A chip package pi defect intelligent detection method and system based on multi-modal images

Abstract

This invention provides an intelligent detection method and system for PI defects in chip packaging based on multimodal images, mainly relating to the field of chip packaging defect detection technology. The method comprises the following steps: First, at least two modal images of the chip packaging site are acquired, including a color image mainly characterizing the surface morphology of the chip and a fluorescence image mainly characterizing the physical properties of the chip surface material; then, the color image and fluorescence image to be detected are input into a multimodal target detection model to perform defect detection and obtain preliminary detection results; finally, based on the preliminary detection results and combined with the grayscale feature analysis of the fluorescence image, intelligent posterior decision-making is performed to obtain the intelligent detection result of PI defects in chip packaging. This invention combines the joint detection of color and fluorescence images with intelligent posterior decision-making, solving the problems of inaccurate detection of abnormal PI thickness defects, high dependence on manual image interpretation, and difficulty in automatically distinguishing multiple types of PI defects in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of chip packaging defect detection technology, and more specifically, to a method and system for intelligent detection of chip packaging PI defects based on multimodal images. Background Technology

[0002] In integrated circuit packaging, polyimide (PI) serves as an interlayer dielectric or protective layer, and its coating quality directly affects the reliability and yield of the chip. PI defects mainly include PI residue, PI residue caused by foreign matter, and abnormal PI thickness. These defects face numerous challenges in traditional inspection methods. Traditional automated optical inspection (AOI) systems are mostly based on a single imaging mode (usually color images), while PI defects exhibit different characteristics under different imaging principles: PI residual defects are invisible in color images but show obvious luminescence in fluorescence images; PI residues caused by foreign matter appear as foreign matter morphology in color images and emit light simultaneously in fluorescence images; pure foreign matter is visible in color images but does not emit light in fluorescence images; in addition, abnormal PI thickness appears normal in color images but appears as a large area of ​​reduced background grayscale in fluorescence images.

[0003] Existing detection technologies have the following main shortcomings: ①Limitations of single-modal perception: A single imaging mode cannot fully capture the multi-dimensional features of defects. In particular, for defects such as PI thickness anomalies, which manifest as global grayscale changes, detection methods based on color images are basically ineffective.

[0004] ② Insufficient detection accuracy: Even with deep learning technology, a single model still suffers from high false negative and false positive rates in detecting PI thickness anomalies. Experiments show that the traditional YOLOv5 model has a recall rate of only about 60% and a precision rate of about 70% for PI thickness anomalies.

[0005] ③Low efficiency of human-machine collaboration: Current detection systems require a large amount of manual review, especially for detection results with confidence levels in the middle range, where the rate of manual intervention is usually as high as 30% or more. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a method and system for intelligent detection of PI defects in chip packaging based on multimodal images.

[0007] In a first aspect, embodiments of the present invention provide an intelligent detection method for PI defects in chip packaging based on multimodal images, comprising the following steps: Acquire at least two modal images of the chip packaging site, including color images that primarily characterize the morphological features of the chip surface and fluorescence images that primarily characterize the physical properties of the chip surface material; The color image and fluorescence image to be detected are input into the multimodal target detection model to perform defect detection and obtain preliminary detection results; Based on the preliminary detection results, combined with the grayscale feature analysis of the fluorescence image, intelligent posterior decision-making is performed to obtain intelligent detection results of PI defects in chip packaging.

[0008] In the above embodiments, the present invention inputs color images and fluorescence images into a network for detection and combines intelligent posterior decision-making. While significantly simplifying the system architecture, it achieves high-precision and high-efficiency detection of complex defects such as abnormal PI thickness in chip packaging. This solves the technical problems of single-modal perception limitations, insufficient detection accuracy, and low human-machine collaboration efficiency in traditional methods.

[0009] As some optional embodiments of this application, before inputting the color image and fluorescence image to be detected into the multimodal target detection model to perform defect detection, the multimodal target detection model needs to be trained; the training process of the multimodal target detection model is as follows: Acquire color images and corresponding fluorescence images of the chip packaging process stations; Foreign object morphological defects in the color image are marked, and residual luminescent areas and abnormal grayscale areas of PI thickness in the fluorescence image are marked. Multimodal joint training of the multimodal target detection model is performed based on labeled color images and fluorescence images.

[0010] In the above embodiments, the present invention simultaneously labels the foreign object category in the color image and the defect category in the fluorescence image, and uses multimodal images to jointly train the model, enabling the model to automatically learn and fully utilize the complementary features of the two imaging modes, thereby significantly improving the model's detection accuracy and generalization ability for complex PI defects.

[0011] As some optional embodiments of this application, when annotating the abnormal grayscale region of PI thickness in the fluorescence image, the annotation box needs to cover the transition region of grayscale difference.

[0012] In the above embodiments, the annotation method of the present invention enables the model to learn the complete gray-scale gradient features of PI thickness anomalies from normal to abnormal regions, thereby significantly improving the model's ability to detect large-area, low-contrast defects and the accuracy of boundary recognition.

[0013] As some optional implementations of this application, the preliminary detection results include defect category, confidence level, and corresponding bounding box location information.

[0014] In the above embodiments, the preliminary detection results of the present invention provide complete defect information, which facilitates subsequent processing. The confidence level and location information contained therein provide a direct and reliable quantitative basis for intelligent posterior decision-making and accurate defect analysis.

[0015] As some optional implementations of this application, the intelligent posterior decision is as follows: If the preliminary detection results of the fluorescence image contain PI residue defects with a confidence level greater than the first threshold, then the preliminary detection results of the color image are combined for determination: if the color image detects foreign object defects at the same location, it is determined to be PI residue caused by foreign object; otherwise, it is determined to be PI residue defect. If the preliminary detection results of the fluorescence image do not contain PI residual defects but contain PI thickness abnormality defects with a confidence level greater than the second threshold, then extract the area in the fluorescence image whose gray value is within the first preset range and calculate its area ratio: if the area ratio is less than the third threshold, then output the manual judgment instruction; otherwise, judge it as a PI thickness abnormality defect. If the preliminary detection results of the fluorescence image and the color image do not detect any defects, the product is determined to be normal, and a missed detection suppression check is performed.

[0016] In the above embodiments, the present invention combines the preliminary detection results of the deep learning model with traditional image processing (grayscale and area analysis) through intelligent posterior decision logic. By analyzing the area ratio of specific grayscale regions in the fluorescence image, it intelligently distinguishes between true positive defects and over-detection false positives caused by the model. At the same time, it can detect large-area grayscale anomalies that the model misses, fundamentally suppressing the under-detection and over-detection of PI thickness anomalies.

[0017] As some optional embodiments of this application, the missed detection suppression check is as follows: extract the region in the fluorescence image whose gray value is within a second preset range, and if its area ratio is greater than a fourth threshold, output a manual judgment instruction.

[0018] In the above embodiments, the present invention significantly reduces labor costs while suppressing missed detections, and makes the image judgment process standardized and traceable.

[0019] As some optional implementations of this application, the intelligent posterior decision also includes priority determination, and the priority order from high to low is: PI residual defects caused by foreign matter, PI residual defects, PI thickness abnormality defects, and other foreign matter defects.

[0020] In the above embodiments, the present invention employs this priority determination mechanism to ensure that when multiple defects coexist, the most critical defect category can be output based on the severity of the defect and the priority of its impact on the process. This effectively avoids the risk of misjudging minor defects by masking major defects, thereby providing an accurate and reliable quality basis for subsequent process decisions.

[0021] In a second aspect, the present invention provides an intelligent detection system for PI defects in chip packaging based on multimodal images, the system comprising: An image acquisition unit is used to acquire at least two modal images of the chip packaging site, including a color image that mainly characterizes the morphological features of the chip surface and a fluorescence image that mainly characterizes the physical properties of the chip surface material. A preliminary detection unit is used to input the color image and fluorescence image to be detected into a multimodal target detection model to perform defect detection and obtain preliminary detection results; The post-decision unit, based on the preliminary detection results and combined with the grayscale feature analysis of the fluorescence image, performs intelligent posterior decision-making to obtain intelligent detection results of PI defects in chip packaging.

[0022] In a third aspect, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the aforementioned intelligent detection method for chip packaging PI defects based on multimodal images.

[0023] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned intelligent detection method for chip packaging PI defects based on multimodal images.

[0024] The beneficial effects of this invention are as follows: 1. The detection accuracy of this invention is significantly improved: By training color images and fluorescence images in a multimodal manner, the model can learn the complementary features of different defects in the two imaging modes at the same time, which significantly improves the ability to perceive complex PI defects, especially PI thickness abnormal defects.

[0025] 2. This invention effectively solves the problems of missed detections and overdetections: By using intelligent posterior decision logic, the preliminary detection results of the deep learning model are combined with traditional image processing (grayscale and area analysis). By analyzing the area ratio of specific grayscale regions in the fluorescence image, it intelligently distinguishes between true positive defects and false positives caused by the model. At the same time, it can detect large-area grayscale anomalies that the model misses, fundamentally suppressing missed detections and overdetections of PI thickness anomalies.

[0026] 3. This invention achieves refined classification of defects: By logically combining the detection results of color images and fluorescence images (such as position overlap judgment), it can accurately distinguish defects with similar forms but different properties, such as PI residual defects, PI residual defects caused by foreign objects, and foreign object defects. The classification ability is far superior to a single model.

[0027] 4. This invention improves the level of automation and intelligence: By setting confidence thresholds, priority rules and clear conditions for transferring to manual processing, it maximizes the proportion of automated image judgment while ensuring high accuracy, significantly reduces labor costs, and makes the image judgment process standardized and traceable. Attached Figure Description

[0028] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments are briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a flowchart of the intelligent detection method for PI defects in chip packaging according to an embodiment of the present invention; Figure 2 This is a flowchart of the intelligent posterior decision-making process described in an embodiment of the present invention. Detailed Implementation

[0030] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0031] To address the limitations of existing detection technologies, such as single-modal perception, insufficient detection accuracy, and low efficiency of human-machine collaboration, this invention provides an intelligent detection method and system for PI defects in chip packaging based on multimodal images. Please refer to [link to relevant documentation]. Figure 1 , Figure 1 The flowchart of the intelligent detection method for PI defects in chip packaging is as follows: (1) Multimodal data acquisition and preprocessing.

[0032] This invention simultaneously acquires color and fluorescence images at the same detection location at key process stations in chip packaging; wherein the color image mainly characterizes the surface morphology of the chip, and the fluorescence image mainly characterizes the distribution and thickness information of the PI material on the chip surface.

[0033] Specifically, the acquired color and fluorescence images are preprocessed, including but not limited to: ① Distortion correction: lens distortion is corrected according to camera calibration parameters; ② Spatial registration: a SIFT feature-based registration algorithm is used to ensure accurate alignment of the color and fluorescence images, with a registration error of less than 2 pixels; ③ Illumination normalization: the effects of uneven illumination are eliminated; ④ Noise filtering: adaptive median filtering is used to remove noise.

[0034] (2) Multimodal target detection.

[0035] This invention inputs the color image and fluorescence image to be detected into a multimodal target detection model to perform defect detection and obtain preliminary detection results.

[0036] Before performing multimodal target detection, this invention requires image registration of the color image and the fluorescence image. This is because there may be slight differences in the installation positions of the color camera and the fluorescence camera, so subpixel-level registration of the two images is required. Specifically, a feature point-based registration algorithm can be used for feature detection and matching, then the homography transformation matrix is ​​calculated, and the fluorescence image is geometrically corrected based on the homography transformation matrix to make it accurately aligned with the color image.

[0037] Furthermore, an improved YOLOv5 model is used as a multimodal target detection model for defect detection. The multimodal target detection model includes an input layer, a backbone network, and an output layer.

[0038] The input layer is used to input 4-channel images (color images as 3 channels and fluorescence images as 1 channel).

[0039] The Backbone network employs the YOLOv5 model's CSPDarknet53 architecture for feature extraction, and adds a fluorescence feature enhancement submodule and a cross-modal attention submodule to the Backbone network. The fluorescence feature enhancement submodule specifically extracts and enhances important features in fluorescence images through a series of convolutional layers and activation functions. Considering that abnormal PI thickness in fluorescence images manifests as large-area grayscale changes, this submodule uses a larger receptive field convolutional kernel to capture regional grayscale features. The cross-modal attention submodule dynamically calculates the importance weights of color and fluorescence features based on an attention mechanism. This module first concatenates color and fluorescence features along the channel dimension, then obtains global information through global average pooling, and finally learns the attention weights for each channel through fully connected layers. The learned weights are used to reweight the original features, enabling the network to focus on features more important to the current detection task.

[0040] The output layer is used for preliminary detection results of color and fluorescence images; wherein, the preliminary detection results include defect category, confidence level and corresponding bounding box location information.

[0041] Furthermore, the multimodal target detection model needs to be pre-trained, and the training process is as follows: ① Collect color images and corresponding fluorescence images of the chip packaging process sites.

[0042] ② The foreign object morphology defects and other types of defects in the color image are labeled, and the PI residual luminescence region and PI thickness abnormal grayscale region in the fluorescence image are labeled. When labeling the PI thickness abnormal grayscale region in the fluorescence image, the label box needs to cover the transition region of grayscale difference, which serves as the basis for model training.

[0043] ③ The multimodal target detection model is jointly trained using labeled color images and fluorescence images, and the model parameters are adjusted and saved.

[0044] (3) Intelligent posterior decision-making.

[0045] Based on preliminary detection results and combined with grayscale feature analysis of the fluorescence image, this invention performs intelligent posterior decision-making to obtain intelligent detection results for PI defects in chip packaging.

[0046] For details, please refer to Figure 2 , Figure 2 The flowchart for the execution of intelligent posterior decision-making is as follows: ①If the preliminary detection results of the fluorescence image contain PI residual defects with a confidence level greater than the first threshold, then the preliminary detection results of the color image are combined for determination: if the color image detects a foreign object at the same position, it is determined to be a PI residual defect caused by a foreign object; otherwise, it is determined to be a PI residual defect. ② If the preliminary detection results of the fluorescence image do not contain PI residual defects but contain PI thickness abnormality defects with a confidence level greater than the second threshold, then extract the area in the fluorescence image whose gray value is within the first preset range and calculate its area ratio: if the area ratio is less than the third threshold, then output the manual judgment instruction; otherwise, judge it as a PI thickness abnormality defect. ③ If no defects are detected in the preliminary detection results of the fluorescence image and the color image, the product is determined to be normal and a missed detection suppression check is performed: extract the area in the fluorescence image whose gray value is within the second preset range, and if its area ratio is greater than the fourth threshold, output a manual judgment instruction.

[0047] The first, second, third, and fourth thresholds can be set according to actual conditions. Preferably, the first threshold is set to 0.35; the second threshold is set to 0.35; the third threshold is 1 / 30 of the area of ​​the image to be detected; and the fourth threshold is 1 / 20 of the area of ​​the image to be detected. Specifically, fluorescent images of abnormal PI thickness defects are collected, and the average gray value of the abnormal area in such images is approximately 90, with a maximum value of approximately 120. The maximum gray value of the normal structure of the product can be 20. Therefore, the first and second preset ranges can be set to 30~110.

[0048] Furthermore, by analyzing the preliminary results output by the model, this invention can achieve defect detection along three paths: Path A: If the fluorescence image detects PI residue with a confidence level higher than 0.35 (suppressing overdetection), then proceed to branch judgment. At this time, query the detection results of the color image of the same sample. If foreign objects are also detected at the same location (judged by calculating the overlap index such as the bounding box IoU), then it is finally determined to be a PI residue defect caused by foreign objects; if no foreign objects are detected at the same location in the color image, then it is determined to be a pure PI residue defect. Path B: If the fluorescence image does not detect PI residue defects, but detects PI thickness abnormality defects with a confidence level higher than 0.35, then the overdetection suppression logic is activated: read the fluorescence image, extract the pixel region with a gray value between 30 and 110 (this range is based on the gray value statistical analysis of a large number of PI thickness abnormality samples), calculate the proportion of the area of ​​this region to the area of ​​the entire image. If the proportion is less than 1 / 30, it is considered that it may be model overdetection or a small normal fluctuation, and the manual instruction is output; if the proportion is greater than or equal to 1 / 30, it is confirmed as a valid PI thickness abnormality defect. Path C: If the model initially determines that there are no defects in the color image and the fluorescence image (determined as OK), then the missed detection suppression logic is activated: Similarly, the region with gray values ​​between 30 and 110 in the fluorescence image is extracted and the area ratio is calculated. If this ratio is greater than 1 / 20, it means that there is a large range of gray value abnormalities in the image, and the model may have missed detections. Therefore, the output is sent to the manual reviewer for verification; otherwise, the product is confirmed as a normal product.

[0049] In addition, the intelligent post-hoc decision-making also sets defect priorities (e.g., PI residual defects caused by foreign objects > PI residual defects > PI thickness abnormality defects > foreign object defects). When multiple judgment possibilities exist, the result is output according to the highest priority to ensure that critical defects are not covered up.

[0050] Specifically, some parameters in the above embodiments (such as confidence threshold of 0.35, grayscale range of 30-110, area ratios of 1 / 30 and 1 / 20, etc.) can be optimized and adjusted according to the actual conditions of different production lines and processes, and these adjustments all fall within the protection scope of this invention. Similarly, the multimodal target detection model is not limited to the improved YOLOv5.

[0051] In summary, this invention combines color and fluorescence image detection with intelligent posterior decision-making, solving the problems of inaccurate detection of PI thickness anomalies, high dependence on manual image interpretation, and difficulty in automatically distinguishing multiple types of PI defects in existing technologies.

[0052] Furthermore, in one embodiment, based on the same inventive concept as the foregoing embodiments, this embodiment of the invention provides an intelligent detection system for chip packaging PI defects based on multimodal images. The system corresponds one-to-one with the method described above, and includes: An image acquisition unit is used to acquire at least two modal images of the chip packaging site, including a color image that mainly characterizes the morphological features of the chip surface and a fluorescence image that mainly characterizes the physical properties of the chip surface material. A preliminary detection unit is used to input the color image and fluorescence image to be detected into a multimodal target detection model to perform defect detection and obtain preliminary detection results; The post-decision unit, based on the preliminary detection results and combined with the grayscale feature analysis of the fluorescence image, performs intelligent posterior decision-making to obtain intelligent detection results of PI defects in chip packaging.

[0053] It should be noted that each unit in the chip packaging PI defect intelligent detection system in this embodiment corresponds one-to-one with each step in the chip packaging PI defect intelligent detection method in the aforementioned embodiment. Therefore, the specific implementation method and the technical effects achieved in this embodiment can be referred to the implementation method of the aforementioned chip packaging PI defect intelligent detection method, and will not be repeated here.

[0054] Furthermore, in one embodiment, this application also provides a computer device, the computer device including a processor, a memory, and a computer program stored in the memory, the computer program being executed by the processor to implement the methods in the foregoing embodiments.

[0055] In addition, in one embodiment, this application also provides a computer storage medium storing a computer program that is executed by a processor to implement the methods described in the foregoing embodiments.

[0056] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a device including one or any combination of the above-mentioned memories. The computer may be a variety of computing devices, including smart terminals and servers.

[0057] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0058] As an example, executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborative files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0059] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0060] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0061] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0062] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk) and includes several instructions to cause a multimedia terminal device (which may be a mobile phone, computer, television receiver, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0063] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for intelligent detection of PI defects in chip packaging based on multimodal images, characterized in that, Includes the following steps: Acquire at least two modal images of the chip packaging site, including color images that primarily characterize the morphological features of the chip surface and fluorescence images that primarily characterize the physical properties of the chip surface material; The color image and fluorescence image to be detected are input into the multimodal target detection model to perform defect detection and obtain preliminary detection results; Based on the preliminary detection results, combined with the grayscale feature analysis of the fluorescence image, intelligent posterior decision-making is performed to obtain intelligent detection results of PI defects in chip packaging.

2. The intelligent detection method for PI defects in chip packaging based on multimodal images according to claim 1, characterized in that, Before inputting the color image and fluorescence image to be detected into the multimodal target detection model to perform defect detection, the multimodal target detection model needs to be trained; the training process of the multimodal target detection model is as follows: Acquire color images and corresponding fluorescence images of the chip packaging process stations; Foreign object morphological defects in the color image are marked, and residual luminescent areas and abnormal grayscale areas of PI thickness in the fluorescence image are marked. Multimodal joint training of the multimodal target detection model is performed based on labeled color images and fluorescence images.

3. The intelligent detection method for chip packaging PI defects based on multimodal images according to claim 2, characterized in that, When annotating abnormal grayscale regions of PI thickness in the fluorescence image, the annotation box must cover the transition area of ​​grayscale difference.

4. The intelligent detection method for PI defects in chip packaging based on multimodal images according to claim 1, characterized in that, The preliminary detection results include defect category, confidence level, and corresponding bounding box location information.

5. The intelligent detection method for PI defects in chip packaging based on multimodal images according to claim 4, characterized in that, The intelligent posterior decision is: If the preliminary detection results of the fluorescence image contain PI residue defects with a confidence level greater than the first threshold, then the preliminary detection results of the color image are combined for determination: if the color image detects foreign object defects at the same location, it is determined to be PI residue caused by foreign object; otherwise, it is determined to be PI residue defect. If the preliminary detection results of the fluorescence image do not contain PI residual defects but contain PI thickness abnormality defects with a confidence level greater than the second threshold, then extract the area in the fluorescence image whose gray value is within the first preset range and calculate its area ratio: if the area ratio is less than the third threshold, then output the manual judgment instruction; otherwise, judge it as a PI thickness abnormality defect. If the preliminary detection results of the fluorescence image and the color image do not detect any defects, the product is determined to be normal, and a missed detection suppression check is performed.

6. The intelligent detection method for PI defects in chip packaging based on multimodal images according to claim 5, characterized in that, The missed detection suppression check is as follows: extract the region in the fluorescence image whose gray value is within the second preset range, and if its area ratio is greater than the fourth threshold, output a manual judgment instruction.

7. The intelligent detection method for PI defects in chip packaging based on multimodal images according to claim 5, characterized in that, The intelligent post-hoc decision also includes priority determination, with the priority order from high to low as follows: PI residual defects caused by foreign matter, PI residual defects, PI thickness abnormality defects, and other foreign matter defects.

8. A chip packaging PI defect intelligent detection system based on multimodal images for implementing the method of claim 1, characterized in that, The system includes: An image acquisition unit is used to acquire at least two modal images of the chip packaging site, including a color image that mainly characterizes the morphological features of the chip surface and a fluorescence image that mainly characterizes the physical properties of the chip surface material. A preliminary detection unit is used to input the color image and fluorescence image to be detected into a multimodal target detection model to perform defect detection and obtain preliminary detection results; The post-decision unit, based on the preliminary detection results and combined with the grayscale feature analysis of the fluorescence image, performs intelligent posterior decision-making to obtain intelligent detection results of PI defects in chip packaging.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the intelligent detection method for chip packaging PI defects based on multimodal images as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the intelligent detection method for chip packaging PI defects based on multimodal images as described in any one of claims 1-7.

Images