Automobile headrest sorting method and system based on confidence matching

Abstract

The invention relates to the technical field of intelligent sorting, in particular to an automobile headrest sorting method and system based on confidence matching, and the method comprises the steps: collecting a target headrest image, and carrying out the preprocessing of the target headrest image; performing multi-dimensional physical feature extraction on the preprocessed target headrest image to generate a physical feature vector of the target headrest; calculating a confidence coefficient between the physical feature vector of the target headrest and each template feature vector in each preset headrest type feature library; the type of the target headrest is judged based on the calculated confidence degree set of all the headrest types and an introduced judgment threshold value; and executing corresponding operation according to the judgment result of the target headrest. According to the method and the device, the identification precision and stability can be improved, and automatic and extensible headrest type identification without labels is ensured to be realized in a flexible production environment.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent sorting, and particularly to a method and system for sorting automotive headrests based on confidence matching. Background Art

[0002] With the continuous development of the automotive manufacturing and aftermarket industries, the types, materials, and assembly methods of automotive parts have become increasingly diverse. Especially in the field of intelligent sorting and logistics automation, how to quickly and accurately identify parts with similar structures but different attributes has become an important research direction in intelligent factories. During the automotive manufacturing and parts assembly processes, as an important part of the seat system, the types, materials, and installation methods of headrests show diversity due to factors such as automotive energy types, vehicle configurations, and interior styles. Although the headrests used in different vehicle models may be similar in overall structure, each headrest usually has certain physical feature differences due to differences in color, material, position, and matching vehicle models. These differences are reflected in aspects such as visual appearance, structural form, and material texture, objectively forming the inherent and distinguishable physical features of each type of headrest.

[0003] In existing automotive production lines and parts logistics systems, the sorting of automotive headrests is mainly achieved through the following methods: One is to attach barcodes, labels, or RFID chips to the headrests and read the additional information through dedicated scanning devices to achieve identification and sorting; the other is to use an image recognition system to recognize headrest images, but mostly rely on predefined feature templates for image comparison.

[0004] However, with the increasing richness of vehicle models and the improvement of user customization requirements, the differences in the types, materials, and colors of automotive headrests are constantly increasing, making the identification task in the sorting link of headrests increasingly complex. The existing methods based on label recognition or image template matching often rely on external identifiers or rule settings for limited features, and it is difficult to ensure the accuracy and stability of sorting in scenarios with diverse types, subtle differences, label shedding, or image changes. Especially in production lines lacking manual participation or with a high degree of automation, the existing recognition means often struggle to adapt to the rapidly changing types and attributes of headrests. This limitation directly leads to an increase in the recognition error rate and a decrease in sorting efficiency, thereby affecting the continuity and reliability of the overall assembly process. Summary of the Invention

[0005] The present application provides a method and system for sorting automotive headrests based on confidence matching, which can improve the recognition accuracy and stability, and ensure label-free, automated, and scalable headrest type recognition in a flexible production environment. The present application provides the following technical solutions:

[0006] In a first aspect, the present application provides a method for sorting automotive headrests based on confidence matching, the method comprising:

[0007] Collect the target headrest image and preprocess the target headrest image;

[0008] Extract multi-dimensional physical features from the preprocessed target headrest image to generate a physical feature vector of the target headrest;

[0009] Calculate the confidence of the physical feature vector of the target headrest and each template feature vector in the preset headrest type feature libraries;

[0010] Judge the type of the target headrest based on the set of confidences of all headrest types obtained by calculation and the introduced decision threshold;

[0011] Execute corresponding operations according to the judgment result of the target headrest.

[0012] In a specific feasible implementation, the preprocessing of the target headrest image includes:

[0013] Normalize the resolution of the image and uniformly scale the images from different devices or different shooting distances to a predetermined size;

[0014] Suppress the noise of the image, calibrate the color through automatic white balance or gamma correction, and combine threshold segmentation, edge detection or depth information masking methods to accurately segment and crop the headrest area in the image, removing the background and irrelevant areas.

[0015] In a specific feasible implementation, the extraction of multi-dimensional physical features from the preprocessed target headrest image to generate a physical feature vector of the target headrest includes:

[0016] Perform edge detection on the target headrest image, remove noise points and close the edges through morphological dilation and erosion operations, and call the contour extraction algorithm to identify the main contour curve of the headrest;

[0017] Calculate geometric descriptors based on the identified contour curve, convert the headrest area into a grayscale image, and at the same time extract local texture microstructures using local binary patterns;

[0018] Extract color features from the color image, splice the extracted geometric, texture and color features into a high-dimensional feature vector in a predetermined order, and then normalize the feature values of each dimension to generate a physical feature vector of the target headrest.

[0019] In a specific feasible implementation, the calculation of the confidence of the physical feature vector of the target headrest and each template feature vector in the preset headrest type feature libraries includes:

[0020] In the pre - established headrest type feature library, each template feature vector is obtained by the same physical feature extraction method. The physical feature vector v of the target headrest = [v1, v2, …, v n has the same dimension as each type of template feature vector t j = [t j,1 , t j,2 , …, t j,n , where j = 1, 2, …, N, and N represents the number of known headrest types in the library;

[0021] The weighted Gaussian kernel similarity formula is used to calculate the confidence between the target headrest and each template. The calculation formula is as follows:

[0022]

[0023] where β i represents the importance weight of the i - th dimensional feature, δ j,i is the standard deviation of the template feature vector on the i - th dimensional feature, and ε is an extremely small positive number.

[0024] In a specific feasible implementation, determining the type of the target headrest based on the calculated confidence set of all headrest types and the introduced decision threshold includes:

[0025] Obtain the confidence set {C1, C2, …, C N} of all headrest types calculated, and sort it from large to small to obtain the maximum confidence value C (1) , the corresponding type index is denoted as p, and the second - largest confidence value C (2) , the corresponding type index is denoted as q. Introduce two sets of decision thresholds, the basic matching threshold T j and the ratio threshold M j ;

[0026] When the maximum confidence C (1) reaches or exceeds the basic matching threshold T p of type p, and the ratio of the sub - optimal confidence to the optimal confidence R = C (2) / C (1) is less than or equal to the ratio threshold M p , classify the headrest as type p;

[0027] If C (1) ≥T p , but the ratio of the sub - optimal confidence to the optimal confidence R exceeds the ratio threshold M p , mark the sample as a fuzzy type;

[0028] When the maximum confidence C (1) is lower than the corresponding threshold T pWhen it is, mark it as an unknown type.

[0029] In a specific feasible implementation, the determination of the type of the target headrest based on the set of confidences of all headrest types obtained by calculation and the introduced determination threshold further includes:

[0030] Design an online threshold adaptive update strategy, taking the basic matching threshold T of type j as an example. j Let the current time be the threshold before the t-th update, denoted as T. j (t) When a certain classification is verified as correct, take the actual confidence of this time. to update the threshold:

[0031]

[0032] where the attenuation factor λ is used to balance historical experience and the latest observation.

[0033] In a specific feasible implementation, the execution of corresponding operations according to the judgment result of the target headrest includes:

[0034] If it is determined to be a definite type, directly classify the target headrest into the definite type and push it to the subsequent business process;

[0035] If it is determined to be a fuzzy type, send it to the manual review system;

[0036] If it is determined to be an unknown type, trigger the process of expanding the database, prompting the technical staff to perform manual modeling, feature annotation on this sample, or incorporate it into the feature database update process.

[0037] In a second aspect, the present application provides a vehicle headrest sorting system based on confidence matching, adopting the following technical solution:

[0038] A vehicle headrest sorting system based on confidence matching, comprising:

[0039] An image acquisition module, configured to acquire a target headrest image and preprocess the target headrest image;

[0040] A feature extraction module, configured to extract multi-dimensional physical features from the preprocessed target headrest image to generate a physical feature vector of the target headrest;

[0041] A confidence calculation module, configured to calculate the confidence between the physical feature vector of the target headrest and each template feature vector in the preset feature library of each headrest type;

[0042] A type judgment module, configured to judge the type of the target headrest based on the set of confidences of all headrest types obtained by calculation and the introduced determination threshold;

[0043] A result execution module, configured to perform corresponding operations according to the judgment result of the target headrest.

[0044] In a third aspect, the present application provides an electronic device, which includes a processor and a memory; a program is stored in the memory, and the program is loaded and executed by the processor to implement a method for sorting automotive headrests based on confidence matching as described in the first aspect.

[0045] In a fourth aspect, the present application provides a computer-readable storage medium, in which a program is stored, and when the program is executed by a processor, it is used to implement a method for sorting automotive headrests based on confidence matching as described in the first aspect.

[0046] In summary, the beneficial effects of the present application at least include:

[0047] (1) Through multi-angle image acquisition and image standardization preprocessing technology, the accuracy and robustness of headrest recognition are significantly improved. By standardizing the acquired images, the influence of lighting changes and shooting angles on the image quality is eliminated, ensuring the consistency of the input data. In addition, by combining the extraction of multiple feature dimensions such as shape, texture, and color, the model can more comprehensively understand and identify different types of headrests, thereby enhancing the ability to distinguish headrest types in complex scenarios and reducing the probability of misrecognition.

[0048] (2) The weighted Gaussian kernel similarity algorithm and the maximum confidence screening strategy introduced in the present application further improve the stability and accuracy of recognition. By assigning reasonable weights to different features, the algorithm can adjust according to the contribution degree of each feature to the classification result, thereby achieving higher accuracy in the classification tasks of different types of headrests. At the same time, the maximum confidence screening mechanism further reduces the possibility of classification errors, ensuring that the system has high reliability in practical applications. Especially in the automated production process, accurate and efficient headrest type recognition is achieved.

[0049] By collecting the target headrest images and performing preprocessing, multi-dimensional physical features of the headrest are extracted from the preprocessed images, including contour shape parameters, gray-level co-occurrence matrix, local binary pattern, color histogram, color moments, etc., and stitched to generate a normalized high-dimensional feature vector. The feature vector of the target headrest is compared with each template vector in the preset feature library, and the weighted Gaussian kernel similarity formula is used to combine standard deviation normalization and feature importance weights to calculate the confidence between the target headrest and each known type. Based on the confidence results of all types, the confidence results of each type are input into the decision module, and the type with the highest confidence is selected as the recognition result; a threshold can be set to judge whether the confidence is reliable, and when necessary, enter the review process. The recognition result is output to the execution system to control the subsequent headrest sorting action. This application performs multi-angle image acquisition, image standardization preprocessing, combines multi-dimensional feature extraction such as shape, texture, and color, and uses the weighted Gaussian kernel similarity algorithm to accurately classify and identify the headrest. By introducing the feature weight mechanism and the maximum confidence screening, the recognition accuracy and stability are improved, ensuring label-free, automated, and scalable headrest type recognition in a flexible production environment, thus effectively solving the problems of low accuracy, difficult type differentiation, and excessive manual intervention.

[0050] The above description is only an overview of the technical solution of this application. In order to understand the technical means of this application more clearly and implement it according to the content of the specification, the following takes the preferred embodiments of this application and combines the accompanying drawings to describe in detail as follows. Brief Description of the Drawings

[0051] Figure 1 is a schematic flowchart of the method for sorting automotive headrests based on confidence matching in an embodiment of this application.

[0052] Figure 2 is a structural block diagram of the system for sorting automotive headrests based on confidence matching in an embodiment of this application.

[0053] Figure 3 is a block diagram of an electronic device for sorting automotive headrests based on confidence matching in an embodiment of this application. Detailed Embodiments

[0054] The following combines the accompanying drawings and embodiments to further describe the detailed embodiments of this application in detail. The following embodiments are used to illustrate this application, but are not used to limit the scope of this application.

[0055] Optionally, this application takes the method for sorting automotive headrests based on confidence matching provided in each embodiment as an example for illustration in an electronic device. The electronic device is a terminal or a server, and the terminal can be a mobile phone, a computer, a tablet computer, etc. The type of the electronic device is not limited in this embodiment.

[0056] Refer to Figure 1, which is a schematic flowchart of a car headrest sorting method based on confidence matching provided by an embodiment of this application. This method at least includes the following steps:

[0057] Step S101: Collect the target headrest image and preprocess the target headrest image.

[0058] In step S101, first, use a pre-calibrated industrial camera to collect images of the target headrests to be sorted. To ensure the collection effect, a high-resolution RGB camera or an RGB-D camera with depth perception ability can be selected for the industrial camera, and it is fixedly installed on both sides of the conveyor line or in a top-down position to obtain the surface information of the headrests from multiple angles without blind spots.

[0059] In implementation, the exposure time, aperture size, and lighting conditions of the camera can be adjusted automatically or manually according to the ambient light in the workshop to ensure that the captured images are clear and there is no overexposure or underexposure phenomenon. In addition, to prevent incomplete shooting caused by the position offset of the headrest during transportation, a position detection threshold can be set in the image trigger module, and the shooting action is triggered only when the headrest completely enters the preset shooting area, so as to obtain high-quality original image data.

[0060] Subsequently, the collected original image data will enter the preprocessing process to improve the robustness and accuracy of subsequent feature extraction. In step S101, the preprocessing includes normalizing the resolution of the image, uniformly scaling images from different devices or different shooting distances to a predetermined size; using algorithms such as median filtering or bilateral filtering to suppress noise in the image to remove sensor noise and environmental interference; calibrating the color through automatic white balance or gamma correction to eliminate color cast caused by differences in light source color temperature; and combining methods such as threshold segmentation, edge detection, or depth information masking to accurately segment and crop the headrest area in the image, removing the background and irrelevant areas, and finally outputting a standardized image with consistent size, low noise, true color, and only containing the headrest target, laying a solid foundation for subsequent physical feature extraction.

[0061] Step S102: Extract multi-dimensional physical features from the preprocessed target headrest image to generate a physical feature vector of the target headrest.

[0062] In step S102, extract physical features from the preprocessed target headrest image to obtain multi-dimensional features such as shape, texture, and color that can distinguish different headrests.

[0063] Specifically, the Canny or Sobel operator is first used to perform edge detection on the target headrest image, and morphological dilation and erosion operations are used to remove noise and close edges. Then, the contour extraction algorithm is called to identify the main contour curve of the headrest. Based on the identified contour curve, geometric descriptors are calculated, including indicators such as headrest area, perimeter, aspect ratio, roundness, and invariant moments, to characterize the overall shape characteristics. Next, the headrest area is converted into a grayscale image, and texture features such as contrast, correlation, energy, and homogeneity are calculated through the grayscale co-occurrence matrix. At the same time, the local binary pattern is used to extract the local texture microstructure to enhance the sensitivity to surface detail differences.

[0064] In the same implementation, color features are also extracted from color images: the image is kept in RGB space or mapped to HSV space, and the color histogram and first-order and second-order color moments of each channel are counted to reflect the color distribution differences caused by different materials or dyeing processes. After all manually extracted geometric, texture and color features are spliced ​​into high-dimensional feature vectors in a predetermined order, the feature values ​​of each dimension are normalized to eliminate dimensional differences and improve matching stability, and finally a unified physical feature vector of the target headrest with good discrimination is generated, providing reliable input for subsequent confidence calculation and type matching.

[0065] Step S103: Calculate the confidence between the physical feature vector of the target headrest and each template feature vector in the preset feature library of each headrest type.

[0066] In step S103, the physical feature vector of the target headrest extracted in step S102 is compared with the pre-established feature library of each headrest type, focusing on calculating the confidence between each headrest type and the target image. In the pre-established headrest type feature library, each template feature vector is obtained by the same physical feature extraction method, ensuring that the physical feature vector v of the target headrest is [v1, v2, …, v n ] and each type of template feature vector t j =[t j,1 ,t j,2 ,…,t j,n ] have the same dimensions, where j = 1, 2, ..., N, where N represents the number of known headrest types in the library. This number can be set dynamically according to actual needs and is not required to be fixed. At the same time, for the physical characteristics of each dimension, the standard deviation σ of this type is obtained in advance through statistics of a large number of sample data j,i (i=1,2,…,n) so as to reflect the discrete degree of each feature in the calculation.

[0067] In the implementation, an improved weighted Gaussian kernel similarity formula is used to calculate the confidence between the target headrest and each template. The calculation formula is as follows:

[0068]

[0069] Among them, β i represents the importance weight of the i-th dimensional feature, which can usually be preset according to the contribution degree of the feature when differentiating different headrest types; σ j,i is the standard deviation of the template feature vector on the i-th dimensional feature, used to measure the volatility of this feature within the same type; ε is an extremely small positive number, used to avoid division-by-zero errors. The design of this formula is based on the principle of Gaussian kernel. The differences between the target and the template in each dimension are mapped to the interval from 0 to 1 after being normalized by the weighted sum of squares. The closer the value is to 1, the more similar the target headrest is to this template in physical features, and vice versa, the lower the similarity. By calculating the confidence C j .

[0070] In the above formula, this formula divides the feature difference of each dimension by the corresponding template standard deviation, realizing the normalization process based on statistical distribution. In implementation, this means automatically reducing the impact of features with large fluctuations within the same type on the overall similarity, while giving greater weights to features with small fluctuations and high discrimination. Compared with the ordinary Euclidean distance that cannot distinguish the fluctuation ranges of each dimension, this design effectively suppresses the interference of large noise or outliers on the confidence calculation, improving the stability and reliability. The introduction of the feature importance weight can be preset or adjusted online according to the sensitivity of different headrest types during differentiation or business experience. In implementation, designers can assign higher weights to some physical features (such as texture details or color distribution) that are particularly critical for type differentiation, making these features dominant in the similarity calculation; while reducing the weights of secondary features to avoid the interference of redundant information. Ordinary formulas often treat all features equally and lack this flexible and controllable ability.

[0071] In addition, mapping the weighted difference to the interval from 0 to 1 through an exponential function realizes a sensitive response to small differences and a rapid decay to large differences. In implementation, this smooth similarity curve makes the confidence rapidly approach 1 when approaching the template, and rapidly decrease when the difference is obvious, facilitating the subsequent setting of a unified and easily adjustable matching threshold. Compared with linear distance metrics or threshold segmentation, exponential mapping can reduce the classification instability caused by "critical jitter".

[0072] Step S104: Determine the type of the target headrest based on the set of confidences of all headrest types calculated and the introduced determination threshold.

[0073] In step S104, first obtain the set of confidences of all headrest types {C1, C2, …, C N} calculated in step S103, and sort them from large to small to obtain the maximum confidence value C(1) , the corresponding type index is denoted as p, and the second-largest confidence value is C (2) , the corresponding type index is denoted as q. At this time, two key pieces of information are known: one is the confidence C of the "most likely" type p (1) , and the other is the confidence C of its closest competing type q (2) . In order to ensure a sufficient confidence level during classification and avoid making hasty conclusions when the confidences of the two types are too close, two sets of decision thresholds are introduced in this step, the basic matching threshold T j and the ratio threshold M j , which are preset or dynamically updated for each type of headrest respectively

[0074] In implementation, the decision logic is divided into three cases: First, when the maximum confidence C (1) not only reaches or exceeds the basic matching threshold T of type p p , and at this time the ratio of the sub-optimal confidence to the optimal confidence R = C (2) / C (1) is less than or equal to the ratio threshold M p , it indicates that the target headrest has an obvious advantage in physical characteristics over type p and the gap from any other type is large enough. At this time, the headrest is directly classified as type p without further manual intervention or supplementary verification

[0075] Second, if C (1) ≥ T p , but the ratio of the sub-optimal confidence to the optimal confidence R exceeds the ratio threshold M p , it indicates that although the confidence of the target headrest in the optimal type p is high, the confidence difference from the sub-optimal type q is not large, and there is potential classification ambiguity. To avoid misclassification, the sample is marked as "ambiguous type" and the manual review process is triggered to obtain more information for the final decision

[0076] Finally, when the maximum confidence C (1) is lower than the corresponding threshold T p , it indicates that the similarity of the target headrest to any known type is not sufficient to meet the most basic confidence level, and it is marked as "unknown type". This mark can be used to prompt manual inspection, supplement new templates or expand the feature library to ensure that new models or abnormal parts can be included in the recognition system in a timely manner

[0077] In addition, preferably, in order to enable the decision-making mechanism to be continuously optimized according to the actual operation situation of the production site, an online threshold adaptive update strategy is designed. Taking the basic matching threshold T of type j j as an example, assume that the current time is the t-th update, and the threshold before the update is T j (t) . When a certain classification is verified to be correct, the actual confidence of this time can be taken (i.e., C when the classification is correct j ) to update the threshold:

[0078]

[0079] Among them, the attenuation factor λ is used to balance historical experience and the latest observation. The larger the value, the smoother the update; the smaller the value, the more sensitive the response to new data. Similarly, the ratio threshold can also be updated in the same way. Through this mechanism, during continuous operation, it can learn the subtle changes in the on-site data distribution, automatically adjust the judgment criteria, which not only reduces the manual calibration cost but also ensures stable recognition performance in scenarios where new and old models are mixed or there are large batch differences.

[0080] In the above design, the dynamic threshold judgment of the confidence ratio is derived from the boundary learning theory in the fuzzy classification problem. Traditional multi-class classification algorithms often have difficulty making a robust decision when faced with two highly similar samples, while the ratio judgment provides a more detailed perspective: if the first-highest and second-highest confidences are very close (the ratio is close to 1), it indicates that the system may be in a "fuzzy boundary". Dynamically update the threshold of this ratio to let the system define by itself "how close is considered fuzzy" to achieve more personalized judgment boundary learning. At the same time, make the threshold be able to "think" instead of being fixed. Through the memory of each successful recognition, let the system continuously optimize its judgment ability in a dynamic environment.

[0081] Step S105: Execute corresponding operations according to the judgment result of the target headrest.

[0082] In step S105, according to the classification result of the target headrest type determined in step S104, execute a differentiated processing strategy to ensure the recognition accuracy and the continuous evolution ability of the recognition system. The classification result includes three situations: determined type, fuzzy type, and unknown type, and the following responses are made respectively:

[0083] First of all, if it is determined to be the determined type, it means that the physical characteristics of the target headrest are highly consistent with the determined type, and there is sufficient confidence for automated processing. At this time, directly classify the target headrest into the determined type and push it to the subsequent business processes, including but not limited to: generating recognition labels for production line tracking, recording the recognition information into the quality inspection report, starting the assembly or packaging process of the corresponding model, etc.

[0084] Secondly, if it is determined to be a fuzzy type, it means that there is an overlap in feature similarity between the target headrest in two types, and there is a large uncertainty in the judgment result. At this time, to prevent subsequent process errors caused by misidentification, the following processing actions are performed: Automatically record the current image and physical features, mark the recognition result as "fuzzy pending confirmation", and send it to the manual review system. At the same time, it can trigger a prompt for optimizing the acquisition angle or a retake process to obtain a clearer image for reference. Once the review conclusion is confirmed, this sample can be included in the sample library for future model optimization or threshold re-learning.

[0085] Finally, if it is determined to be an unknown type, it means that the target headrest does not have a sufficient confidence match with all types in the existing feature library. This situation usually indicates the following possibilities: The product is a new model that has not been modeled, there are abnormalities in the image, such as incorrect shooting angle, insufficient lighting, or component defects, and there are omissions or obsolescence problems in the feature library. At this time, the library expansion process can be triggered to prompt the technician to perform manual modeling and feature annotation on this sample, or include it in the feature library update process.

[0086] In summary, this application preprocesses the collected target headrest image, extracts multi-dimensional physical features of the headrest from the preprocessed image, including contour shape parameters, gray-level co-occurrence matrix, local binary pattern, color histogram, and color moment, etc., and splices them to generate a normalized high-dimensional feature vector. Compare the feature vector of the target headrest with each template vector in the preset feature library, and use the weighted Gaussian kernel similarity formula combined with standard deviation normalization and feature importance weights to calculate the confidence between the target headrest and each known type. Based on the confidence results of all types, input the confidence results of each type into the decision module, and select the type with the highest confidence as the recognition result; a threshold can be set to judge whether the confidence is reliable, and enter the review process if necessary. Output the recognition result to the execution system to control the subsequent headrest sorting action. This application accurately classifies and identifies the headrest through multi-angle image acquisition, image standardization preprocessing, combined with the extraction of multi-dimensional features such as shape, texture, and color, and uses the weighted Gaussian kernel similarity algorithm. By introducing the feature weight mechanism and the maximum confidence screening, the recognition accuracy and stability are improved, ensuring label-free, automated, and scalable headrest type recognition in a flexible production environment, thus effectively solving the problems of low accuracy, difficult type differentiation, and excessive manual intervention.

[0087] Figure 2 FIG. 10 is a structural block diagram of an automobile headrest sorting system based on confidence matching provided by an embodiment of the present application. The system at least includes the following modules:

[0088] An image acquisition module, configured to acquire a target headrest image and preprocess the target headrest image;

[0089] A feature extraction module, configured to perform multi-dimensional physical feature extraction on the preprocessed target headrest image to generate a physical feature vector of the target headrest;

[0090] A confidence calculation module, configured to calculate the confidence between the physical feature vector of the target headrest and each template feature vector in the preset feature libraries of each headrest type;

[0091] A type determination module, configured to determine the type of the target headrest based on the set of confidences of all headrest types obtained through calculation and the introduced determination threshold;

[0092] A result execution module, configured to perform corresponding operations according to the determination result of the target headrest.

[0093] For relevant details, refer to the above method embodiments.

[0094] Figure 3 It is a block diagram of an electronic device provided by an embodiment of the present application. The device at least includes a processor 401 and a memory 402.

[0095] The processor 401 may include one or more processing cores, such as: a 4-core processor, an 8-core processor, etc. The processor 401 may be implemented in at least one of the following hardware forms: DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 401 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the wake state, also known as the CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 401 may be integrated with a GPU (Graphics Processing Unit), and the GPU is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 401 may further include an AI (Artificial Intelligence) processor, and the AI processor is used to process calculation operations related to machine learning.

[0096] The memory 402 may include one or more computer-readable storage media, which may be non-transitory. The memory 402 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices and flash storage devices. In some embodiments, the non-transitory computer-readable storage media in the memory 402 is used to store at least one instruction for being executed by the processor 401 to implement the method for sorting automotive headrests based on confidence matching provided in the method embodiments of the present application.

[0097] In some embodiments, the electronic device may further optionally include: a peripheral device interface and at least one peripheral device. The processor 401, the memory 402, and the peripheral device interface may be connected through a bus or signal lines. Each peripheral device may be connected to the peripheral device interface through a bus, signal lines, or a circuit board. Schematically, the peripheral devices include, but are not limited to: a radio frequency circuit, a touch display screen, an audio circuit, and a power supply, etc.

[0098] Certainly, the electronic device may also include fewer or more components, and this embodiment does not make any limitations thereto.

[0099] Optionally, the present application also provides a computer-readable storage medium, in which a program is stored, and the program is loaded and executed by a processor to implement the method for sorting automotive headrests based on confidence matching in the above method embodiments.

[0100] Optionally, the present application also provides a computer product, which includes a computer-readable storage medium, in which a program is stored, and the program is loaded and executed by a processor to implement the method for sorting automotive headrests based on confidence matching in the above method embodiments.

[0101] The technical features of the above embodiments may be combined arbitrarily. For the sake of brevity of description, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0102] The above embodiments only represent several implementation manners of the present application, and the description thereof is relatively specific and detailed, but it should not be understood as a limitation to the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.

Claims

1. A method for sorting car headrests based on confidence matching, characterized in that, The method includes: Collecting a target headrest image and preprocessing the target headrest image; Performing multi-dimensional physical feature extraction on the preprocessed target headrest image to generate a physical feature vector of the target headrest; Calculating the confidence of the physical feature vector of the target headrest and each template feature vector in the preset feature libraries of each headrest type; Judging the type of the target headrest based on the set of confidences of all headrest types obtained by calculation and the introduced decision threshold; Performing corresponding operations according to the judgment result of the target headrest.

2. The method for sorting automotive headrests based on confidence matching according to claim 1, wherein, The preprocessing of the target headrest image includes: Normalizing the resolution of the image to uniformly scale images from different devices or different shooting distances to a predetermined size; Suppressing noise in the image, calibrating the color through automatic white balance or gamma correction, and combining methods such as threshold segmentation, edge detection, or depth information masking to accurately segment and crop the headrest area in the image, removing the background and irrelevant areas.

3. The method for sorting automotive headrests based on confidence matching according to claim 1, characterized in that The multi-dimensional physical feature extraction of the preprocessed target headrest image to generate a physical feature vector of the target headrest includes: Performing edge detection on the target headrest image, removing noise points and closing edges through morphological dilation and erosion operations, and calling a contour extraction algorithm to identify the main contour curve of the headrest; Calculating geometric descriptors based on the identified contour curve, converting the headrest area into a grayscale image, and simultaneously extracting local texture microstructures using local binary patterns; Performing color feature extraction on the color image, splicing the extracted geometric, texture, and color features into a high-dimensional feature vector in a predetermined order, and then normalizing the feature values of each dimension to generate a physical feature vector of the target headrest.

4. The method for sorting automotive headrests based on confidence matching according to claim 1, wherein The calculation of the confidence of the physical feature vector of the target headrest and each template feature vector in the preset feature libraries of each headrest type includes: In the pre - established headrest type feature library, each template feature vector is obtained by the same physical feature extraction method. The physical feature vector v of the target headrest = [v1, v2, …, v n has the same dimension as each type of template feature vector t j = [t j,1 , t j,2 , …, t j,n , where j = 1, 2, …, N, and N represents the number of known headrest types in the library; Using a weighted Gaussian kernel similarity formula to calculate the confidence between the target headrest and each template, and the calculation formula is as follows: Among them, β i represents the importance weight of the i-th dimension feature, σ j,i is the standard deviation of the template feature vector in the i-th dimension feature, and ε is a very small positive number.

5. The method for sorting car headrests based on confidence matching according to claim 4, characterized in that, The judgment of the type of the target headrest based on the set of confidences of all headrest types obtained by calculation and the introduced decision threshold includes: Obtain the set of confidence levels {C1, C2, …, C N} calculated for all headrest types, and sort them in descending order to obtain the maximum confidence level value C (1) . Denote the corresponding type index as p, and the second-largest confidence level value C (2) . Denote the corresponding type index as q. Introduce two sets of decision thresholds, the basic matching threshold T j and the ratio threshold M j ; When the maximum confidence C (1) reaches or exceeds the basic matching threshold T of type p p , and the ratio R of the sub-optimal confidence to the optimal confidence is R = C (2) / C (1) less than or equal to the ratio threshold M p the headrest is classified as type p; If C (1) ≥T p , but the ratio R of the sub - optimal confidence level to the optimal confidence level exceeds the ratio threshold M p , mark the sample as the fuzzy type; When the maximum confidence C (1) is lower than the corresponding threshold T p it is marked as an unknown type.

6. The method for sorting automotive headrests based on confidence matching according to claim 5, wherein The judgment of the type of the target headrest based on the set of confidences of all headrest types obtained by calculation and the introduced decision threshold further includes: Design an online threshold adaptive update strategy. Taking the basic matching threshold T of type j as an example, assume that the threshold before the t-th update at the current moment is j When a certain classification is verified as correct, take the actual confidence of this time to update the threshold: ​ Among them, the attenuation factor λ is used to balance historical experience and the latest observation.

7. The method for sorting automotive headrests based on confidence matching according to claim 5, wherein The performing corresponding operations according to the judgment result of the target headrest includes: If it is determined to be a definite type, directly classify the target headrest into the definite type and push it to the subsequent business process; If it is determined to be a fuzzy type, send it to the manual review system; If it is determined to be an unknown type, trigger an expansion library process, prompting the technical staff to perform manual modeling, feature annotation on this sample, or incorporate it into the feature library update process.

8. An automobile headrest sorting system based on confidence matching, characterized in that, Including: An image acquisition module for collecting a target headrest image and preprocessing the target headrest image; A feature extraction module for performing multi-dimensional physical feature extraction on the preprocessed target headrest image to generate a physical feature vector of the target headrest; A confidence calculation module for calculating the confidence of the physical feature vector of the target headrest and each template feature vector in the preset feature libraries of each headrest type; A type judgment module, configured to judge the type to which the target headrest belongs based on the set of confidence levels of all headrest types obtained through calculation and the introduced determination threshold; A result execution module, configured to execute corresponding operations according to the judgment result of the target headrest.

9. An electronic device, characterized in that, The device includes a processor and a memory; a program is stored in the memory, and the program is loaded and executed by the processor to implement a method for sorting automotive headrests based on confidence matching as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A program is stored in the storage medium, and when the program is executed by a processor, it is used to implement a method for sorting automotive headrests based on confidence matching as described in any one of claims 1 to 7.

Images