, damage risk indicators, and subsequent interference indicators in calculating the probe priority score based on the acquired production target information and batch attribute information. For example, when the production target is high throughput, the weight of execution cost indicators (especially time consumption) may be increased; when the batch attributes of the component to be inspected indicate that it is a high-value or fragile component, the weight of damage risk indicators will be significantly increased. In this way, the calculation of probe priority scores can be more flexible and intelligent.

[0062] Optionally, the adjustment steps for the uncertainty threshold range and the maximum number of iterations include: Obtain batch attribute information of the component to be tested; Obtain defect type identification information for the current region of the component to be inspected; Adjust the uncertainty threshold range based on batch attribute information and defect type identification information; Adjust the maximum number of iterations based on defect type identification information and cost-effectiveness information of performed probe operations.

[0063] Specifically, in the iterative verification process, the first step is to obtain the batch attribute information of the component under test. Batch attribute information can be understood as metadata related to the batch to which the component belongs, such as production date, production line, material supplier, batch quality level, and historical defect rate. This information reflects the overall quality trend and potential risks of the batch. Simultaneously, it is also necessary to obtain defect type identification information for the current area of ​​the component under test. Defect type identification information refers to the preliminary identification or classification of possible defect types in the current area by the system during initial judgment or previous iterations, such as short circuits, open circuits, cold solder joints, misalignment, and foreign objects.

[0064] Based on the acquired batch attribute information and defect type identification information, the uncertainty threshold range can be adjusted. For example, for batches with a known low historical defect rate, or for areas identified as non-critical defects, the uncertainty threshold range can be appropriately widened, allowing iteration to terminate at a relatively high level of uncertainty to improve detection efficiency. Conversely, for high-risk batches or critical defect areas, the uncertainty threshold range should be tightened, requiring a lower level of judgment uncertainty to terminate iteration, to ensure the accuracy of the judgment.

[0065] In practical applications, the maximum number of iterations can be adjusted based on defect type identification information and cost-effectiveness information of the probe operations already performed. For example, for defect types identified as easy to judge, or when the cost of the probe operations performed is high but the information gain is limited, the maximum number of iterations can be appropriately reduced to avoid unnecessary resource consumption. Conversely, for complex or difficult-to-judge defect types, or when the cost of the probe operations is low and the expected information gain is significant, the maximum number of iterations can be increased to provide more opportunities to reduce judgment uncertainty.

[0066] Optionally, the steps to terminate iterative verification include: Obtain the adjusted uncertainty threshold range and the adjusted maximum number of iterations; After each probe operation and multi-source fusion processing is performed, the current uncertainty is calculated and the corresponding state probability information is updated. Determine whether the uncertainty falls within the adjusted uncertainty threshold range, and determine whether the number of probe operations performed has reached the adjusted maximum number of iterations; When the uncertainty falls within the adjusted uncertainty threshold range, the final judgment result is output based on the currently updated state probability information. When the number of probe operations performed reaches the adjusted maximum number of iterations and the uncertainty is still not within the adjusted uncertainty threshold range, a termination flag indicating that the uncertainty cannot be eliminated will be output, along with a prompt for manual review or rework and retesting.

[0067] Specifically, in the iterative verification process, the uncertainty threshold range and the adjusted maximum number of iterations, adjusted according to the above scheme, are first obtained. These parameters are dynamically determined based on the batch attribute information and defect type identification information of the component to be detected, aiming to provide the most suitable termination condition for the current detection task. Subsequently, after each probe operation and multi-source fusion processing, the system calculates the judgment uncertainty of the current region in real time and updates the corresponding state probability information based on the fused information. This process ensures continuous monitoring and evaluation of the detection status. Next, the system performs a dual judgment: first, it determines whether the currently calculated judgment uncertainty has fallen within the adjusted uncertainty threshold range, i.e., whether the judgment uncertainty has been reduced to an acceptable level; second, it determines whether the number of probe operations performed has reached the adjusted maximum number of iterations to prevent infinite loops. When the judgment uncertainty falls within the adjusted uncertainty threshold range, it indicates that the uncertainty has been successfully reduced to an acceptable level through iterative verification. At this point, the system outputs the final defect judgment result based on the currently updated state probability information. However, if the number of probe operations performed reaches the adjusted maximum number of iterations, but the uncertainty still falls outside the adjusted uncertainty threshold, it means that under the current resources and strategies, the system cannot further eliminate uncertainty through automation. In this case, the system will output a termination flag indicating that uncertainty cannot be eliminated, and simultaneously output a prompt for manual review or rework / retesting to guide subsequent manual intervention or further detection processing.

[0068] This application also discloses a SMT (Surface Mount Technology) image recognition system for chip manufacturing, used to perform SMT image recognition for chip manufacturing, combined with... Figure 3 As shown, the SMT (Surface Mount Technology) surface mount quality image recognition system 1 for chip manufacturing includes: The component information acquisition module 11 is used to acquire the surface optical characterization information of the component to be tested by applying various illumination conditions, and to determine the component type information of the component to be tested based on the surface optical characterization information. The image acquisition execution module 12 is used to acquire the detection image information of the component to be detected after adjusting the image acquisition parameters according to the component type information; The defect feature extraction module 13 is used to process the detected image information and extract defect image feature information related to the defect; The discrimination criterion determination module 14 is used to determine the defect discrimination criterion information based on the component type information; The preliminary judgment execution module 15 is used to make a preliminary judgment on the defects of the component to be detected based on the defect image feature information and defect discrimination standard information, and generate preliminary judgment result information. The quality judgment execution module 16 is used to obtain supplementary detection information and perform multi-source fusion processing and state probability reassessment on the supplementary detection information when the preliminary judgment result information indicates that there is uncertainty in judgment, so as to reduce the uncertainty in judgment, and make a final judgment on the defect based on the result of reducing the uncertainty in judgment, thereby obtaining defect judgment information and quality judgment information.

[0069] To facilitate understanding of the technical solution of this application, the key terms, system structure and operating environment involved are further explained below.

[0070] The component under inspection (SDI) refers to the electronic component that requires quality inspection during the SMT assembly process. This can include electronic devices in various package forms, such as chips, resistors, and capacitors. Surface optical characterization information refers to characteristic data reflecting the surface physical properties of the component under inspection, obtained through optical imaging or measurement methods. This includes reflectivity, transmittance, scattering characteristics, and their spatial distribution. This information reflects attributes such as the component's material composition, surface roughness, and coating uniformity. Component type information identifies the component's model, package structure, and material characteristics, serving as the basis for subsequent image acquisition parameter settings and defect identification strategy selection. Image acquisition parameters refer to camera control parameters that affect the quality of the inspection image, including exposure time, gain, white balance, focal length, and aperture settings. Inspection image information refers to the component image data acquired by the image acquisition device for quality inspection after the image acquisition parameters are determined. Defect image feature information is visual feature information extracted from the inspection image information that is related to potential defect states such as cold solder joints, misalignment, short circuits, and contamination. This includes edge features, texture features, color distribution features, and brightness variation features. Defect discrimination criteria information refers to the rules or models used to determine whether a component under inspection has a defect. These criteria can be differentiated based on component type and defect type. Preliminary judgment results are preliminary detection conclusions derived from defect image feature information and defect discrimination criteria information. Judgment uncertainty refers to situations where the preliminary judgment results are insufficient to clearly distinguish whether a component is in a defective or normal state. Supplementary detection information refers to auxiliary information obtained through additional detection methods when judgment uncertainty exists. This can include images from different angles, different spectral imaging data, or other auxiliary detection data. Multi-source fusion processing refers to the joint analysis of information from different detection sources to form a more comprehensive description of the component's state. State probability reassessment refers to updating the probability distribution of the component under inspection in different states based on multi-source fusion processing to reduce judgment uncertainty.

[0071] The implementation environment of this application is typically an automated optical inspection system, which integrates various lighting devices, high-resolution image acquisition equipment, image processing units, and decision control units, thereby constructing an online or offline quality inspection platform suitable for SMT assembly lines.

[0072] This application proposes an image recognition system for SMT (Surface Mount Technology) chip manufacturing to address the challenges posed by diverse component types, complex defect morphologies, and varying imaging environments under complex production conditions. In the system, a component information acquisition module applies various illumination conditions to the component under inspection at the initial stage to acquire its surface optical characterization information and determines the corresponding component type based on this information. This module can integrate multiple illumination methods such as ring light, coaxial light, backlight, or polarized light, and combine them with an image sensor to acquire imaging data of the component under different illumination conditions. By analyzing reflection characteristics, scattering distribution, or spectral response, the component information acquisition module can identify the component packaging material and its surface structural characteristics, thus providing a type basis for subsequent inspection processes.

[0073] The image acquisition module, after obtaining component type information, adjusts the image acquisition parameters based on this information and then acquires the detection image information of the component to be detected. This module can set appropriate exposure time, gain, and imaging focal length for different component types by calling a preset parameter library or adaptively calculating the parameters, ensuring that the detected image is in a suitable state in terms of brightness, contrast, and sharpness. In this way, the system can avoid problems such as overexposure, underexposure, or loss of detail caused by fixed parameter settings.

[0074] The defect feature extraction module processes the detected image information to extract defect image features related to potential defects. This module performs image denoising, enhancement, region segmentation, and feature calculation, extracting multi-dimensional visual features to characterize the defect state. The discrimination criterion determination module determines the defect discrimination criterion information matching the currently detected component based on the component type information and provides this criterion to subsequent judgment modules. Through the correspondence between component type and discrimination criteria, the system can avoid misjudgments caused by using the same discrimination threshold for components with different packaging forms or material properties.

[0075] The preliminary judgment module performs a preliminary defect judgment on the component under inspection based on defect image feature information and defect discrimination criteria information, and generates preliminary judgment result information. The preliminary judgment result information may include defect type, defect location, and corresponding confidence level. When the preliminary judgment result information indicates uncertainty, the quality judgment module is triggered to perform further processing. This quality judgment module is used to acquire supplementary detection information and perform multi-source fusion processing on the supplementary detection information, existing surface optical characterization information, and detection image information, thereby re-evaluating the component's state probability information. By introducing multi-dimensional information and performing state probability re-evaluation, the system can conduct a more in-depth analysis of the component's true state when a clear conclusion cannot be reached in the preliminary judgment stage, thereby reducing the risk of misjudgment.

[0076] Through the coordinated operation of the aforementioned modules, this application constructs an SMT (Surface Mount Technology) quality image recognition system with adaptive parameter adjustment capabilities, support for multi-stage judgment, and uncertainty resolution mechanisms. This system effectively overcomes the problems of fixed parameters, rigid judgment standards, and insufficient ability to identify minute defects in traditional detection schemes. It maintains high detection accuracy and reliability even with new packaging materials and complex production environments, thereby significantly improving the quality control level in the SMT assembly process.

[0077] The above are merely embodiments of this application and are not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for SMT (Surface Mount Technology) surface mount quality image recognition in chip manufacturing, characterized in that, include: By applying various illumination conditions to the component under test, the surface optical characterization information of the component under test is obtained, and the component type information of the component under test is determined based on the surface optical characterization information; Based on the component type information, the image acquisition parameters are adjusted, and then the detection image information of the component to be detected is acquired. The detected image information is processed to extract defect image feature information related to the defect; Based on the component type information, determine the defect identification criteria information; Based on the defect image feature information and the defect discrimination standard information, a preliminary judgment is made on the defects of the component to be detected and a preliminary judgment result information is generated; When the preliminary judgment result indicates that there is uncertainty in the judgment, supplementary detection information is obtained and multi-source fusion processing and state probability reassessment are performed on the supplementary detection information to reduce the uncertainty in the judgment. Based on the result of reducing the uncertainty in the judgment, the defect is finally judged to obtain defect judgment information and quality judgment information.

2. The SMT (Surface Mount Technology) image recognition method for chip manufacturing according to claim 1, characterized in that, When the preliminary judgment result indicates the existence of judgment uncertainty, the steps of obtaining supplementary detection information and performing multi-source fusion processing and state probability reassessment on the supplementary detection information include: Analyze and determine the source of uncertainty, and identify the category of uncertainty source information as at least one of internal material fluctuations, environmental disturbance artifacts, or overlap of minor defects with normal characteristics; Based on the uncertainty source category information, a target probe operation for reducing the uncertainty of judgment is selected from the preset probe operation library; the preset probe operation library is a pre-established probe operation collection library, which at least stores the type of each probe operation, the applicable uncertainty source category for triggering, and the corresponding execution parameters; Perform the target probe operation to obtain supplementary detection information; The supplementary detection information, surface optical characterization information, and detection image information are fused together to obtain fused information. Based on the fused information, the state probability information of the current region of the element to be detected is re-evaluated; The defects are finally judged based on the reassessed state probability information. During the final judgment, if the state probability information of one defect state is higher than that of a non-defect state, it is judged as a real defect. If the state probability information of a non-defect state is higher than that of a defect state, it is judged as normal. If the state probability information of multiple states is the same, the iterative verification process is entered to further reduce the uncertainty of the judgment.

3. The SMT (Surface Mount Technology) image recognition method for chip manufacturing according to claim 2, characterized in that, The iterative verification process includes: Acquire surface temperature information, local charge distribution information, and microstructure information of the current region of the element under test; Based on the surface temperature information, the local charge distribution information, and the microstructure information, the instantaneous or cumulative impact of the preceding probe operation on the surface or internal structure of the element under test is evaluated, and the impact evaluation result information is obtained. Based on the type and intensity information of the preceding probe operation, predict the optical artifact information introduced by the preceding probe operation; Based on the impact assessment results and the optical artifact information, the parameters of subsequent probe operations are adjusted, and the adjusted subsequent probe operations are executed to obtain supplementary detection information.

4. The SMT (Surface Mount Technology) image recognition method for chip manufacturing according to claim 2, characterized in that, The iterative verification process includes: The material type information, the sequence of probe operations performed and the corresponding parameter information, and the real-time environmental perception data of the component to be detected are obtained to construct the uncertainty state description information of the current area of ​​the component to be detected. For the uncertainty state description information, the information gain index expected to be brought about by each probe operation under the current uncertainty state is calculated from the preset probe operation library; the calculation of the information gain index takes into account the instantaneous or cumulative effects caused by the preceding probe operation and the introduced optical artifact information. Based on the information gain index, the probe operation with the largest information gain and the total execution cost index meeting the preset threshold is selected; when multiple probe operations have the same information gain index, the probe operation with the lowest damage risk index is selected first. Perform the selected probe operation to obtain supplementary detection information; The supplementary detection information is fused with historical detection information to obtain the fused information. Based on the fused information, the state probability information of various states is re-evaluated; When the uncertainty is reduced to the uncertainty threshold or the maximum number of iterations is reached, the iterative verification is terminated and the final judgment result is output.

5. The SMT (Surface Mount Technology) image recognition method for chip manufacturing according to claim 4, characterized in that, The steps that take into account the instantaneous or cumulative effects of preceding probe operations and the introduced optical artifact information include: Acquire information on the surface temperature distribution, local charge distribution, and instantaneous changes in the microstructure of the current region of the element under test; The surface temperature distribution information, the local charge distribution information, and the instantaneous change data of the microstructure are correlated with the type, intensity, and duration information of the probe operation performed to obtain the correlation analysis results. Based on the correlation analysis results, assess the instantaneous or cumulative impact caused by the preceding probe operation to obtain impact assessment results information; Based on the type and intensity information of the preceding probe operation, the corresponding optical artifact information is predicted. Based on the impact assessment results and the optical artifact information, adjust the parameters of subsequent probe operations and execute the adjusted subsequent probe operations; when the impact assessment results indicate that the potential damage risk exceeds a preset threshold, terminate the iterative verification and output an alarm message.

6. The SMT (Surface Mount Technology) image recognition method for chip manufacturing according to claim 4, characterized in that, The iterative verification process also includes the following steps: Calculate the execution cost index for each probe operation; the execution cost index includes time consumption, energy consumption, and the degree of potential micro-damage to the component under test; Evaluate the subsequent interference indicators of each probe operation on subsequent operations; The information gain index, the execution cost index, the damage risk index, and the subsequent interference index are comprehensively considered to assign probe priority scores to each probe operation. The probe operation with the highest probe priority score is selected for execution; where, when multiple probe operations have the same probe priority score, the probe operation with the lowest damage risk index is selected first. Perform the selected probe operation to obtain supplementary detection information, and perform multi-source fusion processing on the supplementary detection information and historical detection information to update the state probability information; When the uncertainty is reduced to the uncertainty threshold or the maximum number of iterations is reached, the iterative verification is terminated and the final judgment result is output.

7. The SMT (Surface Mount Technology) image recognition method for chip manufacturing according to claim 6, characterized in that, The step of assigning probe priority scores to each probe operation includes: Obtain production target information for the current production stage; Obtain the batch attribute information of the component to be detected; Based on the production target information and the batch attribute information, adjust the weights of the information gain index, the execution cost index, the damage risk index, and the subsequent interference index in the probe priority score calculation; Based on the adjusted weights, the information gain index, the execution cost index, the damage risk index, and the subsequent interference index are weighted and summed to obtain the probe priority score for each probe operation.

8. The SMT (Surface Mount Technology) image recognition method for chip manufacturing according to claim 6, characterized in that, The steps for adjusting the uncertainty threshold range and the maximum number of iterations include: Obtain batch attribute information of the component to be tested; Obtain defect type identification information for the current region of the component to be inspected; The uncertainty threshold range is adjusted based on the batch attribute information and the defect type identification information; The maximum number of iterations is adjusted based on the defect type identification information and the cost-effectiveness information of the probe operations already performed.

9. The SMT (Surface Mount Technology) image recognition method for chip manufacturing according to claim 8, characterized in that, The steps for terminating iterative verification include: Obtain the adjusted uncertainty threshold range and the adjusted maximum number of iterations; After each probe operation and multi-source fusion processing is performed, the current uncertainty is calculated and the corresponding state probability information is updated. Determine whether the uncertainty falls within the adjusted uncertainty threshold range, and determine whether the number of probe operations performed has reached the adjusted maximum number of iterations; When the uncertainty of the judgment falls within the adjusted uncertainty threshold range, the final judgment result is output based on the currently updated state probability information; When the number of probe operations performed reaches the adjusted maximum number of iterations and the uncertainty judgment still does not fall within the adjusted uncertainty threshold range, a termination flag indicating that the uncertainty cannot be eliminated is output, along with a prompt message for manual review or rework and retesting.

10. An SMT (Surface Mount Technology) image recognition system for chip manufacturing, used to perform SMT image recognition for chip manufacturing, characterized in that, include: The component information acquisition module is used to acquire the surface optical characterization information of the component under test by applying various illumination conditions to the component under test, and to determine the component type information of the component under test based on the surface optical characterization information; The image acquisition execution module is used to acquire the detection image information of the component to be detected after adjusting the image acquisition parameters according to the component type information; The defect feature extraction module is used to process the detected image information and extract defect image feature information related to the defect; The discrimination criterion determination module is used to determine defect discrimination criterion information based on the component type information; The preliminary judgment execution module is used to make a preliminary judgment on the defects of the component to be detected based on the defect image feature information and the defect discrimination standard information, and generate preliminary judgment result information. The quality judgment execution module is used to obtain supplementary detection information and perform multi-source fusion processing and state probability reassessment on the supplementary detection information when the preliminary judgment result information indicates that there is judgment uncertainty, so as to reduce the judgment uncertainty, and make a final judgment on the defect based on the result of reducing the judgment uncertainty, thereby obtaining defect judgment information and quality judgment information.

Images