Dynamic compensation method and system for measurement error of optical encoder and storage medium

Abstract

The invention relates to a quantum dot optical encoder measurement error dynamic compensation method and system and a storage medium. The method comprises the following steps: acquiring a multi-color-gamut code channel image according to a quantum dot optical encoder; constructing a sub-pixel level compensation network model; a code channel fine feature extraction network, a causal inference network, a decision network, a sub-pixel level positioning network, a coding and decoding network and Q learning are introduced, so that a single-factor measurement error compensation agent is constructed; parallel decomposition multi-subtask model training is adopted, the output of a causal inference network is combined, and the compensation agent strategy proportion of each influence factor is calculated, so that a measurement error compensation agent under the influence of multiple factors is constructed; and measuring error dynamic compensation is completed. By adopting the method, decoupling of coupling errors of internal and external factors can be realized, and error compensation can be performed in a more targeted manner by analyzing the strategy ratio of each compensation sub-agent, so that the measurement precision is improved.

Description

Technical Field

[0001] This application relates to the field of high-precision measurement, and in particular to a method, system, and storage medium for dynamic compensation of measurement errors of quantum dot optical encoders. Background Technology

[0002] Optical encoders, as key components in micro-nano ultra-precision manufacturing, are widely used in high-end precision instruments and equipment such as industrial micro-nano robots, mid-to-high-end CNC machine tools, and precision measuring instruments, determining the measurement and control accuracy of these instruments and equipment. Currently, optical encoders primarily employ "black and white" chrome film technology to create black and white code tracks on a base scale. For the black and white chrome film gratings, a "0-1" binary encoding mode is mainly used for encoding. The drawback is that when representing the same length of displacement, binary encoding requires a large number of code elements (0 or 1) (taking quaternary and binary encoding as examples, the information capacity of quaternary absolute position encoding is 1 / 2, while that of binary is 1 / 2). In actual products, typically 16 binary code elements are used to represent one absolute position code area to achieve encoding for measurement strokes exceeding 0.5 m. This binary encoding mechanism leads to excessively long encoding code areas, severely limiting the measurement stroke of the optical encoder and making it difficult to achieve high-precision measurement with large strokes.

[0003] Based on the current state of research on optical encoders, the research group took a different approach, fabricating quantum dot self-emissive gratings using micro-nano additive manufacturing technology to develop a quantum dot optical encoder. Further in-depth research revealed that by introducing multi-color gamut code tracks (i.e., multi-color code tracks made from a material called quantum dots) to recompile the absolute position code of the quantum dot optical encoder, replacing the black and white code tracks in traditional optical encoders with multi-color gamut code tracks, a larger encoding / decoding capacity, more easily distinguishable absolute position codes, and more accurate absolute position measurement precision can be achieved. This improves decoding accuracy and measurement stability, better meeting the measurement needs of current high-end manufacturing industries.

[0004] Current research on optical encoder error compensation methods mainly focuses on the comprehensive error compensation of "black and white" optical encoders. However, this patent focuses on quantum dot optical encoders, which employ a multi-color-domain, multi-level encoding method for code track compilation. This differs significantly from the binary encoding method of traditional "black and white" optical encoders, which relies solely on "0" and "1," thus introducing new problems and technical challenges. Specifically, because quantum dot optical encoders use multi-level encoding, involving the combination and differentiation of different color code tracks, this inevitably introduces multi-color-domain mixing phenomena, affecting the precise subdivision and positioning of the code tracks. During measurement, this multi-color-domain mixing phenomenon is further exacerbated by external factors (such as vibration and temperature), greatly restricting the measurement accuracy of quantum dot optical encoders. However, as a novel displacement measurement encoder, there is currently no research on dynamic compensation for its measurement errors in the measurement field. Given the increasing demand for high-precision measurement, research on error compensation methods for quantum dot optical encoders under the influence of various internal and external factors is of great significance. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, system, and storage medium for dynamic compensation of measurement errors in quantum dot optical encoders to address the aforementioned technical problems.

[0006] This invention provides a method for dynamic compensation of measurement errors in a quantum dot optical encoder, comprising: S1: Obtain multi-color field code track images based on quantum dot optical encoder; S2: Construct a sub-pixel level compensation network model; S3: In the subpixel-level compensation network model, a code channel fine feature extraction network, a causal inference network, a decision network, a subpixel-level localization network, a code-decode network, and Q-learning are introduced to construct a single-factor measurement error compensation agent; the decision network parameters are obtained through the decision network. S4: In the single-factor measurement error compensation agent, a multi-sub-task model with parallel decomposition is trained. Combined with the output of the causal inference network, the proportion of compensation agent strategies for each influencing factor is calculated, thereby constructing a measurement error compensation agent under the influence of multiple factors. S5: The multi-color-gamut code track image and decision network parameters are dynamically compensated for measurement errors by a measurement error compensation agent under the influence of multiple factors.

[0007] Furthermore, in step S2, the subpixel-level compensation network model is the Unet++ network; the Unet++ network contains multiple downsampling layers, upsampling layers, and skip connections.

[0008] Furthermore, the subpixel-level compensation network model is input as a feature map of a multi-color domain code channel image. The feature map is processed by Unet++ through multiple downsampling, upsampling, and skip connections to determine the compensation direction. Based on this compensation direction, the compensation network is optimized, and the final output is an optimized image with subpixel precision compensation.

[0009] Furthermore, the information transmission process of the sub-pixel level compensation network model is as follows: The multi-color-domain code track image is input into the Unet++ network, and after passing through 5 downsampling layers of the backbone network, the result is obtained. , , , , The above The input to the previous layer is obtained through convolution, max pooling, and ReLU activation functions; the specific details of each... The relationship is as follows.

[0010]

[0011]

[0012] As shown in the formula above, the input is... Its dimensions are (1024, 1024, 3), and it produces five outputs through five downsampling layers. The dimensions are (1024, 1024, 16). The dimensions are (512, 512, 32). The dimensions are (256, 256, 64). The dimensions are (128, 128, 128). The dimensions are (64, 64, 256).

[0013] Correspondingly, the upsampling layer is constructed through upsampling, skip concatenation, and convolution. For example... It is by and The formula obtained through the above process is as follows:

[0014] Finally, through the above process, the compensated result is obtained. Its size is the same as the input, which is (1024, 1024, 3).

[0015] Correspondingly, the upsampling layer is constructed through upsampling, skip concatenation, and convolution. For example... It is by and The formula obtained through the above process is as follows:

[0016] Finally, through the above process, the compensated result is obtained. Its size is the same as the input, which is (1024, 1024, 3).

[0017] Furthermore, in step S3, the code channel fine feature extraction network, sub-pixel level localization network, and code encoding / decoding network are used to extract image features, locate code channels, and encode / decode code channels, respectively. The causal inference network is used to analyze the influence values ​​of various factors on the measurement from internal and external factor data. The decision network determines the error compensation direction based on the obtained influence values ​​and informs the compensation network through network parameters. Finally, Q-learning introduces an experience playback mechanism, which is beneficial for handling work with a small number of dataset samples.

[0018] Furthermore, in step S3, both the code track fine feature extraction network and the sub-pixel level localization network employ the H / S / V three-channel feature extraction method, with each channel being handled by the basic Unet network for feature extraction. Additionally, a conventional lightweight image detection network is used for the encoding and decoding network to implement the code track encoding and decoding process, thereby obtaining measurement values.

[0019] The causal inference network mainly consists of a text encoder, one-dimensional convolutions, two-dimensional convolutions, Transformer blocks, linear layers, and a cross-attention fusion module. The cross-attention fusion module's role is to fuse features from text and image data related to internal and external factors, directly obtaining a causal vector between these factors and the measured value. This vector provides a basis for subsequently evaluating the proportion of influence of different internal and external factors on the measured value. Furthermore, the decision network determines the direction of error compensation based on the aforementioned causal vector and transmits relevant information to the compensation network through network parameters. Finally, the introduction of Q-learning enables the agent to employ an experience replay mechanism, which is particularly advantageous for processing datasets with few samples and can further advance the compensation process, improving measurement accuracy.

[0020] Furthermore, in step S4, the measurement error compensation agent under the influence of multiple factors is composed of compensation agents under various influencing factors through weighting, where the weighting parameter is obtained through parallel decomposition. First, the internal and external factor data are analyzed through a causal inference network to obtain the causal vector between each factor and the measured value. This causal vector is combined with the output of the fine feature extraction module and input into the decision network. The output is used to train the parallel decomposition multi-subtask model and calculate the proportion value of the compensation agent strategy under various influencing factors. The proportion value is the proportion weight of the compensation agent under various factors. Finally, a measurement error compensation agent under the influence of multiple factors is formed to achieve decoupling and compensation for the influence of multiple interference factors at the same time. Finally, it is combined with sub-pixel level positioning and encoding / decoding network to obtain the final accurate measurement value.

[0021] Furthermore, in step S4, the specific compensation process is as follows:

[0022]

[0023] As shown in the formula above, This indicates the percentage of influence of each internal and external factor on the measured value, obtained through the causal inference network. This represents the central decision-making network. Indicates parallel decomposition. This represents the proportion of strategies employed by the compensating agent in response to various internal and external factors. The compensation agents corresponding to various internal and external factors are represented. Through weighted summation, the final error compensation agent for the coupled influence of multiple factors is obtained. The code track image is input into the compensating agent, and the resulting output image is then passed through a codec network to obtain the final accurate measurement value.

[0024] Furthermore, this invention provides a method for dynamic compensation of measurement errors in quantum dot optical encoders, including: Acquisition module: Acquires multi-color-domain code track images based on the quantum dot optical encoder; Model building module: Constructs subpixel level compensation network models; Single agent module: In the subpixel level compensation network model, a code channel fine feature extraction network, a causal inference network, a decision network, a subpixel level localization network, a code-decode network, and Q-learning are introduced to construct a single-factor measurement error compensation agent; the decision network parameters are obtained through the decision network. Multi-factor agent module: In the single-factor measurement error compensation agent, a multi-sub-task model with parallel decomposition is trained. Combined with the output of the causal inference network, the proportion of compensation agent strategies for each influencing factor is calculated, thereby constructing a measurement error compensation agent under the influence of multiple factors. Compensation module: The multi-color-gamut code track image and decision network parameters are dynamically compensated for measurement errors by a measurement error compensation agent under the influence of multiple factors.

[0025] Furthermore, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0026] In constructing a measurement error compensation agent under the influence of multiple factors, this invention uses a parallel decomposition multi-subtask model training method to decouple the coupled errors of internal and external factors. By analyzing the policy proportion of each compensation sub-agent, error compensation can be performed more effectively, thereby improving measurement accuracy. Attached Figure Description

[0027] Figure 1 This is a flowchart of a method for dynamic compensation of measurement errors in a quantum dot optical encoder in one embodiment; Figure 2 This is a schematic diagram of the subpixel-level compensation network model structure in one embodiment; Figure 3 This is a schematic diagram of a single-factor measurement error compensation agent structure in one embodiment. Figure 4 This is a schematic diagram of the causal inference network structure in one embodiment; Figure 5 This is a schematic diagram of a measurement error compensation agent structure under the influence of multiple factors in one embodiment. Figure 6 This is a block diagram of a quantum dot optical encoder measurement error dynamic compensation system in one embodiment; Detailed Implementation To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0028] Example 1: This embodiment provides, as follows: Figure 1 The method for dynamic compensation of measurement errors in quantum dot optical encoders, as shown, includes: S1: Obtain multi-color field code track images based on quantum dot optical encoder; S2: Construct a sub-pixel level compensation network model; S3: In the subpixel-level compensation network model, a code channel fine feature extraction network, a causal inference network, a decision network, a subpixel-level localization network, a code-decode network, and Q-learning are introduced to construct a single-factor measurement error compensation agent; the decision network parameters are obtained through the decision network. S4: In the single-factor measurement error compensation agent, a multi-sub-task model with parallel decomposition is trained. Combined with the output of the causal inference network, the proportion of compensation agent strategies for each influencing factor is calculated, thereby constructing a measurement error compensation agent under the influence of multiple factors. S5: The multi-color-gamut code track image and decision network parameters are dynamically compensated for measurement errors by a measurement error compensation agent under the influence of multiple factors.

[0029] In this embodiment, when constructing a measurement error compensation agent under the influence of multiple factors, a parallel decomposition multi-subtask model training method is used to decouple the coupling errors of internal and external factors. By analyzing the strategy proportion of each compensation sub-agent, error compensation can be performed more effectively, thereby improving measurement accuracy.

[0030] Example 2: This embodiment further discloses information based on Embodiment 1: Furthermore, in step S2, the subpixel-level compensation network model is the Unet++ network; the Unet++ network contains multiple downsampling layers, upsampling layers, and skip connections.

[0031] Furthermore, the subpixel-level compensation network model is input as a feature map of a multi-color domain code channel image. The feature map is processed by Unet++ through multiple downsampling, upsampling, and skip connections to determine the compensation direction. Based on this compensation direction, the compensation network is optimized, and the final output is an optimized image with subpixel precision compensation.

[0032] Furthermore, the information transmission process of the sub-pixel level compensation network model is as follows: The multi-color-domain code track image is input into the Unet++ network, and after passing through 5 downsampling layers of the backbone network, the result is obtained. , , , , The above The input to the previous layer is obtained through convolution, max pooling, and ReLU activation functions; the specific details of each... The relationship is as follows.

[0033]

[0034]

[0035] As shown in the formula above, the input is... Its dimensions are (1024, 1024, 3), and it produces five outputs through five downsampling layers. The dimensions are (1024, 1024, 16). The dimensions are (512, 512, 32). The dimensions are (256, 256, 64). The dimensions are (128, 128, 128). The dimensions are (64, 64, 256).

[0036] Correspondingly, the upsampling layer is constructed through upsampling, skip concatenation, and convolution. For example... It is by and The formula obtained through the above process is as follows:

[0037] Finally, through the above process, the compensated result is obtained. Its size is the same as the input, which is (1024, 1024, 3).

[0038] In step S3, a code channel fine feature extraction network, a causal inference network, a decision network, a sub-pixel level localization network, a code-decoding network, and Q-learning are introduced into the sub-pixel level compensation network model to construct a network such as... Figure 3 The single-factor measurement error compensation agent shown includes: The code channel fine feature extraction network, sub-pixel level localization network, and code encoding / decoding network are used to extract image features, locate code channels, and encode / decode code channels, respectively. The causal inference network is used to analyze the influence of various factors on the measurement from internal and external factor data. The decision network determines the error compensation direction based on the obtained influence values ​​and informs the compensation network through the network parameters. Finally, Q-learning introduces an experience playback mechanism, which is beneficial for handling tasks with a small number of dataset samples.

[0039] In step S3, both the code track fine feature extraction network and the sub-pixel level localization network employ the H / S / V three-channel feature extraction method, with each channel's feature extraction handled by the basic Unet network. Furthermore, the encoding / decoding network utilizes a conventional lightweight image detection network to implement the code track encoding and decoding process, thereby obtaining measurement values.

[0040] like Figure 4 The causal inference network shown mainly consists of a text encoder, one-dimensional convolution, two-dimensional convolution, a Transformer block, linear layers, and a cross-attention fusion module. The cross-attention fusion module's role is to fuse features from text and image data related to internal and external factors, directly obtaining a causal vector between these factors and the measured value. This vector provides a basis for subsequently evaluating the proportion of influence of different internal and external factors on the measured value. Furthermore, the decision network determines the direction of error compensation based on the aforementioned causal vector and transmits relevant information to the compensation network through network parameters. Finally, the introduction of Q-learning enables the agent to employ an experience replay mechanism, which is particularly advantageous for processing datasets with few samples and can further advance the compensation process, improving measurement accuracy.

[0041] In step S4, the measurement error compensation agent under multiple factors is constructed by weighting the compensation agents under various influencing factors. This weighting parameter is obtained through parallel decomposition. First, internal and external factor data are analyzed through a causal inference network to obtain the causal vector between each factor and the measured value. This causal vector is combined with the output of the fine feature extraction module and input into the decision network. The resulting output participates in the parallel decomposition multi-subtask model training, calculating the proportion of compensation agent strategies for various influencing factors. This proportion represents the weight of the compensation agent under each factor, ultimately constructing a... Figure 5The measurement error compensation agent shown is designed to compensate for multiple factors simultaneously, thereby decoupling and compensating for the influence of various interference factors. Finally, it combines sub-pixel-level positioning and a codec network to obtain the final accurate measurement value.

[0042] The specific compensation process is as follows:

[0043]

[0044] As shown in the formula above, This indicates the percentage of influence of each internal and external factor on the measured value, obtained through the causal inference network. This represents the central decision-making network. Indicates parallel decomposition. This represents the proportion of strategies employed by the compensating agent in response to various internal and external factors. The compensation agents corresponding to various internal and external factors are represented. Through weighted summation, the final error compensation agent for the coupled influence of multiple factors is obtained. The code track image is input into the compensating agent, and the resulting output image is then passed through a codec network to obtain the final accurate measurement value.

[0045] This embodiment addresses the problem of increased measurement error in quantum dot optical encoders due to the coupling effect of internal and external factors. Traditional compensation methods for optical encoders lack decoupling capabilities, and there is currently no research on error compensation specifically for quantum dot optical encoders. Therefore, this paper proposes a dynamic compensation method for measurement errors in quantum dot optical encoders for the first time. This method mainly includes three stages: constructing a sub-pixel-level compensation network model, constructing a single-factor measurement error compensation agent, and constructing a measurement error compensation agent under the influence of multiple factors. First, the sub-pixel-level compensation network model adopts the Unet++ network, with multi-color domain code track images and decision network parameters as inputs, and the output being the compensated and optimized image. Second, based on this model, a causal inference network, a decision network, a localization network, and a codec network are introduced to form a single-factor measurement error compensation agent. Furthermore, based on the above single-factor compensation agent, a parallel decomposition multi-subtask model is used for training, and the proportion of compensation agent strategies for each influencing factor is calculated by combining the output of the causal inference network. Finally, a measurement error compensation agent capable of coping with the coupling effect of multiple factors is constructed, achieving decoupling and compensation for multiple interference factors, thereby significantly improving measurement accuracy. When performing error compensation, a multimodal data system composed of textual data and measurement images representing both internal and external factors is used to train the compensation network. This method effectively avoids the loss of key information during the compensation process, thus significantly improving the compensation effect. When establishing the error compensation network, a causal inference network is used to analyze the impact of various influencing factors on the measurement error. This process can analyze the proportion of influence of various interference factors on the measurement, thereby formulating a more effective compensation strategy for the error task. When constructing the measurement error compensation agent under the influence of multiple factors, a parallel decomposition multi-subtask model training method is used to decouple the coupled errors of internal and external factors. By analyzing the strategy proportion of each compensation sub-agent, error compensation can be performed more specifically, thereby improving measurement accuracy.

[0046] Example 3: This embodiment provides, as follows: Figure 6 The quantum dot optical encoder measurement error dynamic compensation system shown includes: Acquisition module: Acquires multi-color-domain code track images based on the quantum dot optical encoder; Model building module: Constructs subpixel level compensation network models; Single agent module: In the subpixel level compensation network model, a code channel fine feature extraction network, a causal inference network, a decision network, a subpixel level localization network, a code-decode network, and Q-learning are introduced to construct a single-factor measurement error compensation agent; the decision network parameters are obtained through the decision network. Multi-factor agent module: In the single-factor measurement error compensation agent, a multi-sub-task model with parallel decomposition is trained. Combined with the output of the causal inference network, the proportion of compensation agent strategies for each influencing factor is calculated, thereby constructing a measurement error compensation agent under the influence of multiple factors. Compensation module: The multi-color-gamut code track image and decision network parameters are dynamically compensated for measurement errors by a measurement error compensation agent under the influence of multiple factors.

[0047] In this embodiment, when constructing a measurement error compensation agent under the influence of multiple factors, a parallel decomposition multi-subtask model training method is used to decouple the coupling errors of internal and external factors. By analyzing the strategy proportion of each compensation sub-agent, error compensation can be performed more effectively, thereby improving measurement accuracy.

[0048] Example 4: This embodiment provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.

[0049] In this embodiment, when constructing a measurement error compensation agent under the influence of multiple factors, a parallel decomposition multi-subtask model training method is used to decouple the coupling errors of internal and external factors. By analyzing the strategy proportion of each compensation sub-agent, error compensation can be performed more effectively, thereby improving measurement accuracy.

Claims

1. A method for dynamic compensation of measurement errors in a quantum dot optical encoder, characterized in that, include: S1: Obtain multi-color field code track images based on quantum dot optical encoder; S2: Construct a sub-pixel level compensation network model; S3: In the subpixel-level compensation network model, a code channel fine feature extraction network, a causal inference network, a decision network, a subpixel-level localization network, a code-decode network, and Q-learning are introduced to construct a single-factor measurement error compensation agent; the decision network parameters are obtained through the decision network. S4: In the single-factor measurement error compensation agent, a multi-sub-task model with parallel decomposition is trained. Combined with the output of the causal inference network, the proportion of compensation agent strategies for each influencing factor is calculated, thereby constructing a measurement error compensation agent under the influence of multiple factors. S5: The multi-color-gamut code track image and decision network parameters are dynamically compensated for measurement errors by a measurement error compensation agent under the influence of multiple factors.

2. The method for dynamic compensation of measurement errors of a quantum dot optical encoder according to claim 1, characterized in that, In step S2, the subpixel-level compensation network model is the Unet++ network; the Unet++ network contains multiple downsampling layers, upsampling layers, and skip connections.

3. The method for dynamic compensation of measurement errors of a quantum dot optical encoder according to claim 2, characterized in that, The subpixel-level compensation network model takes the feature map of a multi-color gamut code channel image as input. The feature map is processed by Unet++ through multiple downsampling, upsampling, and skip connections to determine the compensation direction. Based on this compensation direction, the compensation network is optimized, and the final output is an optimized image with subpixel precision compensation.

4. The method for dynamic compensation of measurement errors of a quantum dot optical encoder according to claim 2, characterized in that, The information transmission process of the sub-pixel level compensation network model is as follows: The multi-color-domain code track image is input into the Unet++ network, and after passing through 5 downsampling layers of the backbone network, the result is obtained. , , , , The above The input to the previous layer is obtained through convolution, max pooling, and ReLU activation functions; the specific details of each... The relationship is as follows: As shown in the formula above, the input is... Its dimensions are (1024, 1024, 3), and it produces five outputs through five downsampling layers. The dimensions are (1024, 1024, 16). The dimensions are (512, 512, 32). The dimensions are (256, 256, 64). The dimensions are (128, 128, 128). The dimensions are (64, 64, 256); Correspondingly, the upsampling layer is constructed through upsampling, skip concatenation, and convolution; such as It is by and The formula obtained through the above process is as follows: Finally, through the above process, the compensated result is obtained. Its size is the same as the input, which is (1024, 1024, 3).

5. The method for dynamic compensation of measurement errors of a quantum dot optical encoder according to claim 1, characterized in that, In step S3, the code channel fine feature extraction network, sub-pixel level localization network, and code encoding / decoding network are used to extract image features, locate code channels, and encode / decode code channels, respectively. The causal inference network is used to analyze the influence values ​​of various factors on the measurement from internal and external factor data. The decision network determines the error compensation direction based on the obtained influence values ​​and informs the compensation network through network parameters. Finally, Q-learning introduces an experience playback mechanism.

6. The method for dynamic compensation of measurement errors of a quantum dot optical encoder according to claim 1, characterized in that, In step S3, both the code channel fine feature extraction network and the sub-pixel level localization network adopt the H / S / V three-channel feature extraction method, and each channel is handled by the basic Unet network for feature extraction; in addition, the encoding and decoding network uses a conventional lightweight image detection network to realize the encoding and decoding process of the code channel, thereby obtaining the measurement value. The causal inference network consists of a text encoder, one-dimensional convolution, two-dimensional convolution, a Transformer block, a linear layer, and a cross-attention fusion module. The cross-attention fusion module is responsible for fusing features from text and image data related to internal and external factors, thereby directly obtaining a causal vector between internal and external factors and the measured value. This provides a basis for subsequently evaluating the proportion of influence of different internal and external factors on the measured value. In addition, the decision network determines the direction of error compensation based on the aforementioned causal vector and transmits relevant information to the compensation network through network parameters.

7. The method for dynamic compensation of measurement errors of a quantum dot optical encoder according to claim 1, characterized in that, In step S4, the measurement error compensation agent under the influence of multiple factors is composed of compensation agents under various influencing factors through weighting. The weighting parameter is obtained through parallel decomposition. First, the internal and external factor data are analyzed by the causal inference network to obtain the causal vector between each factor and the measurement value. This causal vector is combined with the output of the fine feature extraction module and input into the decision network. The output is used to train the parallel decomposition multi-subtask model and calculate the proportion value of the compensation agent strategy under various influencing factors. The proportion value is the proportion weight of the compensation agent under various factors. Finally, a measurement error compensation agent under the influence of multiple factors is formed to achieve decoupling and compensation for the influence of multiple interference factors at the same time. Finally, it is combined with sub-pixel level positioning and encoding / decoding network to obtain the final accurate measurement value.

8. The method for dynamic compensation of measurement errors of a quantum dot optical encoder according to claim 7, characterized in that, The specific compensation process is as follows: As shown in the formula above, This indicates the percentage of influence of each internal and external factor on the measured value, obtained through the causal inference network. This represents the central decision-making network. Indicates parallel decomposition. This represents the proportion of strategies employed by the compensating agent in response to various internal and external factors. The compensation agents corresponding to various internal and external factors are represented. Through weighted summation, the final error compensation agent for the coupled influence of multiple factors is obtained. The code track image is input into the compensation agent, and the resulting output image is passed through a code-decoder network to obtain the final accurate measurement value.

9. A dynamic compensation system for measurement errors of a quantum dot optical encoder, characterized in that, include: Acquisition module: Acquires multi-color-domain code track images based on the quantum dot optical encoder; Model building module: Constructs subpixel level compensation network models; Single agent module: In the subpixel level compensation network model, a code channel fine feature extraction network, a causal inference network, a decision network, a subpixel level localization network, a code-decode network, and Q-learning are introduced to construct a single-factor measurement error compensation agent; the decision network parameters are obtained through the decision network. Multi-factor agent module: In the single-factor measurement error compensation agent, a multi-sub-task model with parallel decomposition is trained. Combined with the output of the causal inference network, the proportion of compensation agent strategies for each influencing factor is calculated, thereby constructing a measurement error compensation agent under the influence of multiple factors. Compensation module: The multi-color-gamut code track image and decision network parameters are dynamically compensated for measurement errors by a measurement error compensation agent under the influence of multiple factors.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images