Product quality analysis method, apparatus, device, and medium

Abstract

The product quality analysis method, device, equipment and medium provided in the application relate to the product quality analysis technical field.In the application, first, a lithium hydroxide particle image is acquired, and Fourier transform is performed on the lithium hydroxide particle image to form first particle frequency domain data;second, at least one second particle frequency domain data is formed by performing at least one low-frequency mask on the first particle frequency domain data;then, inverse Fourier transform is performed on each second particle frequency domain data to form at least one lithium hydroxide conversion image; further, the lithium hydroxide particle image is associatedly coded based on the at least one lithium hydroxide conversion image and the at least one second particle frequency domain data to form image coding features; finally, the image coding features are decoded to form image decoding data.Based on the above, the problem that the reliability of the quality analysis of the lithium hydroxide particles is relatively low in the prior art can be improved.

Description

Technical Field

[0001] This application relates to the field of product quality analysis technology, and more specifically, to a product quality analysis method, apparatus, equipment, and medium. Background Technology

[0002] Quality analysis of lithium hydroxide particles is a crucial step in evaluating their physical properties and performance. Existing methods for lithium hydroxide particle quality analysis typically rely on microscopic imaging techniques and image processing algorithms. Traditional image processing algorithms mainly depend on techniques such as contour recognition. However, edge detection methods are highly sensitive to factors such as image noise, particle overlap, and uneven illumination. These factors can lead to edge detection failures or the detection of incorrect edges, thus affecting the accuracy of dimensional measurements. Furthermore, edge detection algorithms usually depend on the settings of multiple parameters, such as the threshold of the edge detection operator and the filter size. Different parameter settings can lead to inconsistent detection results, increasing the instability of the method. In addition, for irregularly shaped particles, edge detection may struggle to accurately capture all edge information, resulting in inaccurate dimensional measurements. In other words, existing technologies suffer from relatively low reliability in the quality analysis of lithium hydroxide particles. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a product quality analysis method, apparatus, equipment and medium to improve the problem of relatively low reliability of quality analysis of lithium hydroxide particles in the prior art.

[0004] To achieve the above objectives, this application adopts the following technical solution:

[0005] A product quality analysis method, comprising:

[0006] Acquire images of lithium hydroxide particles and perform Fourier transform on the lithium hydroxide particle images to form first particle frequency domain data. The lithium hydroxide particle images are formed by image acquisition operations on lithium hydroxide particles using an electron microscope.

[0007] The first particle frequency domain data is subjected to at least one low-frequency mask to form at least one second particle frequency domain data, wherein one low-frequency mask corresponds to one second particle frequency domain data.

[0008] Perform inverse Fourier transform on each of the frequency domain data of the second particle to form at least one lithium hydroxide transformation image;

[0009] Based on the at least one lithium hydroxide conversion image and the at least one second particle frequency domain data, the lithium hydroxide particle image is correlated and encoded to form image coding features;

[0010] The image encoding features are decoded to form image decoding data, wherein the image decoding data is used to characterize the quality of the lithium hydroxide particles in the particle size distribution dimension.

[0011] In a preferred embodiment of this application, in the above-described product quality analysis method, the step of performing at least one low-frequency masking on the first particle frequency domain data to form at least one second particle frequency domain data includes:

[0012] Based on the size of the first particle frequency domain data, an initial parameter matrix is ​​generated, wherein the size of the initial parameter matrix is ​​the same as the size of the first particle frequency domain data, and each parameter in the initial parameter matrix is ​​equal to 1;

[0013] Using the center point of the initial parameter matrix as the center, randomly determine a rectangular region with a size smaller than the size of the initial parameter matrix, and update all parameters in the rectangular region to 0;

[0014] By performing the step of randomly determining a rectangular region with a size smaller than the size of the initial parameter matrix centered at the center point of the initial parameter matrix, and updating each parameter in the rectangular region to 0 at least once, at least one update parameter matrix is ​​formed. Wherein, when the number of executions of this step is greater than or equal to 2, the size of the rectangular region determined randomly in each two executions is different.

[0015] Each of the updated parameter matrices and the first particle frequency domain data are multiplied bitwise to achieve a low-frequency mask, forming at least one second particle frequency domain data.

[0016] In a preferred embodiment of this application, in the above-described product quality analysis method, the step of correlating and encoding the lithium hydroxide particle image based on the at least one lithium hydroxide conversion image and the at least one second particle frequency domain data to form image coding features includes:

[0017] Semantic coding is performed on the at least one lithium hydroxide converted image, the at least one second particle frequency domain data and the lithium hydroxide particle image respectively to form at least one converted image coding feature, at least one frequency domain data coding feature and particle image coding feature;

[0018] Based on the at least one transformed image coding feature, the particle image coding feature is subjected to a first association coding to form an image association coding feature;

[0019] Based on the at least one frequency domain data coding feature, the image association coding feature is subjected to a second association coding to form an image coding feature, wherein the second association coding includes feature space mapping of the frequency domain data coding feature.

[0020] In a preferred embodiment of this application, in the above-described product quality analysis method, the step of performing a first association encoding on the particle image encoding features based on the at least one transformed image encoding feature to form an image association encoding feature includes:

[0021] The coding features of multiple transformed images are sorted in descending order of the range of the low-frequency mask to form the first coding feature sequence;

[0022] At the a-th time step, the first encoded feature sequence is traversed in a forward direction to form the transformed image encoded feature at the a-th time step, where a is an odd number greater than or equal to 1;

[0023] At the b-th time step, the first encoded feature sequence is traversed in reverse to form the transformed image encoded feature at the b-th time step, where b is an even number greater than 1. The traversal stops when the transformed image encoded feature at the current time step coincides with the transformed image encoded feature at the next time step.

[0024] Based on the order in which each of the transformed image coding features is traversed, the particle image coding features are sequentially subjected to first association coding to form image association coding features.

[0025] In a preferred embodiment of this application, in the above-mentioned product quality analysis method, the step of sequentially performing a first association encoding on the particle image encoding features based on the order in which each of the transformed image encoding features is traversed, to form image association encoding features, includes:

[0026] For the first transformed image coding feature traversed, a fully connected process is performed on the transformed image coding feature to obtain a fully connected feature. The fully connected feature is then activated to form an activated feature. Each parameter in the activated feature is used as a weight coefficient of the parameter at the corresponding position in the particle image coding feature to adjust the parameters of the particle image coding feature and form an adjusted particle image coding feature. Finally, the particle image coding feature and the adjusted particle image coding feature are added together to form the local association coding feature corresponding to the first transformed image coding feature.

[0027] For the transformed image coding features other than the first one traversed, the transformed image coding features are fully connected to obtain fully connected features. The fully connected features are then activated to form activated features. Each parameter in the activated features is used as a weight coefficient of the parameter at the corresponding position in the local association coding feature corresponding to the previous transformed image coding feature to adjust the parameters of the local association coding feature to form an adjusted local association coding feature. Finally, the local association coding feature and the adjusted local association coding feature are added together to form the local association coding feature corresponding to the current transformed image coding feature.

[0028] Based on the local correlation coding features corresponding to the coding features of the last transformed image, image correlation coding features are formed.

[0029] In a preferred embodiment of this application, in the above-described product quality analysis method, the step of performing a second association encoding on the image association encoding features based on the at least one frequency domain data encoding feature to form image encoding features includes:

[0030] The encoding features of multiple frequency domain data are sorted in descending order of the range of the low-frequency mask to form a second encoding feature sequence;

[0031] At the c-th time step, the second coding feature sequence is traversed in a forward direction to form the frequency domain data coding feature at the c-th time step, where c is an odd number greater than or equal to 1;

[0032] At the d-th time step, the second coding feature sequence is traversed in reverse to form the frequency domain data coding feature at the d-th time step, where d is an even number greater than 1. The traversal stops when the frequency domain data coding feature at the current time step coincides with the frequency domain data coding feature at the next time step.

[0033] Based on the order in which each frequency domain data coding feature is traversed, the image association coding features are sequentially subjected to second association coding to form image coding features.

[0034] In a preferred embodiment of this application, in the above-described product quality analysis method, the step of sequentially performing second association encoding on the image association encoding features based on the order in which each of the frequency domain data encoding features is traversed to form image encoding features includes:

[0035] For the first frequency domain data coding feature traversed, the image association coding feature is adjusted based on the attention parameter between the frequency domain data coding feature and the image association coding feature to form an adjusted image association coding feature. Then, the image association coding feature and the adjusted image association coding feature are added together to form the local association coding feature corresponding to the first frequency domain data coding feature.

[0036] For frequency domain data coding features other than the first one traversed, the parameters of the local correlation coding features are adjusted based on the attention parameters between the frequency domain data coding features and the local correlation coding features corresponding to the previous frequency domain data coding features, forming the adjusted local correlation coding features. The local correlation coding features and the adjusted local correlation coding features are then added together to form the local correlation coding features corresponding to the current frequency domain data coding features.

[0037] Image coding features are formed based on the local correlation coding features corresponding to the last frequency domain data coding features.

[0038] This application also provides a product quality analysis device, comprising:

[0039] The frequency domain data acquisition module is used to acquire images of lithium hydroxide particles and perform Fourier transform on the lithium hydroxide particle images to form first particle frequency domain data. The lithium hydroxide particle images are formed by image acquisition operations on lithium hydroxide particles using an electron microscope.

[0040] A low-frequency masking module is used to perform at least one low-frequency masking on the first particle frequency domain data to form at least one second particle frequency domain data, wherein one low-frequency masking corresponds to one second particle frequency domain data.

[0041] The inverse Fourier transform module is used to perform inverse Fourier transform on each of the frequency domain data of the second particle to form at least one lithium hydroxide transformed image;

[0042] An image association coding module is used to perform association coding on the lithium hydroxide particle image based on the at least one lithium hydroxide converted image and the at least one second particle frequency domain data to form image coding features;

[0043] The encoding feature decoding module is used to decode the image encoding features to form image decoding data, wherein the image decoding data is used to characterize the quality of the lithium hydroxide particles in the particle size distribution dimension.

[0044] Based on the above, this application also provides an electronic device, including:

[0045] Memory, used to store computer programs;

[0046] A processor connected to the memory is used to execute the computer program stored in the memory to implement the product quality analysis method described above.

[0047] Based on the above, this application also provides a computer-readable storage medium storing a computer program that, when executed, performs the various steps of the product quality analysis method described above.

[0048] The product quality analysis method, apparatus, equipment, and medium provided in this application first acquire images of lithium hydroxide particles and perform Fourier transform on the lithium hydroxide particle images to form first particle frequency domain data; second, perform at least one low-frequency masking on the first particle frequency domain data to form at least one second particle frequency domain data; then, perform inverse Fourier transform on each second particle frequency domain data to form at least one lithium hydroxide transformed image; further, perform correlation encoding on the lithium hydroxide particle images based on at least one lithium hydroxide transformed image and at least one second particle frequency domain data to form image encoding features; finally, decode the image encoding features to form image decoded data. Based on the above, since low-frequency components are usually related to the overall distribution of particles and background information, while high-frequency components are related to the surface details and edges of particles, the low-frequency masking allows the formed second particle frequency domain data to focus on high-frequency components, thereby achieving focused mining of edge information. This enables the formed encoding features to accurately characterize edge information, thus ensuring the reliability of the quality in the particle size (size) distribution dimension obtained from decoding. Furthermore, an inverse Fourier transform will be performed, enabling the lithium hydroxide particle images to be correlated and encoded based on the lithium hydroxide transformed image in the spatial domain and the second particle frequency domain data in the time domain, respectively. This further improves the reliability of the correlation encoding, thereby addressing the problem of relatively low reliability in the quality analysis of lithium hydroxide particles in the existing technology. Attached Figure Description

[0049] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings.

[0050] Figure 1 A structural block diagram of an electronic device provided in an embodiment of this application.

[0051] Figure 2 This is a flowchart illustrating the product quality analysis method provided in an embodiment of this application.

[0052] Figure 3 A schematic diagram of the first association encoding provided in the embodiments of this application.

[0053] Figure 4 A schematic diagram of the second association encoding provided in the embodiments of this application.

[0054] Figure 5 A block diagram of the product quality analysis device provided in the embodiments of this application. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0056] 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 the 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.

[0057] like Figure 1 As shown in the illustration, this application provides an electronic device. The electronic device may include a memory, a processor, and a product quality analysis device.

[0058] In detail, the memory and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, the memory and the processor can be electrically connected via one or more communication buses or signal lines. The product quality analysis device includes at least one software functional module stored in the memory in the form of software or firmware. The processor is used to execute executable computer programs stored in the memory, such as the software functional modules and computer programs included in the product quality analysis device, to implement the product quality analysis method provided in the embodiments of this application.

[0059] Optionally, the memory may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0060] Optionally, the processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0061] Understandable. Figure 1 The structure shown is for illustrative purposes only; the electronic device may also include components that are more advanced than those shown. Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown may include, for example, a communication unit for exchanging information with other devices, such as an electron microscope.

[0062] Combination Figure 2 This application also provides a product quality analysis method applicable to the aforementioned electronic device. The method steps defined in the process of the product quality analysis method can be implemented by the electronic device. The following will describe... Figure 2 The specific process shown will be explained in detail.

[0063] Step S110: Obtain an image of lithium hydroxide particles and perform a Fourier transform on the lithium hydroxide particle image to form first particle frequency domain data.

[0064] In this embodiment, the electronic device can acquire images of lithium hydroxide particles and perform a Fourier transform on the images (which converts the image from the spatial domain to the frequency domain, generating a spectrum) to form first particle frequency domain data. The lithium hydroxide particle images are formed by image acquisition using an electron microscope (considering the micrometer-scale size of lithium hydroxide particles, a high-resolution electron microscope can be used).

[0065] Step S120: Perform at least one low-frequency mask on the first particle frequency domain data to form at least one second particle frequency domain data.

[0066] In this embodiment, after forming the first particle frequency domain data, the electronic device can perform at least one low-frequency masking on the first particle frequency domain data to form at least one second particle frequency domain data. Each low-frequency masking corresponds to one second particle frequency domain data, where low-frequency masking refers to masking low-frequency components. Furthermore, the spectrum (i.e., frequency domain data) is a two-dimensional frequency distribution map, with the central region corresponding to low-frequency components and the edge regions corresponding to high-frequency components. Therefore, the central region can be masked to form the corresponding second particle frequency domain data.

[0067] Step S130: Perform inverse Fourier transform on each of the frequency domain data of the second particle to form at least one lithium hydroxide transformed image.

[0068] In this embodiment of the application, after the second particle frequency domain data is formed, the electronic device can perform an inverse Fourier transform (which is the inverse of the Fourier transform, i.e., from the frequency domain to the spatial domain) on each of the second particle frequency domain data to form at least one lithium hydroxide conversion image. In other words, one second particle frequency domain data corresponds to one lithium hydroxide conversion image.

[0069] Step S140: Based on the at least one lithium hydroxide conversion image and the at least one second particle frequency domain data, the lithium hydroxide particle image is correlated and encoded to form image coding features.

[0070] In this embodiment, after forming the lithium hydroxide conversion image, the electronic device can perform associative encoding on the lithium hydroxide particle image based on the at least one lithium hydroxide conversion image and the at least one second particle frequency domain data to form image coding features. That is, semantic information of the lithium hydroxide particle image can be correlated and mined based on the spatial and frequency domain semantic information of high-frequency components, resulting in higher accuracy of the formed image coding features, i.e., greater attention is paid to edge information.

[0071] Step S150: Decode the image encoding features to form image decoding data.

[0072] In this embodiment, after forming the image encoding features, the electronic device can decode the image encoding features to form image decoding data. The image decoding data is used to characterize the quality of the lithium hydroxide particles in the particle size distribution dimension. For example, the image encoding features can be processed through a fully connected network layer to obtain corresponding output features (size can be 1*1). Then, the output features are processed (such as identity mapping or linear mapping) to obtain image decoding data, such as a value from 1 to 10, where a higher value indicates higher quality and a lower value indicates lower quality.

[0073] It should be noted that steps S140 and S150 above can be implemented using a corresponding neural network model. This neural network model can be formed by learning from the sample lithium hydroxide particle image and the corresponding quality label (e.g., a concentrated size distribution within the target range indicates higher quality, while a non-concentrated size distribution or not belonging to the target range indicates lower quality). Specifically, the neural network model can include an encoding network and a decoding network. The encoding network can be used for associative encoding, and the decoding network can include fully connected network layers for decoding.

[0074] Based on the above, since low-frequency components are generally related to the overall distribution of particles and background information, while high-frequency components are related to the surface details and edges of particles, low-frequency masking allows the resulting second particle frequency domain data to focus on high-frequency components. This enables targeted mining of edge information, ensuring that the resulting encoded features accurately represent edge information and guaranteeing the reliability of the decoded data in terms of particle size distribution. Furthermore, inverse Fourier transform is performed, allowing for correlation coding of lithium hydroxide particle images based on the spatial domain lithium hydroxide transformed image and the temporal domain second particle frequency domain data, further improving the reliability of the correlation coding and addressing the relatively low reliability of lithium hydroxide particle quality analysis in existing technologies.

[0075] Firstly, regarding step S120, it should be noted that the specific method of performing at least one low-frequency mask on the frequency domain data of the first particle is not limited and can be selected according to actual needs.

[0076] For example, in an alternative implementation, low-frequency masking can be performed based on a defined frequency threshold; for instance, regions with frequencies below the threshold can be masked. This frequency threshold can be set empirically or formed during training as a parameter of the aforementioned neural network model.

[0077] For example, in another alternative implementation, in order to improve the overall reliability of the low-frequency mask in characterizing high-frequency components, i.e., to avoid the inherent bias caused by a fixed frequency threshold, step S120 above may further include the following:

[0078] First, an initial parameter matrix can be generated based on the size of the first particle frequency domain data, wherein the size of the initial parameter matrix is ​​the same as the size of the first particle frequency domain data, and each parameter in the initial parameter matrix is ​​equal to 1;

[0079] Secondly, taking the center point of the initial parameter matrix as the center, a rectangular region with a size smaller than the size of the initial parameter matrix is ​​randomly determined, and each parameter in the rectangular region is updated to 0. In this way, a corresponding updated parameter matrix can be formed.

[0080] Then, by performing the step of randomly determining a rectangular region with a size smaller than the size of the initial parameter matrix centered on the center point of the initial parameter matrix and updating each parameter in the rectangular region to 0 at least once, at least one updated parameter matrix is ​​formed. When the number of executions of this step is greater than or equal to 2, the size of the rectangular region determined twice is different. In this way, low-frequency components at different scales can be suppressed. This multi-scale processing can more comprehensively capture the characteristics of particles at different scales and improve the comprehensiveness of detection.

[0081] Finally, each of the updated parameter matrices and the first particle frequency domain data can be multiplied bitwise to achieve a low-frequency mask and form at least one second particle frequency domain data. For example, the first updated parameter matrix can be multiplied with the first particle frequency domain data to form the first second particle frequency domain data, and the second updated parameter matrix can be multiplied with the first particle frequency domain data to form the second second particle frequency domain data.

[0082] Secondly, regarding step S140, it should be noted that the specific method for associating and encoding the lithium hydroxide particle image based on the at least one lithium hydroxide conversion image and the at least one second particle frequency domain data is not limited and can be selected according to the actual situation.

[0083] For example, in an alternative implementation, cross-attention processing can be performed on the coding features corresponding to the lithium hydroxide particle images based on the coding features corresponding to each lithium hydroxide conversion image and the coding features corresponding to each second particle frequency domain data, to form each corresponding cross-attention feature. Then, each cross-attention feature can be averaged or summed to form image coding features.

[0084] For example, in another alternative implementation, in order to improve the reliability of the association coding, the above step S140 may further include steps S141, S142 and S143, the specific contents of each step are as follows.

[0085] Step S141: Semantic encoding is performed on the at least one lithium hydroxide converted image, the at least one second particle frequency domain data, and the lithium hydroxide particle image respectively to form at least one converted image encoding feature, at least one frequency domain data encoding feature, and particle image encoding feature.

[0086] In this embodiment, semantic encoding is performed on the at least one lithium hydroxide converted image, the at least one second particle frequency domain data, and the lithium hydroxide particle image to form at least one converted image encoding feature, at least one frequency domain data encoding feature, and a particle image encoding feature. For example, a lithium hydroxide converted image can be convolved to form a converted image encoding feature, a second particle frequency domain data can be convolved to form a frequency domain data encoding feature, and the lithium hydroxide particle image can be convolved to form a particle image encoding feature.

[0087] Step S142: Based on the at least one transformed image coding feature, perform a first association coding on the particle image coding feature to form an image association coding feature.

[0088] In this embodiment of the application, after forming the transformed image coding feature and the particle image coding feature, the particle image coding feature can be first associated with the at least one transformed image coding feature to form an image association coding feature.

[0089] Step S143: Based on the at least one frequency domain data coding feature, perform a second association coding on the image association coding feature to form an image coding feature.

[0090] In this embodiment, after forming the transformed image coding features and the particle image coding features, a second association coding can be performed on the image association coding features based on the at least one frequency domain data coding feature to form image coding features. The second association coding includes feature space mapping of the frequency domain data coding features. That is, since both the transformed image coding features and the particle image coding features belong to the semantic features of the spatial domain, association coding can be performed first to ensure the reliability of the association coding. Furthermore, since the frequency domain data coding features belong to the semantic features of the frequency domain, which are different feature domains from the image association coding features of the spatial domain, feature space mapping is required to perform association coding in a similar feature domain (i.e., a similar semantic space), thereby improving the reliability of the second association coding.

[0091] It is understood that the specific method of performing the first association encoding on the particle image encoding features in step S142 above is not limited. For example, in an alternative implementation, in order to achieve the mining of complex latent semantic relationships through the first association encoding, step S142 above may further include steps S142a, S142b, S142c, and S142d, the specific contents of each step of which are as follows (in conjunction with...). Figure 3 ).

[0092] Step S142a: Sort the multiple converted image coding features in descending order of the range of the low-frequency mask to form the first coding feature sequence.

[0093] In this embodiment of the application, multiple converted image coding features can be sorted in descending order of the range of the low-frequency mask to form a first coding feature sequence, that is, sorted in descending order of the size of the aforementioned rectangular regions.

[0094] In step S142b, at the a-th time step, the first encoded feature sequence is traversed in a forward direction to form the transformed image encoded features at the a-th time step.

[0095] In this embodiment of the application, after the first coding feature sequence is formed, the first coding feature sequence can be traversed in a forward direction at the a-th time step to form the transformed image coding feature at the a-th time step, where a is an odd number greater than or equal to 1, such as the first time step, the third time step, the fifth time step, the seventh time step, etc.

[0096] In step S142c, at the b-th time step, the first encoded feature sequence is traversed in reverse to form the transformed image encoded feature at the b-th time step.

[0097] In this embodiment of the application, after forming the first encoded feature sequence, the first encoded feature sequence can be reversed at the b-th time step to form the transformed image encoded feature at the b-th time step. Here, b is an even number greater than 1, such as the second time step, the fourth time step, the sixth time step, the eighth time step, etc. The traversal stops when the transformed image encoded feature at the current time step coincides with the transformed image encoded feature at the next time step.

[0098] Step S142d: Based on the order in which each of the transformed image coding features is traversed, the particle image coding features are sequentially subjected to first association coding to form image association coding features.

[0099] In this embodiment, the particle image coding features can be sequentially associated with each transformed image coding feature based on the order in which they are traversed, forming image association coding features, such as the first transformed image coding feature, the last transformed image coding feature, the second transformed image coding feature, and so on. This ensures that the difference between two adjacent transformed image coding features gradually decreases over time, allowing for a gradual increase in information fusion. Furthermore, the size of the rectangular region corresponding to the intermediate transformed image coding features is moderate, avoiding the problem of excessive loss of semantic information from high-frequency components or excessive focus on semantic information from low-frequency components caused by ending association coding at the beginning or end of the sequence.

[0100] It is understood that in step S142d above, the specific method of sequentially performing the first association encoding on the particle image encoding features is not limited. For example, in an alternative implementation, considering that both the transformed image encoding features and the particle image encoding features belong to the semantic features of the spatial domain, using one as the adjustment basis for the other has relatively reliable stability and can ensure the reliability of the first association encoding. Thus, step S142d above may include:

[0101] First, for the first transformed image coding feature encountered, a fully connected process is performed on the transformed image coding feature (i.e., the transformed image coding feature is multiplied by the weight matrix of the fully connected layer to obtain the corresponding multiplied feature; then, the multiplied feature is added to the bias parameters of the fully connected layer to complete the fully connected process), resulting in a fully connected feature. Next, the fully connected feature is activated (e.g., using the sigmoid function) to form an activated feature. Then, each parameter in the activated feature is used as a weight coefficient for the corresponding position in the particle image coding feature to adjust the parameters of the particle image coding feature (i.e., the weight coefficient is multiplied by the corresponding parameter), forming an adjusted particle image coding feature. Finally, the particle image coding feature and the adjusted particle image coding feature are added together (to avoid the loss of global semantic information) to form the local association coding feature corresponding to the first transformed image coding feature.

[0102] Secondly, for the transformed image coding features other than the first one traversed, the transformed image coding features are fully connected to obtain fully connected features. The fully connected features are then activated to form activated features. Each parameter in the activated features is used as a weight coefficient of the parameter at the corresponding position in the local association coding feature corresponding to the previous transformed image coding feature to adjust the parameters of the local association coding feature to form an adjusted local association coding feature. Finally, the local association coding feature and the adjusted local association coding feature are added together to form the local association coding feature corresponding to the current transformed image coding feature.

[0103] Finally, image association coding features can be formed based on the local association coding features corresponding to the last transformed image coding features. For example, the local association coding features corresponding to the last transformed image coding features can be directly used as the image association coding features.

[0104] It is understood that in step S143 above, the specific method of performing the second association encoding on the image association encoding features is not limited. For example, in an alternative implementation, in order to achieve the mining of complex latent semantic relationships through the second association encoding, step S143 above may further include steps S143a, S143b, S143c, and S143d, the specific contents of each step of which are as follows (in conjunction with...). Figure 4 ).

[0105] Step S143a: Sort the multiple frequency domain data coding features in descending order of the range of the low-frequency mask to form a second coding feature sequence.

[0106] In this embodiment of the application, multiple frequency domain data coding features can be sorted in descending order of the range of the low-frequency mask to form a second coding feature sequence, that is, sorted in descending order of the size of the aforementioned rectangular regions.

[0107] In step S143b, at the c-th time step, the second coding feature sequence is traversed in a forward direction to form the frequency domain data coding feature at the c-th time step.

[0108] In this embodiment of the application, after forming the second coded feature sequence, the second coded feature sequence can be traversed forward at the c-th time step to form the frequency domain data coded feature at the c-th time step. Here, c is an odd number greater than or equal to 1, such as the first time step, the third time step, the fifth time step, the seventh time step, etc.

[0109] In step S143c, at the d-th time step, the second coding feature sequence is traversed in reverse to form the frequency domain data coding feature at the d-th time step.

[0110] In this embodiment of the application, after forming the second coded feature sequence, the second coded feature sequence can be reversed at the d-th time step to form the frequency domain data coded feature at the d-th time step. Here, d is an even number greater than 1, such as the second time step, the fourth time step, the sixth time step, the eighth time step, etc. The traversal stops when the frequency domain data coded feature of the current time step coincides with the frequency domain data coded feature of the next time step.

[0111] Step S143d: Based on the order in which each frequency domain data coding feature is traversed, the image association coding features are sequentially subjected to second association coding to form image coding features.

[0112] In this embodiment, the image association coding features can be sequentially associated with each frequency domain data coding feature in the order they are traversed, forming image coding features, such as the first frequency domain data coding feature, the last frequency domain data coding feature, the second frequency domain data coding feature, and so on. This ensures that the difference between two adjacent frequency domain data coding features gradually decreases over time, allowing for a gradual increase in information fusion. Furthermore, the size of the rectangular region corresponding to the intermediate frequency domain data coding features is moderate, avoiding the problem of excessive loss of semantic information from high-frequency components or excessive focus on semantic information from low-frequency components caused by ending association coding at the beginning or end of the sequence.

[0113] It is understood that in step S143d above, the specific method of sequentially performing the second association coding on the image association coding features is not limited. For example, in an alternative implementation, considering that the frequency domain data coding features and the image association coding features belong to the semantic features of the frequency domain and the semantic features of the spatial domain, respectively, it is difficult to guarantee the reliability of the second association coding if one is directly used as the adjustment basis for the other. Thus, step S143d above includes:

[0114] First, for the first frequency domain data coding feature encountered, the image association coding feature is adjusted based on the attention parameter between the frequency domain data coding feature and the image association coding feature (i.e., matrix multiplication of the query vector corresponding to the frequency domain data coding feature and the transpose of the key vector corresponding to the image association coding feature to obtain the corresponding attention parameter). This adjustment involves weighted summation based on the attention parameter, thus enabling cross-attention processing of the image association coding feature based on the frequency domain data coding feature. By performing cross-attention processing, complex latent semantic relationships can be captured, thereby improving the reliability of the obtained coding features and forming the adjusted image association coding feature. Finally, the image association coding feature and the adjusted image association coding feature are added together (to avoid the loss of global semantic information) to form the local association coding feature corresponding to the first frequency domain data coding feature.

[0115] Secondly, for the frequency domain data coding features other than the first one traversed, the parameters of the local correlation coding features are adjusted based on the attention parameters between the frequency domain data coding features and the local correlation coding features corresponding to the previous frequency domain data coding features, forming the adjusted local correlation coding features. Then, the local correlation coding features and the adjusted local correlation coding features are added together to form the local correlation coding features corresponding to the current frequency domain data coding features.

[0116] Finally, image coding features can be formed based on the local correlation coding features corresponding to the last frequency domain data coding features. For example, the local correlation coding features corresponding to the last frequency domain data coding features can be directly used as the image coding features.

[0117] Combination Figure 5 This application also provides a product quality analysis device applicable to the aforementioned electronic equipment. The product quality analysis device may include a frequency domain data acquisition module, a low-frequency mask module, an inverse Fourier transform module, an image association coding module, and a coded feature decoding module.

[0118] Specifically, the frequency domain data acquisition module can be used to acquire images of lithium hydroxide particles and perform Fourier transform on the lithium hydroxide particle images to form first particle frequency domain data. The lithium hydroxide particle images are formed by image acquisition operations on lithium hydroxide particles using an electron microscope. In this embodiment, the frequency domain data acquisition module can be used to perform... Figure 2 The relevant content regarding the frequency domain data acquisition module in step S110 shown can be found in the previous description of step S110.

[0119] Specifically, the low-frequency masking module can be used to perform at least one low-frequency masking on the first particle frequency domain data to form at least one second particle frequency domain data, wherein one low-frequency masking corresponds to one second particle frequency domain data. In the embodiments of this application, the low-frequency masking module can be used to perform... Figure 2 The relevant content regarding the low-frequency mask module in step S120 shown can be found in the previous description of step S120.

[0120] Specifically, the inverse Fourier transform module can be used to perform an inverse Fourier transform on each of the second particle's frequency domain data, forming at least one lithium hydroxide conversion image. In this embodiment, the inverse Fourier transform module can be used to perform... Figure 2 The relevant content regarding the inverse Fourier transform module in step S130 shown can be found in the previous description of step S130.

[0121] Specifically, the image association coding module can be used to perform association coding on the lithium hydroxide particle image based on the at least one lithium hydroxide converted image and the at least one second particle frequency domain data to form image coding features. In the embodiments of this application, the image association coding module can be used to perform... Figure 2 The relevant content regarding the image association coding module in step S140 shown can be found in the preceding description of step S140.

[0122] Specifically, the encoding feature decoding module can be used to decode the image encoding features to form image decoding data, wherein the image decoding data is used to characterize the quality of the lithium hydroxide particles in the particle size distribution dimension. In the embodiments of this application, the encoding feature decoding module can be used to perform... Figure 2 The relevant content regarding the encoding feature decoding module in step S150 shown can be found in the previous description of step S150.

[0123] In this embodiment of the application, corresponding to the product quality analysis method applied to the electronic device described above, a computer-readable storage medium is also provided, which stores a computer program that executes the various steps of the product quality analysis method when the computer program is run.

[0124] The steps executed by the aforementioned computer program during runtime will not be described in detail here, but can be found in the explanation of the product quality analysis method above.

[0125] In summary, the product quality analysis method, apparatus, equipment, and medium provided in this application firstly acquire images of lithium hydroxide particles and perform Fourier transform on these images to form first particle frequency domain data. Secondly, the first particle frequency domain data undergoes at least one low-frequency masking to form at least one second particle frequency domain data. Then, each second particle frequency domain data undergoes an inverse Fourier transform to form at least one lithium hydroxide transformed image. Further, the lithium hydroxide particle images are correlated and encoded based on at least one lithium hydroxide transformed image and at least one second particle frequency domain data to form image encoding features. Finally, the image encoding features are decoded to form image decoding data. Based on the above, since low-frequency components are usually related to the overall particle distribution and background information, while high-frequency components are related to the surface details and edges of the particles, low-frequency masking allows the formed second particle frequency domain data to focus on high-frequency components, thereby enabling focused mining of edge information. This allows the formed encoding features to accurately characterize edge information, ensuring the reliability of the quality in the particle size distribution dimension obtained from decoding. Furthermore, an inverse Fourier transform will be performed, enabling the lithium hydroxide particle images to be correlated and encoded based on the lithium hydroxide transformed image in the spatial domain and the second particle frequency domain data in the time domain, respectively. This further improves the reliability of the correlation encoding, thereby addressing the problem of relatively low reliability in the quality analysis of lithium hydroxide particles in the existing technology.

[0126] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0127] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0128] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, electronic device, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks. 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 apparatus 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 apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0129] The above description is merely a preferred embodiment of this application and is not intended to limit 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 protection scope of this application.

Claims

1. A product quality analysis method, characterized in that, include: Acquire images of lithium hydroxide particles and perform Fourier transform on the lithium hydroxide particle images to form first particle frequency domain data. The lithium hydroxide particle images are formed by image acquisition operations on lithium hydroxide particles using an electron microscope. The first particle frequency domain data is subjected to at least one low-frequency mask to form at least one second particle frequency domain data, wherein one low-frequency mask corresponds to one second particle frequency domain data. Perform inverse Fourier transform on each of the frequency domain data of the second particle to form at least one lithium hydroxide transformation image; The lithium hydroxide particle image is correlated and encoded based on the at least one lithium hydroxide converted image and the at least one second particle frequency domain data to form image coding features. This includes: semantically encoding the at least one lithium hydroxide converted image, the at least one second particle frequency domain data, and the lithium hydroxide particle image respectively to form at least one converted image coding feature, at least one frequency domain data coding feature, and a particle image coding feature; performing a first correlation coding on the particle image coding feature based on the at least one converted image coding feature to form image correlation coding features; and performing a second correlation coding on the image correlation coding feature based on the at least one frequency domain data coding feature to form image coding features, wherein the second correlation coding includes feature space mapping of the frequency domain data coding feature. The image encoding features are decoded to form image decoding data, wherein the image decoding data is used to characterize the quality of the lithium hydroxide particles in the particle size distribution dimension.

2. The product quality analysis method according to claim 1, characterized in that, The step of performing at least one low-frequency mask on the first particle frequency domain data to form at least one second particle frequency domain data includes: Based on the size of the first particle frequency domain data, an initial parameter matrix is ​​generated, wherein the size of the initial parameter matrix is ​​the same as the size of the first particle frequency domain data, and each parameter in the initial parameter matrix is ​​equal to 1; Using the center point of the initial parameter matrix as the center, randomly determine a rectangular region with a size smaller than the size of the initial parameter matrix, and update all parameters in the rectangular region to 0; By performing the step of randomly determining a rectangular region with a size smaller than the size of the initial parameter matrix centered at the center point of the initial parameter matrix, and updating each parameter in the rectangular region to 0 at least once, at least one update parameter matrix is ​​formed. Wherein, when the number of executions of this step is greater than or equal to 2, the size of the rectangular region determined randomly in each two executions is different. Each of the updated parameter matrices and the first particle frequency domain data are multiplied bitwise to achieve a low-frequency mask, forming at least one second particle frequency domain data.

3. The product quality analysis method according to claim 1, characterized in that, The step of performing a first association coding on the particle image coding features based on the at least one transformed image coding feature to form image association coding features includes: The coding features of multiple transformed images are sorted in descending order of the range of the low-frequency mask to form the first coding feature sequence; At the a-th time step, the first encoded feature sequence is traversed in a forward direction to form the transformed image encoded feature at the a-th time step, where a is an odd number greater than or equal to 1; At the b-th time step, the first encoded feature sequence is traversed in reverse to form the transformed image encoded feature at the b-th time step, where b is an even number greater than 1. The traversal stops when the transformed image encoded feature at the current time step coincides with the transformed image encoded feature at the next time step. Based on the order in which each of the transformed image coding features is traversed, the particle image coding features are sequentially subjected to first association coding to form image association coding features.

4. The product quality analysis method according to claim 3, characterized in that, The step of performing a first association encoding on the particle image encoding features sequentially based on the order in which each of the transformed image encoding features is traversed, to form image association encoding features, includes: For the first transformed image coding feature traversed, a fully connected process is performed on the transformed image coding feature to obtain a fully connected feature. The fully connected feature is then activated to form an activated feature. Each parameter in the activated feature is used as a weight coefficient of the parameter at the corresponding position in the particle image coding feature to adjust the parameters of the particle image coding feature and form an adjusted particle image coding feature. Finally, the particle image coding feature and the adjusted particle image coding feature are added together to form the local association coding feature corresponding to the first transformed image coding feature. For the transformed image coding features other than the first one traversed, the transformed image coding features are fully connected to obtain fully connected features. The fully connected features are then activated to form activated features. Each parameter in the activated features is used as a weight coefficient of the parameter at the corresponding position in the local association coding feature corresponding to the previous transformed image coding feature to adjust the parameters of the local association coding feature to form an adjusted local association coding feature. Finally, the local association coding feature and the adjusted local association coding feature are added together to form the local association coding feature corresponding to the current transformed image coding feature. Based on the local correlation coding features corresponding to the coding features of the last transformed image, image correlation coding features are formed.

5. The product quality analysis method according to claim 1, characterized in that, The step of performing a second association coding on the image association coding features based on the at least one frequency domain data coding feature to form image coding features includes: The encoding features of multiple frequency domain data are sorted in descending order of the range of the low-frequency mask to form a second encoding feature sequence; At the c-th time step, the second coding feature sequence is traversed in a forward direction to form the frequency domain data coding feature at the c-th time step, where c is an odd number greater than or equal to 1; At the d-th time step, the second coding feature sequence is traversed in reverse to form the frequency domain data coding feature at the d-th time step, where d is an even number greater than 1. The traversal stops when the frequency domain data coding feature at the current time step coincides with the frequency domain data coding feature at the next time step. Based on the order in which each frequency domain data coding feature is traversed, the image association coding features are sequentially subjected to second association coding to form image coding features.

6. The product quality analysis method according to claim 5, characterized in that, The step of performing a second association encoding on the image association encoding features sequentially based on the order in which each of the frequency domain data encoding features is traversed, to form image encoding features, includes: For the first frequency domain data coding feature traversed, the image association coding feature is adjusted based on the attention parameter between the frequency domain data coding feature and the image association coding feature to form an adjusted image association coding feature. Then, the image association coding feature and the adjusted image association coding feature are added together to form the local association coding feature corresponding to the first frequency domain data coding feature. For frequency domain data coding features other than the first one traversed, the parameters of the local correlation coding features are adjusted based on the attention parameters between the frequency domain data coding features and the local correlation coding features corresponding to the previous frequency domain data coding features, forming the adjusted local correlation coding features. The local correlation coding features and the adjusted local correlation coding features are then added together to form the local correlation coding features corresponding to the current frequency domain data coding features. Image coding features are formed based on the local correlation coding features corresponding to the last frequency domain data coding features.

7. A product quality analysis device, characterized in that, include: The frequency domain data acquisition module is used to acquire images of lithium hydroxide particles and perform Fourier transform on the lithium hydroxide particle images to form first particle frequency domain data. The lithium hydroxide particle images are formed by image acquisition operations on lithium hydroxide particles using an electron microscope. A low-frequency masking module is used to perform at least one low-frequency masking on the first particle frequency domain data to form at least one second particle frequency domain data, wherein one low-frequency masking corresponds to one second particle frequency domain data. The inverse Fourier transform module is used to perform inverse Fourier transform on each of the frequency domain data of the second particle to form at least one lithium hydroxide transformed image; An image association coding module is used to perform association coding on the lithium hydroxide particle image based on the at least one lithium hydroxide converted image and the at least one second particle frequency domain data to form image coding features. The module includes: performing semantic coding on the at least one lithium hydroxide converted image, the at least one second particle frequency domain data, and the lithium hydroxide particle image respectively to form at least one converted image coding feature, at least one frequency domain data coding feature, and a particle image coding feature; performing a first association coding on the particle image coding feature based on the at least one converted image coding feature to form image association coding features; and performing a second association coding on the image association coding feature based on the at least one frequency domain data coding feature to form image coding features, wherein the second association coding includes feature space mapping of the frequency domain data coding feature. The encoding feature decoding module is used to decode the image encoding features to form image decoding data, wherein the image decoding data is used to characterize the quality of the lithium hydroxide particles in the particle size distribution dimension.

8. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor connected to the memory is used to execute the computer program stored in the memory to implement the product quality analysis method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains a computer program that, when executed, performs the product quality analysis method according to any one of claims 1-6.

Images