Wood-working machine visual inspection system based on AI

Abstract

The invention relates to the technical field of image analysis, in particular to an AI-based wood-working machine visual inspection system, which comprises a data generation module for extracting an original wood image, adjusting variables of a condition generation adversarial network according to a wood type and a processing mode, and circularly generating a wood seam image under a corresponding condition. According to the method, rich and diverse wood seam images are generated by using the conditional generative adversarial network, the problems of data insufficiency and sample singleness are effectively solved, the sample diversity of model training is expanded, and key visual features of wood are accurately captured in the process of converting image features into numerical vectors. The accuracy of classification and detection tasks is improved, the judgment accuracy and the system robustness are enhanced by integrating a plurality of prediction models and dynamically adjusting the weight according to the performance of each model, in addition, the system performance and the environment change are monitored in real time through a self-adaptive adjustment mechanism, model parameters are automatically optimized, and the system performance and the environment change are monitored in real time. And the detection precision and efficiency are continuously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image analysis, and in particular to an AI-based visual inspection system for woodworking machinery. Background Art

[0002] Image analysis is a branch of computer science that deals with techniques and methods for extracting useful information from images. This field uses various algorithms and statistical tools to interpret the visual data in images. Image analysis techniques can be divided into multiple stages, including preprocessing, feature extraction, detection, classification, and post-processing. These techniques are widely used in fields such as medical imaging, satellite remote sensing, industrial inspection, and security monitoring. In industrial applications, image analysis helps improve product quality, reduce errors, and improve production efficiency by automatically detecting defects on production lines.

[0003] Among them, the woodworking machinery visual inspection system refers to a system that uses visual inspection technology to monitor and evaluate the quality and integrity of wood during woodworking machinery processing. This system usually includes a camera, image processing software, and algorithms for performing inspection and analysis. Its uses include detecting wood defects, confirming the accuracy of cutting dimensions, and monitoring the efficiency of assembly lines to ensure product quality and optimization of production processes. Such systems play an important role in improving wood processing accuracy, reducing waste, and increasing production efficiency.

[0004] Existing technologies have shown some deficiencies in actual operation. Relying on limited, predefined data sets makes it ineffective when dealing with unknown wood types or new processing conditions, and may not be able to effectively identify and judge, resulting in reduced detection accuracy. Existing systems often lack flexible feature extraction and model optimization mechanisms, which limits their adaptability and optimization capabilities in changing production environments. These limitations may lead to reduced production efficiency, increased false detection rates, and unstable processing quality. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and propose an AI-based woodworking machine vision inspection system.

[0006] In order to achieve the above purpose, the present invention adopts the following technical solution: The AI-based woodworking machine visual inspection system includes:

[0007] The data generation module extracts the original wood image, adjusts the variables of the conditional generative adversarial network according to the wood type and processing method, and cyclically generates wood joint images under corresponding conditions, covering differentiated wood types and processing conditions, and integrates them into a joint image dataset;

[0008] The feature extraction module processes the seam image data set, analyzes the key texture and color features in the image, converts them into numerical vectors, integrates all vector data, and forms a seam feature vector library;

[0009] The integrated learning module inputs the seam feature vector library into multiple prediction models, obtains output results of the multiple prediction models, adjusts the weight according to the accuracy of each model, and performs weighted fusion on the output results of the multiple prediction models to form a collaborative determination result;

[0010] The adaptive adjustment module monitors the system performance and environmental changes based on the collaborative determination results, collects detection accuracy and environmental data, compares with preset standards, identifies performance differences, adjusts network input parameters and weight coefficients, and continuously optimizes seam detection accuracy.

[0011] As a further solution of the present invention, the step of acquiring the seam image dataset is specifically as follows:

[0012] Screening the original wood images, classifying them according to the differentiated wood types and processing methods, creating an initial index for each category, and generating a preliminary classification index;

[0013] Using the preliminary classification index, fine-tuning latent variables and generator parameters in a conditional generative adversarial network, simulating visual features of differentiated wood joints, and generating a set of simulated joint images;

[0014] Starting from the simulated seam image set, the formula is adopted:

[0015] G new =G prev α+ΔG (1-α)

[0016] Perform multiple rounds of iterative optimization, adjust the generator parameters, and optimize the visual naturalness of the image until each image meets the quality standards to generate a seam image dataset;

[0017] Among them, G new Represents the parameters of the new round of generators. The current parameters obtained after adjustment are used to generate the next round of images. G prev Represents the generator parameters of the previous round, which are the parameters obtained in the previous iteration and are used as the benchmark for the new round of parameter update. ΔG represents the amplitude of parameter adjustment in this round, which is the parameter change value calculated based on the optimization algorithm. α is the balance coefficient, which regulates the fusion ratio of new and old parameters.

[0018] As a further solution of the present invention, the step of acquiring the seam feature vector library is specifically as follows:

[0019] Analyze each image in the seam image data set, extract key texture and color features of the image, convert the features into numerical vectors through feature extraction, and generate a preliminary feature vector set;

[0020] Performing deep learning clustering processing on the preliminary feature vector set, grouping vectors of similar features to ensure that each group of vectors represents a type of texture or color pattern, and obtaining a clustered feature vector set;

[0021] Based on the clustered feature vector set, the formula is adopted:

[0022]

[0023] Perform weighted summation on the vectors in each category, synthesize representative feature vectors of multiple categories, and construct a joint feature vector library;

[0024] Where V represents the representative feature vector of the synthesis, v i is a single vector in the category, w i is the corresponding weight, representing the importance of the vector in the category, and n is the total number of vectors in the category.

[0025] As a further solution of the present invention, the step of obtaining the collaborative determination result is specifically:

[0026] Importing the seam feature vector library into multiple prediction models, running the prediction models to generate preliminary prediction results, each model outputting a set of results, and obtaining a multi-model preliminary prediction result set;

[0027] Analyze the preliminary prediction result set of the multiple models, use statistical analysis methods to evaluate the performance indicators of each model, determine a performance score for each model, and generate a model performance score table;

[0028] Combined with the model performance score table, the formula is used:

[0029]

[0030] The output results of multiple models are weighted and fused according to the performance scores to generate weighted fusion prediction results and form collaborative judgment results;

[0031] Among them, R represents the collaborative judgment result, r i is the output of the i-th model, w i is the weight of the model based on performance evaluation, and k is the total number of models involved in the calculation.

[0032] As a further solution of the present invention, the process of optimizing the seam detection accuracy is specifically as follows:

[0033] Monitor system performance and environmental changes, collect detection accuracy data from the collaborative determination results, and simultaneously capture environmental data to ensure real-time data updates and complete collection, and generate comprehensive monitoring data;

[0034] Comparing the comprehensive monitoring data with preset performance standards, identifying and quantifying performance deviations using a difference analysis method, and generating performance difference results;

[0035] According to the performance difference results, the input parameters and weight coefficients of the deep learning network are adjusted using the formula:

[0036] P new =P old +α(β·ΔP)

[0037] Adjust parameters to optimize seam detection accuracy;

[0038] Among them, P new is the adjusted parameter setting, P old is the original parameter setting, ΔP is the parameter adjustment calculated based on the performance difference, β adjusts the influence of the performance difference, and α controls the step size of the parameter adjustment.

[0039] Compared with the prior art, the advantages and positive effects of the present invention are:

[0040] In the present invention, a conditional generative adversarial network is used to generate rich and diverse wood joint images, which effectively solves the problems of insufficient data and sample uniformity and expands the sample diversity of model training. The process of converting image features into numerical vectors accurately captures the key visual features of wood and improves the accuracy of classification and detection tasks. By integrating multiple prediction models and dynamically adjusting the weights according to the performance of each model, the accuracy of judgment and the robustness of the system are enhanced. In addition, the adaptive adjustment mechanism realizes real-time monitoring of system performance and environmental changes, automatically optimizes model parameters, and continuously improves detection accuracy and efficiency. Combining these strategies, the system's adaptability and long-term operating performance are significantly improved, greatly enhancing quality control and efficiency in the production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a system flow chart of the present invention;

[0042] Figure 2 A flow chart of steps for obtaining a seam image data set of the present invention;

[0043] Figure 3 is a flow chart of the steps of obtaining the seam feature vector library of the present invention;

[0044] Figure 4 A flowchart of the steps for obtaining the collaborative determination result of the present invention;

[0045] Figure 5 A process flow chart for optimizing seam detection accuracy of the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0047] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0048] Embodiment 1:

[0049] See also Figure 1 , the AI-based woodworking machine vision inspection system includes:

[0050] The data generation module extracts the original wood image, adjusts the variables of the conditional generative adversarial network according to the wood type and processing method, and cyclically generates wood joint images under corresponding conditions, which are integrated into the joint image dataset;

[0051] The feature extraction module processes the seam image data set, analyzes the key texture and color features in the image, converts them into numerical vectors, and integrates all vector data to form a seam feature vector library;

[0052] The integrated learning module inputs the joint feature vector library into multiple prediction models, obtains the output results of multiple prediction models, adjusts the weight according to the accuracy of each model, and performs weighted fusion on the output results of multiple prediction models to form a collaborative judgment result;

[0053] The adaptive adjustment module monitors the system performance and environmental changes based on the collaborative judgment results, collects detection accuracy and environmental data, compares with preset standards, identifies performance differences, adjusts the network's input parameters and weight coefficients, and continuously optimizes seam detection accuracy.

[0054] The joint image data set specifically includes wood types and processing conditions; the joint feature vector library specifically includes texture features, color features, and numerical vectors; and the collaborative judgment results specifically include prediction model output, weight adjustment, and fusion results.

[0055] See also Figure 2 , the steps for obtaining the seam image dataset are as follows:

[0056] Screening the original wood images, classifying them according to the differentiated wood types and processing methods, creating an initial index for each category, and generating a preliminary classification index;

[0057] The original wood images are screened and classified, and matching classification indexes are created based on differentiated wood types and processing methods. In this process, image recognition technology is first used to identify wood features, including texture, color, and shape, and then they are classified based on the features. The classification process uses a machine learning algorithm to annotate images to ensure that images of each category have similar features. This method can significantly improve the accuracy and efficiency of classification and generate a preliminary classification index.

[0058] Using the preliminary classification index, we fine-tune the latent variables and generator parameters in the conditional generative adversarial network to simulate the visual characteristics of differentiated wood joints and generate a set of simulated joint images.

[0059] The generated preliminary classification index is used to fine-tune the latent variables and generator parameters of the conditional generative adversarial network (CGAN). By using the classification index, the parameters of the CGAN can be more accurately adjusted to match differentiated types of wood and processing methods, thereby generating a higher quality set of simulated seam images. This adjustment process includes setting the number of latent layer nodes in the network, adjusting weights and biases, so that the generated images are visually closer to the natural state of actual wood joints, improving the realism and wide application of the generated images, and generating a set of simulated seam images.

[0060] Starting from the simulated seam image set, the formula is used:

[0061] G new =G prev ·α+ΔG·(1-α)

[0062] Perform multiple rounds of iterative optimization, adjust the generator parameters, and optimize the visual naturalness of the image until each image meets the quality standards to generate a seam image dataset;

[0063] Among them, G new Represents the parameters of the new round of generators. The current parameters obtained after adjustment are used to generate the next round of images. G prevRepresents the generator parameters of the previous round, which are the parameters obtained in the previous iteration and are used as the benchmark for the new round of parameter update. ΔG represents the amplitude of parameter adjustment in this round, which is the parameter change value calculated based on the optimization algorithm. α is the balance coefficient, which regulates the fusion ratio of new and old parameters.

[0064] formula:

[0065] G new =G prev α+ΔG (1-α)

[0066] The benefit of this formula is that it effectively balances old and new parameters in this way, allowing the network to gradually adjust based on newly acquired data, thereby more accurately simulating the visual effect of wood joints.

[0067] Detailed explanation of the formula and the process of formula calculation and derivation:

[0068] Set the generator parameters G for initial use prev is 0.5, the learning rate α is set to 0.1, and the parameter adjustment ΔG for this round is 0.05. Calculated according to the formula:

[0069] G new =0.5 0.1 + 0.05 (1-0.1) = 0.05 + 0.045 = 0.095

[0070] The results show that: the current generator parameter G new After adjustment, it is 0.095, which means that the generator parameters have increased slightly. Such adjustment helps to generate more natural wood joint images, and the optimized joint image set is more in line with the quality standards.

[0071] See also Figure 3 , the specific steps for obtaining the seam feature vector library are:

[0072] Analyze each image in the seam image dataset, extract the key texture and color features of the image, convert the features into numerical vectors through feature extraction, and generate a preliminary feature vector set;

[0073] When analyzing the seam image dataset, the focus is on extracting key texture and color features from each image. This process locates and quantifies the various texture and color changes in the image through feature extraction algorithms. The feature data is converted into numerical vectors for data processing and analysis. In this way, the texture and color features extracted from the image are converted into structured data that can be used in machine learning models. This step is critical because it directly affects the efficiency and accuracy of subsequent steps. The resulting set of numerical vectors provides basic data for subsequent clustering and classification operations.

[0074] Perform deep learning clustering processing on the preliminary feature vector set, group vectors of similar features to ensure that each group of vectors represents a type of texture or color pattern, and obtain a clustered feature vector set;

[0075] Based on the preliminary feature vector set, the vectors are processed using a deep learning clustering algorithm. The clustering algorithm groups vectors by analyzing the degree of fit between them. This not only enhances the structuring of the data set, but also provides a method for identifying and classifying similar features in images. In addition, the clustering results help identify outliers or irregular patterns in the data, which is extremely critical for data analysis and utilization. Through this step, an effectively organized vector set is obtained, laying a solid foundation for building a feature vector library.

[0076] Based on the clustered feature vector set, the formula is used:

[0077]

[0078] Perform weighted summation on the vectors in each category, synthesize representative feature vectors of multiple categories, and construct a joint feature vector library;

[0079] Where V represents the representative feature vector of the synthesis, v i is a single vector in the category, w i is the corresponding weight, representing the importance of the vector in the category, and n is the total number of vectors in the category.

[0080] formula:

[0081]

[0082] The benefit of the formula is that by introducing a weight parameter, each vector can be weighted according to its contribution and importance in its category, thereby obtaining a representative vector for each category. This method enhances the accuracy and practicality of the vector library.

[0083] Detailed explanation of the formula and the process of formula calculation and derivation:

[0084] Suppose there is a set of vectors v1, v2, ..., v n and corresponding weights w1, w2, ..., w n ,and In order to calculate the weighted average to obtain the representative vector V, the following calculation is performed:

[0085] Take v1 = [1, 2], w1 = 0.5

[0086] Take v2 = [3, 4], w2 = 0.5

[0087] calculate:

[0088] V=w1·v1+w2·v2=0.5·[1,2]+0.5·[3,4]

[0089] =[0.5·1+0.5·3,0.5·2+0.5·4]=[2,3]

[0090] The results show that this method can synthesize representative feature vectors of multiple categories. The vectors not only retain the information of intra-class features, but also balance the influence of multiple vectors, making the generated feature vector library more accurate and practical.

[0091] See also Figure 4 , the specific steps for obtaining the collaborative determination results are:

[0092] Import the joint feature vector library into multiple prediction models, run the prediction models to generate preliminary prediction results, each model outputs a set of results, and obtains a set of preliminary prediction results of multiple models;

[0093] The seam feature vector library is input into multiple prediction models for processing. Each model performs data analysis and processing independently to obtain a preliminary set of prediction results. This process involves standardizing the data input to each model and adjusting the internal parameters of the model to match the characteristics of the input data. Through these steps, it can be ensured that each model can process the data in its optimal state, thereby obtaining reliable preliminary prediction results.

[0094] Analyze the preliminary prediction result set of multiple models, use statistical analysis methods to evaluate the performance indicators of each model, determine a performance score for each model, and generate a model performance score table;

[0095] Evaluate the output results of multiple prediction models. This step includes calculating the accuracy and reliability of each model, analyzing the performance indicators of each model through statistical methods, and then assigning a weight to each model based on the performance indicators. The weight reflects the credibility and influence of the model in the overall prediction process, ensuring that high-performance models occupy a larger proportion in the prediction, thereby improving the accuracy and reliability of the prediction results.

[0096] Combined with the model performance score table, the formula is used:

[0097]

[0098] The output results of multiple models are weighted and fused according to the performance scores to generate weighted fusion prediction results and form collaborative judgment results;

[0099] Among them, R represents the collaborative judgment result, r i is the output of the i-th model, w i is the weight of the model based on performance evaluation, and k is the total number of models involved in the calculation.

[0100] formula:

[0101]

[0102] The benefit of the formula is that it dynamically adjusts the influence of each model in the result according to its accuracy, which not only increases the flexibility of model fusion, but also significantly improves the accuracy of the overall prediction.

[0103] Detailed explanation of the formula and the process of formula calculation and derivation:

[0104] Suppose the output results of the three models are r1=0.85, r2=0.90, r3=0.78, and the corresponding weights are w1=0.5, w2=0.3, w3=0.2. The calculation process is as follows:

[0105] R=(0.5×0.85)+(0.3×0.90)+(0.2×0.78)

[0106] R=0.425+0.27+0.156

[0107] R=0.851

[0108] The results show that after combining the prediction results of the three models, the weighted result is 0.851, which reflects that the combination of multiple models can provide more robust prediction outputs than a single model and help improve the accuracy of decision-making.

[0109] See also Figure 5 ,The process of optimizing seam detection accuracy is as follows:

[0110] Monitor system performance and environmental changes, collect detection accuracy data from collaborative judgment results, and simultaneously capture environmental data to ensure real-time data updates and complete collection, and generate comprehensive monitoring data;

[0111] Monitor system performance and environmental changes, collect data regularly to ensure the effective operation of the monitoring system. Real-time data capture not only helps to understand the current system performance, but also helps to respond to any changes in the environment. The instant update and integrity of the data are carried out through high-performance servers and database systems to ensure that each data capture is correct. In addition, data processing technology is used to pre-process the collected data, including noise removal, data normalization and outlier processing, to ensure that the data quality meets the requirements of subsequent analysis. In this way, the system can better match environmental changes and optimize detection accuracy.

[0112] Compare the comprehensive monitoring data with the preset performance standards, use the difference analysis method to identify and quantify the performance deviation, and generate the performance difference results;

[0113] The comparison between performance monitoring data and preset standards is specially designed to process batch data and extract useful information. Performance differences are identified through quantitative methods, including statistical analysis, data trend tracking and pattern recognition. Using the analysis results, the system can automatically generate performance results. The results provide a scientific basis for decision-making and ensure that each adjustment is based on accurate data analysis, thereby improving the overall performance and reliability of the system.

[0114] According to the performance difference results, the input parameters and weight coefficients of the deep learning network are adjusted using the formula:

[0115] P new =P old +α(β·ΔP)

[0116] Adjust parameters to optimize seam detection accuracy;

[0117] Among them, P new is the adjusted parameter setting, P old is the original parameter setting, ΔP is the parameter adjustment calculated based on the performance difference, β adjusts the influence of the performance difference, and α controls the step size of the parameter adjustment.

[0118] formula:

[0119] P new =P old +α(β·ΔP)

[0120] The benefit of the formula is that the system can flexibly adjust according to the problems encountered in actual operation, and the gradual optimization process ensures that the system adaptability and accuracy are cyclically improved. This formula makes parameter adjustment not only based on a single feedback, but also a comprehensive consideration of multiple factors, which can make more detailed adjustments to performance differences.

[0121] Detailed explanation of the formula and the process of formula calculation and derivation:

[0122] Set the original parameter P old is 0.5, the adjustment amount calculated according to the performance difference ΔP is 0.05, the sensitivity adjustment coefficient β is set to 1.5, and the learning rate α is 0.1, the formula becomes P new =0.5+0.1×(1.5×0.05).

[0123] The calculation process is:

[0124] Calculate the ΔP adjustment: 1.5 × 0.05 = 0.075

[0125] Applied learning rate: 0.1×0.075=0.0075

[0126] Update parameter: 0.5 + 0.0075 = 0.5075

[0127] The results show that the current parameter setting P new It is 0.5075, indicating that after adjustment, the system parameters are slightly improved, which shows that fine-tuning based on actual performance feedback can gradually improve system performance.

[0128] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. AI-based woodworking machine vision inspection system, characterized by: The system comprises: The data generation module is based on the original wood image, and according to the wood type and processing method, adjusts the variables of the conditional generative adversarial network, cyclically generates wood joint images under corresponding conditions, and integrates them into a joint image dataset; The feature extraction module processes the seam image data set, analyzes the key texture and color features in the image, converts them into numerical vectors, integrates all vector data, and forms a seam feature vector library; The integrated learning module inputs the seam feature vector library into multiple prediction models, obtains output results of the multiple prediction models, adjusts the weight according to the accuracy of each model, and performs weighted fusion on the output results of the multiple prediction models to form a collaborative determination result; The adaptive adjustment module monitors the system performance and environmental changes based on the collaborative determination results, collects detection accuracy and environmental data, compares with preset standards, identifies performance differences, adjusts network input parameters and weight coefficients, and continuously optimizes seam detection accuracy.

2. The AI-based woodworking machine vision inspection system according to claim 1 is characterized in that: The steps for obtaining the seam image dataset are specifically as follows: Screening the original wood images, classifying them according to the differentiated wood types and processing methods, creating an initial index for each category, and generating a preliminary classification index; Using the preliminary classification index, fine-tuning latent variables and generator parameters in a conditional generative adversarial network, simulating visual features of differentiated wood joints, and generating a set of simulated joint images; Starting from the simulated seam image set, the formula is adopted: G new =G prev ·α+ΔG·(1-α) Perform multiple rounds of iterative optimization, adjust the generator parameters, and optimize the visual naturalness of the image until each image meets the quality standards to generate a seam image dataset; Among them, G new Represents the parameters of the new round of generators. The current parameters obtained after adjustment are used to generate the next round of images. G prev Represents the generator parameters of the previous round, which are the parameters obtained in the previous iteration and are used as the benchmark for the new round of parameter update. ΔG represents the amplitude of parameter adjustment in this round, which is the parameter change value calculated based on the optimization algorithm. α is the balance coefficient, which regulates the fusion ratio of new and old parameters.

3. The AI-based woodworking machine vision inspection system according to claim 2 is characterized in that: The steps for obtaining the seam feature vector library are specifically as follows: Analyze each image in the seam image data set, extract key texture and color features of the image, convert the features into numerical vectors through feature extraction, and generate a preliminary feature vector set; Performing deep learning clustering processing on the preliminary feature vector set, grouping vectors of similar features to ensure that each group of vectors represents a type of texture or color pattern, and obtaining a clustered feature vector set; Based on the clustered feature vector set, the formula is adopted: Perform weighted summation on the vectors in each category, synthesize representative feature vectors of multiple categories, and construct a joint feature vector library; Where V represents the representative feature vector of the synthesis, v i is a single vector in the category, w i is the corresponding weight, representing the importance of the vector in the category, and n is the total number of vectors in the category.

4. The AI-based woodworking machine vision inspection system according to claim 3 is characterized in that: The steps for obtaining the collaborative determination result are specifically as follows: Importing the seam feature vector library into multiple prediction models, running the prediction models to generate preliminary prediction results, each model outputting a set of results, and obtaining a multi-model preliminary prediction result set; Analyze the preliminary prediction result set of the multiple models, use statistical analysis methods to evaluate the performance indicators of each model, determine a performance score for each model, and generate a model performance score table; Combined with the model performance score table, the formula is used: The output results of multiple models are weighted and fused according to the performance scores to generate weighted fusion prediction results and form collaborative judgment results; Among them, R represents the collaborative judgment result, r i is the output of the i-th model, w i is the weight of the model based on performance evaluation, and k is the total number of models involved in the calculation.

5. The AI-based woodworking machine vision inspection system according to claim 4 is characterized in that: The process of optimizing seam detection accuracy is specifically as follows: Monitor system performance and environmental changes, collect detection accuracy data from the collaborative determination results, and simultaneously capture environmental data to ensure real-time data updates and complete collection, and generate comprehensive monitoring data; Comparing the comprehensive monitoring data with preset performance standards, identifying and quantifying performance deviations using a difference analysis method, and generating performance difference results; According to the performance difference results, the input parameters and weight coefficients of the deep learning network are adjusted using the formula: P new =P old +α(β·ΔP) Adjust parameters to optimize seam detection accuracy; Among them, P new is the adjusted parameter setting, P old is the original parameter setting, ΔP is the parameter adjustment calculated based on the performance difference, β adjusts the influence of the performance difference, and α controls the step size of the parameter adjustment.

Images