Polarizer appearance defect detection method and device

Abstract

The invention relates to the field of polarizer detection technology. In order to improve the comprehensiveness and efficiency of detection and determine the layer position of defects, a method and device for detecting appearance defects of a polarizer are provided. The method comprises the following steps: obtaining N images of a polarizer to be detected, wherein every two images have at least one different acquisition condition; inputting the N images into a defect detection model, wherein the defect detection model introduces an attention mechanism; in the defect detection model, extracting features from each image by a feature extraction unit to obtain feature information corresponding to each image; inputting the feature information of the N images into N target sequence generation units, and generating a sequence of target defects in the image and the layer position of the target defects according to the feature information of the corresponding image and multiple query vectors; inputting the sequence of target defects in the N images into N defect judgment units, and judging the type of the target defects in the image according to the sequence of target defects in the corresponding image.

Description

Technical Field

[0001] The present invention relates to the technical field of polarizer detection, and in particular to a method and device for detecting appearance defects of a polarizer. Background Art

[0002] Polarizer is a thin film material with specific optical properties, which is mainly used to control the propagation direction and intensity of light. In liquid crystal display devices, polarizers are placed on the front and back sides of the liquid crystal panel to convert natural light into a light source suitable for display and control the direction of light transmission, thereby realizing image display.

[0003] Under a certain process, the production process of polarizers mainly includes the following steps: through the cutting machine, the polarizer roll is processed into semi-finished sheets through the corresponding mold; through the single-board machine, the sample processing is quickly and directly formed; through the cleaning machine, the dust-free surface of the sheet is cleaned to ensure cleanliness; through the inkjet printer, the polarizer is sprayed with code to distinguish the front and back sides and the upper and lower plates; through the edge grinding machine, the edge of the sheet is polished and finished to ensure the size and edge sealing effect; through the inspection station, the polarizer is inspected for appearance defects.

[0004] Among them, the inspection station mainly detects scratches, stains, creases, cracks, residual glue, inclusions and other appearance defects of the polarizer. The current inspection methods mainly include manual inspection and machine automatic inspection. Manual inspection requires a lot of time and human resources, and the accuracy depends on the ability and energy of the personnel. Automatic inspection mainly uses machine vision inspection technology. However, due to the wide variety of appearance defects of polarizers, a single image collected in machine vision inspection can only detect a small number of types of appearance defects, and multiple images will take a lot of time in image processing and analysis, which prolongs the stagnation time of the polarizer at the inspection station and affects production efficiency. In addition, the polarizer is a multi-layer structure, and appearance defects may appear in any layer. The current machine vision inspection technology is difficult to determine the layer where the defect is located. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a method and device for detecting appearance defects of polarizers, which can not only improve the comprehensiveness and efficiency of polarizer appearance defect detection, but also detect the layer where the defects are located, providing guidance for pre-feeding, process correction, optimization and subsequent defect repair.

[0006] The technical solution adopted by the present invention is as follows:

[0007] A method for detecting appearance defects of a polarizer comprises the following steps: obtaining N images of a polarizer to be detected, wherein every two of the images have at least one different acquisition condition, and N is an integer greater than 1; inputting the N images into a defect detection model, wherein the defect detection model introduces an attention mechanism; in the defect detection model, extracting features from each of the images through a feature extraction unit to obtain feature information corresponding to each of the images; in the defect detection model, inputting the feature information of the N images into N target sequence generation units one by one, respectively, and in each of the target sequence generation units, generating a sequence of target defects in the image and a layer where the target defects are located according to the feature information of the corresponding image and a plurality of query vectors, wherein the plurality of query vectors correspond one by one to the plurality of layers in the polarizer to be detected; in the defect detection model, inputting the sequences of target defects in the N images into N defect judgment units one by one, respectively, and in each of the defect judgment units, judging the type of the target defect in the image according to the sequence of target defects in the corresponding image.

[0008] Furthermore, the acquisition conditions include: camera type, light source type, light source irradiation angle, and light source-camera position relationship.

[0009] Furthermore, the feature extraction unit includes an encoder, the target sequence generation unit includes a decoder, and the defect judgment unit includes a classifier.

[0010] Furthermore, the defect detection model is obtained by training a training set, wherein the training set includes polarizer defect sample images, and the polarizer defect sample images are annotated with defect types as labels.

[0011] Furthermore, the polarizer defect sample images in the training set are annotated with the defect layer as a label, and the defect detection model is newly trained under this annotation method. After training, multiple feature vectors corresponding to each defect layer are obtained, and the multiple feature vectors corresponding to each defect layer are averaged to obtain the query vector corresponding to the defect layer.

[0012] A polarizer appearance defect detection device comprises: an acquisition module, the acquisition module is used to acquire N images of a polarizer to be detected, wherein every two of the images have at least one different acquisition condition, and N is an integer greater than 1; an input module, the input module is used to input the N images into a defect detection model, wherein the defect detection model introduces an attention mechanism, in the defect detection model, a feature extraction unit performs feature extraction on each of the images to obtain feature information corresponding to each of the images; in the defect detection model, the feature information of the N images is input into N target sequence generation units one by one, respectively, in each of the target sequence generation units, a sequence of target defects in the image and a layer where the target defects are located are generated according to the feature information of the corresponding image and a plurality of query vectors, wherein the plurality of query vectors correspond one by one to the plurality of layers in the polarizer to be detected; in the defect detection model, the sequences of target defects in the N images are input into N defect judgment units one by one, respectively, in each of the defect judgment units, the type of the target defect in the image is judged according to the sequence of target defects in the corresponding image.

[0013] Furthermore, the acquisition conditions include: camera type, light source type, light source irradiation angle, and light source-camera position relationship.

[0014] Furthermore, the feature extraction unit includes an encoder, the target sequence generation unit includes a decoder, and the defect judgment unit includes a classifier.

[0015] Furthermore, the defect detection model is obtained by training a training set, wherein the training set includes polarizer defect sample images, and the polarizer defect sample images are annotated with defect types as labels.

[0016] Furthermore, the polarizer defect sample images in the training set are annotated with the defect layer as a label, and the defect detection model is newly trained under this annotation method. After training, multiple feature vectors corresponding to each defect layer are obtained, and the multiple feature vectors corresponding to each defect layer are averaged to obtain the query vector corresponding to the defect layer.

[0017] Beneficial effects of the present invention:

[0018] The present invention utilizes a defect detection model to simultaneously extract features from multiple images, and uses multiple target sequence generation units and multiple defect judgment units to respectively generate target defect sequences and judge target defect types for multiple images, and obtains the layer where the target defect is located by combining a query vector. Therefore, not only can the simultaneous detection of various appearance defects of the polarizer be achieved through the calculation of a computer, thereby improving the comprehensiveness and efficiency of the appearance defect detection of the polarizer, but also the layer where the defect is located can be detected, providing guidance for pre-feeding, process correction, optimization and subsequent defect repair. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flow chart of a method for detecting appearance defects of a polarizer according to an embodiment of the present invention;

[0020] Figure 2 A schematic diagram of a process flow of a defect detection model for processing an input image according to an embodiment of the present invention;

[0021] Figure 3 Schematic diagram of a block diagram of a device for detecting appearance defects of a polarizer according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] like Figure 1 As shown, the method for detecting appearance defects of a polarizer according to an embodiment of the present invention comprises the following steps:

[0024] S1, acquiring N images of the polarizer to be inspected, wherein every two images have at least one different acquisition condition. N is an integer greater than 1.

[0025] Among them, the acquisition conditions include: camera type, light source type, light source irradiation angle, light source camera position relationship, etc.

[0026] In one embodiment of the present invention, various types of cameras, such as line scan cameras, area array cameras, etc., can be set at the inspection station of the polarizer; various types of light sources, such as surface light sources, strip light sources, etc. can be set; various light source irradiation angles can be set, such as 90° vertical irradiation, 45° oblique irradiation, etc.; and various light source camera position relationships can be set, such as the camera and the light source are on the same side of the polarizer to be detected, that is, the polarizer to be detected is irradiated and the image is collected in a reflection manner, and the camera and the light source are respectively on both sides of the polarizer to be detected, that is, the polarizer to be detected is irradiated and the image is collected in a transmission manner. Under each collection condition, an image of the polarizer to be detected is collected respectively.

[0027] S2, inputting N images into the defect detection model, wherein the defect detection model introduces an attention mechanism.

[0028] In one embodiment of the present invention, the defect detection model may include a feature extraction unit, at least N target sequence generation units, and at least N defect judgment units. The feature extraction unit may include an encoder, the target sequence generation unit may include a decoder, and the defect judgment unit may include a classifier.

[0029] In one embodiment of the present invention, the encoder and decoder can be taken from the Transformer network. In addition to the encoder, the feature extraction unit can also include a backbone network. After the backbone network performs preliminary feature extraction, the extracted features are converted into vectors of fixed length by the encoder, that is, feature information is obtained. In one embodiment of the present invention, the backbone network can adopt a convolutional neural network, etc., and the classifier can adopt a convolutional neural network or a softmax layer, etc. That is to say, in terms of network structure, the defect detection model of a specific embodiment of the present invention can be composed of a convolutional neural network and an improved Transformer network, and the improved Transformer network includes an encoder and multiple decoders.

[0030] In one embodiment of the present invention, the defect detection model can be pre-trained through a training set, which contains a large number of polarizer defect sample images, and the polarizer defect sample images are labeled with defect types, thereby enabling the defect detection model to have the ability to distinguish defect types.

[0031] In actual detection applications, each image collected in step S1 is input into the trained defect detection model, and the defect detection model performs the following steps S3 to S5. Figure 2 shown.

[0032] S3, in the defect detection model, a feature extraction unit is used to extract features from each image to obtain feature information corresponding to each image.

[0033] S4, in the defect detection model, the feature information of N images is input into N target sequence generation units one by one, and in each target sequence generation unit, the sequence of target defects in the image and the layer where the target defects are located are generated according to the feature information of the corresponding image and multiple query vectors, wherein the multiple query vectors correspond one by one to the multiple layers in the polarizer to be detected.

[0034] In one embodiment of the present invention, the polarizer to be detected includes at least a protective film, an upper TAC layer, a PVA layer, a lower TAC layer, a PSA layer and a release film from top to bottom, and the layer where the defect is located in the polarizer to be detected includes the above-mentioned layers. Taking the above-mentioned 6 layers as an example, the query vector is 6 corresponding ones. In one embodiment of the present invention, the query vector can be obtained in another model training process. Specifically, the sample image of the polarizer defect in the training set can also be labeled with the layer where the defect is located, and the defect detection model is newly trained under this labeling method. After training, multiple feature vectors corresponding to the layer where each defect is located are obtained, and the multiple feature vectors corresponding to the layer where each defect is located are averaged to obtain the query vector corresponding to the layer where the defect is located. It should be understood that for each defect, a feature vector of the layer where it is located can be obtained. The average of all feature vectors corresponding to all defects in each layer can obtain the query vector corresponding to the layer.

[0035] By inputting the feature information of N images and multiple query vectors into the target sequence generation unit, the target sequence generation unit can obtain the sequence of target defects in the corresponding image according to the feature information, and at the same time, can obtain the layer where the target defect is located in combination with the query vector. It should be understood that if there is no defect in an image, the sequence of its target defects and the layer where its target defects are located will not be obtained here; if there are multiple defects in an image, the sequence of its multiple target defects can be obtained here, and the layer where each target defect is located can be obtained.

[0036] S5, in the defect detection model, the sequences of target defects in the N images are input into N defect judgment units one by one, and in each defect judgment unit, the type of the target defect in the image is judged according to the sequence of target defects in the corresponding image.

[0037] If at least one sequence of target defects of a certain image is obtained, the at least one sequence is input into a corresponding defect judgment unit, and finally the type of each target defect in the image is output.

[0038] According to the method for detecting appearance defects of polarizers according to an embodiment of the present invention, a defect detection model is used to simultaneously extract features from multiple images, and multiple target sequence generation units and multiple defect judgment units are used to respectively generate target defect sequences and judge target defect types for multiple images, and the layer where the target defect is located is obtained by combining a query vector. As a result, not only can the simultaneous detection of various appearance defects of the polarizer be achieved through the calculation of a computer, thereby improving the comprehensiveness and efficiency of the appearance defect detection of the polarizer, but also the layer where the defect is located can be detected, providing guidance for pre-feeding, process correction, optimization and subsequent defect repair.

[0039] Corresponding to the polarizer appearance defect detection method of the above embodiment, the present invention further provides a polarizer appearance defect detection device.

[0040] like Figure 3 As shown, the appearance defect detection device of the polarizer of the embodiment of the present invention includes an acquisition module and an input module. The acquisition module is used to acquire N images of the polarizer to be detected, wherein every two images have at least one different acquisition condition, and N is an integer greater than 1; the input module is used to input the N images into the defect detection model, wherein the defect detection model introduces an attention mechanism, and in the defect detection model, a feature extraction unit extracts features from each image to obtain feature information corresponding to each image; in the defect detection model, the feature information of the N images is input into the N target sequence generation units one by one, and in each target sequence generation unit, a sequence of target defects in the image and the layer where the target defects are located are generated according to the feature information of the corresponding image and multiple query vectors, wherein the multiple query vectors correspond one by one to multiple layers in the polarizer to be detected; in the defect detection model, the sequences of target defects in the N images are input into the N defect judgment units one by one, and in each defect judgment unit, the type of the target defect in the image is judged according to the sequence of target defects in the corresponding image.

[0041] Among them, the acquisition conditions include: camera type, light source type, light source irradiation angle, light source camera position relationship, etc.

[0042] In one embodiment of the present invention, various types of cameras, such as line scan cameras, area array cameras, etc., can be set at the inspection station of the polarizer; various types of light sources, such as surface light sources, strip light sources, etc. can be set; various light source irradiation angles can be set, such as 90° vertical irradiation, 45° oblique irradiation, etc.; and various light source camera position relationships can be set, such as the camera and the light source are on the same side of the polarizer to be detected, that is, the polarizer to be detected is irradiated and the image is collected in a reflection manner, and the camera and the light source are respectively on both sides of the polarizer to be detected, that is, the polarizer to be detected is irradiated and the image is collected in a transmission manner. Under each collection condition, an image of the polarizer to be detected is collected respectively.

[0043] In one embodiment of the present invention, the defect detection model may include a feature extraction unit, at least N target sequence generation units, and at least N defect judgment units. The feature extraction unit may include an encoder, the target sequence generation unit may include a decoder, and the defect judgment unit may include a classifier.

[0044] In one embodiment of the present invention, the encoder and decoder can be taken from the Transformer network. In addition to the encoder, the feature extraction unit can also include a backbone network. After the backbone network performs preliminary feature extraction, the extracted features are converted into vectors of fixed length by the encoder, that is, feature information is obtained. In one embodiment of the present invention, the backbone network can adopt a convolutional neural network, etc., and the classifier can adopt a convolutional neural network or a softmax layer, etc. That is to say, in terms of network structure, the defect detection model of a specific embodiment of the present invention can be composed of a convolutional neural network and an improved Transformer network, and the improved Transformer network includes an encoder and multiple decoders.

[0045] In one embodiment of the present invention, the defect detection model can be pre-trained through a training set, which contains a large number of polarizer defect sample images, and the polarizer defect sample images are labeled with defect types, thereby enabling the defect detection model to have the ability to distinguish defect types.

[0046] In actual detection applications, the input module inputs each image acquired by the acquisition module into the trained defect detection model. The process of the defect detection model processing the input image is shown in the following figure. Figure 2 shown.

[0047] In one embodiment of the present invention, the polarizer to be detected includes at least a protective film, an upper TAC layer, a PVA layer, a lower TAC layer, a PSA layer and a release film from top to bottom, and the layer where the defect is located in the polarizer to be detected includes the above-mentioned layers. Taking the above-mentioned 6 layers as an example, the query vector is 6 corresponding ones. In one embodiment of the present invention, the query vector can be obtained in another model training process. Specifically, the sample image of the polarizer defect in the training set can also be labeled with the layer where the defect is located, and the defect detection model is newly trained under this labeling method. After training, multiple feature vectors corresponding to the layer where each defect is located are obtained, and the multiple feature vectors corresponding to the layer where each defect is located are averaged to obtain the query vector corresponding to the layer where the defect is located. It should be understood that for each defect, a feature vector of the layer where it is located can be obtained. The average of all feature vectors corresponding to all defects in each layer can obtain the query vector corresponding to the layer.

[0048] By inputting the feature information of N images and multiple query vectors into the target sequence generation unit, the target sequence generation unit can obtain the sequence of target defects in the corresponding image according to the feature information, and at the same time, can obtain the layer where the target defect is located in combination with the query vector. It should be understood that if there is no defect in an image, the sequence of its target defects and the layer where its target defects are located will not be obtained here; if there are multiple defects in an image, the sequence of its multiple target defects can be obtained here, and the layer where each target defect is located can be obtained.

[0049] If at least one sequence of target defects of a certain image is obtained, the at least one sequence is input into a corresponding defect judgment unit, and finally the type of each target defect in the image is output.

[0050] According to the appearance defect detection device of the polarizer according to the embodiment of the present invention, a defect detection model is used to simultaneously extract features from multiple images, and multiple target sequence generation units and multiple defect judgment units are used to respectively generate target defect sequences and judge the target defect types for the multiple images, and the layer where the target defect is located is obtained by combining the query vector. Therefore, not only can the simultaneous detection of various appearance defects of the polarizer be achieved through the calculation of a computer, thereby improving the comprehensiveness and efficiency of the appearance defect detection of the polarizer, but also the layer where the defect is located can be detected, providing guidance for pre-feeding, process correction, optimization and subsequent defect repair.

[0051] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. "Multiple" means two or more, unless otherwise clearly and specifically defined.

[0052] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0053] In the present invention, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may mean that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, a first feature being "above", "above" or "above" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being "below", "below" or "below" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0054] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0055] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0056] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0057] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0058] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0059] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0060] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for detecting appearance defects of a polarizer, characterized in that: The following steps are involved: Acquire N images of the polarizer to be inspected, wherein every two of the images have at least one different acquisition condition, and N is an integer greater than 1; Inputting the N images into a defect detection model, wherein the defect detection model introduces an attention mechanism; In the defect detection model, a feature extraction unit is used to extract features from each of the images to obtain feature information corresponding to each of the images; In the defect detection model, the feature information of the N images is input into N target sequence generation units one by one, and in each of the target sequence generation units, a sequence of target defects in the image and the layer where the target defects are located are generated according to the feature information of the corresponding image and a plurality of query vectors, wherein the plurality of query vectors correspond one by one to the plurality of layers in the polarizer to be detected; In the defect detection model, the sequences of target defects in the N images are input into the N defect judgment units one by one, and in each defect judgment unit, the type of the target defect in the image is judged according to the sequence of target defects in the corresponding image. The feature extraction unit includes an encoder, the target sequence generation unit includes a decoder, and the defect judgment unit includes a classifier. The defect detection model is obtained by training a training set, wherein the training set includes polarizer defect sample images, and the polarizer defect sample images are annotated with defect types as labels. The polarizer defect sample images in the training set are also annotated with the defect layer as a label, and the defect detection model is newly trained under this annotation method. After training, multiple feature vectors corresponding to each defect layer are obtained, and the multiple feature vectors corresponding to each defect layer are averaged to obtain the query vector corresponding to the defect layer.

2. The method for detecting appearance defects of a polarizer according to claim 1, characterized in that: The acquisition conditions include: camera type, light source type, light source irradiation angle and light source camera position relationship.

3. A polarizer appearance defect detection device, characterized in that: include: An acquisition module, the acquisition module is used to acquire N images of the polarizer to be detected, wherein every two of the images have at least one different acquisition condition, and N is an integer greater than 1; An input module, wherein the input module is used to input the N images into a defect detection model, wherein the defect detection model introduces an attention mechanism, and in the defect detection model, a feature extraction unit is used to extract features from each image to obtain feature information corresponding to each image; in the defect detection model, the feature information of the N images is input into N target sequence generation units one by one, and in each target sequence generation unit, a sequence of target defects in the image and a layer where the target defects are located are generated according to the feature information of the corresponding image and a plurality of query vectors, wherein a plurality of query vectors correspond one by one to a plurality of layers of the polarizer to be detected; in the defect detection model, the sequences of target defects in the N images are input into N defect judgment units one by one, and in each defect judgment unit, a type of target defect in the image is judged according to the sequence of target defects in the corresponding image. The feature extraction unit includes an encoder, the target sequence generation unit includes a decoder, and the defect judgment unit includes a classifier. The defect detection model is obtained by training a training set, wherein the training set includes polarizer defect sample images, and the polarizer defect sample images are labeled with defect types as labels. The polarizer defect sample images in the training set are also annotated with the defect layer as a label, and the defect detection model is newly trained under this annotation method. After training, multiple feature vectors corresponding to each defect layer are obtained, and the multiple feature vectors corresponding to each defect layer are averaged to obtain the query vector corresponding to the defect layer.

4. The polarizer appearance defect detection device according to claim 3, characterized in that: The acquisition conditions include: camera type, light source type, light source irradiation angle and light source camera position relationship.

Images