Exhibition subject identifier classification method and system fusing partition color features

Abstract

The invention discloses an exhibition subject identifier classification method and system fusing partition color features, belongs to the technical field of image recognition, and can solve the problem of low classification precision in the prior art. The method comprises the following steps: S1, extracting a first local area corresponding to each connecting piece and a second local area corresponding to each label from a monitoring image in an exhibition hall by using a target detection model; s2, determining a first effective pixel meeting a first preset condition according to the first local area, and determining a second effective pixel meeting a second preset condition according to the second local area; the first preset condition comprises saturation greater than or equal to preset saturation, and the second preset condition comprises saturation greater than or equal to preset saturation and brightness less than or equal to preset brightness; s3, determining a combined hue of each exhibition subject identifier according to the first effective pixel and the second effective pixel; and S4, classifying all the exhibition subject identifiers according to the combined hues. The method is used for counting the number of exhibitors.

Description

Technical Field

[0001] This invention relates to a method and system for classifying exhibition main signage by integrating zoning color features, belonging to the field of image recognition technology. Background Technology

[0002] The current method of classifying exhibitors' tags by color mainly relies on manual methods, which suffers from low efficiency, susceptibility to lighting and angle interference, and large errors. Existing automatic classification methods mostly use a single RGB color space to extract features, without optimizing for interference factors in monitoring scenarios, and lack quantitative modeling of the combined features of lanyards and signs, resulting in insufficient classification accuracy. Summary of the Invention

[0003] This invention provides a method and system for classifying exhibition main signage by integrating zoning color features, which can solve the problems of low classification efficiency and low classification accuracy in existing technologies.

[0004] On one hand, the present invention provides a method for classifying exhibition main body logos by integrating zoning color features. The exhibition main body logo includes a connector and a sign hanging under the connector. The method includes:

[0005] S1. Use the target detection model to extract the first local region corresponding to each connector and the second local region corresponding to each sign from the surveillance images in the exhibition hall;

[0006] S2. Determine a first effective pixel that meets the first preset condition based on the first local region, and determine a second effective pixel that meets the second preset condition based on the second local region; the first preset condition includes saturation greater than or equal to preset saturation, and the second preset condition includes saturation greater than or equal to the preset saturation, and brightness less than or equal to preset brightness.

[0007] S3. Determine the combined color tone of each exhibition main logo based on the first effective pixel and the second effective pixel;

[0008] S4. Classify all exhibition entity logos according to the combined color scheme in order to count the number of exhibition entities.

[0009] Optionally, S2 specifically includes:

[0010] The first local region is denoised and its color space is converted to obtain a first local image. The second local region is denoised and its color space is converted to obtain a second local image.

[0011] The first effective pixel is determined based on the first local image, and the second effective pixel is determined based on the second local image.

[0012] Optionally, S3 specifically includes:

[0013] S31. Determine the first primary color of each connector based on all first valid pixels of each connector, and determine the second primary color of each sign based on all second valid pixels of each sign.

[0014] S32. Determine the combined color scheme of each exhibition main logo based on the first and second primary colors corresponding to each exhibition main logo.

[0015] Optionally, after S4, the method further includes:

[0016] The classification confidence level of each exhibition main identifier is determined based on the first local region, the second local region, the first effective pixel, the second effective pixel, and the color tone type;

[0017] When the classification confidence level meets the first preset interval, the color type of the corresponding exhibition main body logo is determined to be valid; when the classification confidence level meets the second preset interval, the color type of the corresponding exhibition main body logo is re-determined; the first preset interval is greater than the second preset interval.

[0018] Optionally, the classification confidence level of each exhibition main identifier is determined based on the first local region, the second local region, the first effective pixel, the second effective pixel, and the color tone type, specifically including:

[0019] The first confidence level of each exhibition main logo is determined based on the first local area and the second local area corresponding to each exhibition main logo. The second confidence level of each exhibition main logo is determined based on the first effective pixel and the second effective pixel. The third confidence level is determined based on the color tone type of all exhibition main logos.

[0020] The classification confidence level of each exhibition entity identifier is determined based on the first confidence level, the second confidence level, and the third confidence level.

[0021] Optionally, a first confidence level for each exhibition entity identifier is determined based on a first partial region and a second partial region corresponding to that identifier, specifically including:

[0022] Determine the average value of the cross-union ratio of the detection frames of the first local region and the second local region corresponding to each exhibition main identifier, as well as the minimum value between the actual detection confidence of the first local region and the actual detection confidence of the second local region corresponding to each exhibition main identifier;

[0023] The product of the average value and the minimum value corresponding to each exhibition entity identifier is used as the first confidence level of the corresponding exhibition entity identifier.

[0024] Optionally, a second confidence level for each exhibition entity identifier is determined based on the first and second valid pixels of each identifier, specifically including:

[0025] Determine the percentage of the first effective pixel of each exhibition main identifier in its first local image, and the percentage of the second effective pixel in its second local image;

[0026] The average of the first pixel percentage and the second pixel percentage of each exhibition main identifier is used as the second confidence level of the corresponding exhibition main identifier.

[0027] Optionally, when the classification confidence level meets the second preset interval, the color type of the corresponding exhibition main logo is redefined, specifically including:

[0028] When the classification confidence level meets the second preset range, S1 to S3 are repeated for the corresponding exhibition main body logo to update the combined color tone of the corresponding exhibition main body logo;

[0029] The color scheme of the corresponding main exhibition logo will be redefined based on the updated color scheme.

[0030] Optionally, S31 specifically includes:

[0031] The color with the largest pixel proportion among all first effective pixels of each connector is determined as the first primary color of the corresponding connector, and the color with the largest pixel proportion among all second effective pixels of each sign is determined as the second primary color of the corresponding sign.

[0032] On the other hand, the present invention provides a system for classifying exhibition main signage based on the above-mentioned fusion partition color features, the system comprising:

[0033] The local extraction module is used to extract the first local region corresponding to each connector and the second local region corresponding to each sign from the surveillance images in the exhibition hall using the target detection model.

[0034] A pixel determination module is used to determine a first effective pixel that meets a first preset condition based on a first local region, and to determine a second effective pixel that meets a second preset condition based on a second local region; the first preset condition includes saturation greater than or equal to a preset saturation, and the second preset condition includes saturation greater than or equal to a preset saturation and brightness less than or equal to a preset brightness.

[0035] The color tone determination module is used to determine the combined color tone of each exhibition main logo based on the first effective pixel and the second effective pixel.

[0036] The classification module is used to categorize all exhibition entity logos based on their color combinations, in order to count the number of exhibition entities.

[0037] The beneficial effects that this invention can produce include:

[0038] This invention extracts local areas of connectors and signs from surveillance images, determines the main color tone of these areas, and then determines the combined color tone of the main exhibition signage. Based on this combined color tone, the main exhibition signage is classified to facilitate the counting of exhibition entities. Because this invention integrates color features from different areas of the main exhibition signage, it not only makes the features of each main exhibition signage clearer but also increases the distinguishability between different main exhibition signages. Therefore, compared to classification based on the color features of a single area, it significantly improves the accuracy of classification. Simultaneously, this invention utilizes methods such as noise reduction and color space conversion to optimize the image quality of local areas, effectively eliminating interference from lighting, obstructions, etc., on the classification results. Furthermore, this invention evaluates, verifies, and corrects the effectiveness of the classification results based on multi-dimensional confidence levels, reducing the risk of misclassification and improving the reliability of the classification results, thereby achieving accurate and efficient classification of main exhibition signage. Attached Figure Description

[0039] Figure 1 A flowchart illustrating the exhibition main body identification classification method based on the fusion of partition color features provided in this embodiment of the invention. Detailed Implementation

[0040] The present invention will now be described in detail with reference to the embodiments, but the present invention is not limited to these embodiments.

[0041] This invention provides a method for classifying exhibition main signage by integrating zoning color features. The exhibition main signage includes a connector installed on the ceiling of the exhibition hall and a sign hanging below the connector. The connector can be a hanging rope, chain, connecting rod, or other components. This embodiment uses a hanging rope as an example for illustration.

[0042] like Figure 1 As shown, the method includes:

[0043] S0. Construct an object detection model based on a pre-trained YOLOv8n model. Specifically, this includes:

[0044] 1. Construct the training dataset

[0045] 1) Training sample acquisition and enhancement: Select 13 to 25 historical monitoring images with different lighting and angles, and change the image state by random translation (±5%), rotation (±10°), scaling (0.8 to 1.2 times), brightness adjustment (±15%), contrast adjustment (±10%), etc. to generate 50 to 100 training samples, and unify the format of the training samples to PNG / JPG format.

[0046] 2) Training sample annotation and dataset partitioning: Use image annotation tools such as LabelImg, CVAT, Labelme, LabelImg or Labelbox to annotate two types of targets, namely rope and plate, in each training sample, and divide all training samples into training set, validation set and test set in a ratio of 7:2:1 to obtain training dataset.

[0047] 2. Fine-tuning the pre-trained YOLOv8n model

[0048] 1) Model configuration: Build the configuration file for the training dataset, establish the association path of the training samples, and define two types of targets: ropes and signs.

[0049] 2) Training parameter settings: Load the YOLOv8n model based on COCO pre-training, freeze the first 10 layers of the backbone network, train for 50 epochs, initial learning rate of 0.001, batch size of 8, and enable weight decay (0.0005) and early stopping strategy (stop after 10 epochs if there is no improvement).

[0050] 3) Based on model configuration and training parameter settings, the pre-trained YOLOv8n model is fine-tuned using the training dataset, enabling the model to accurately detect two types of targets, namely ropes and signs, from the images of the training samples.

[0051] 4) Model validation: With mAP50≥0.9 as the target, if there are missed detections due to rope hanging, the training samples are increased or the labeled anchor boxes are adjusted, and the model is further fine-tuned and trained to obtain the target detection model of this implementation.

[0052] S1. Using an object detection model, extract the first local region corresponding to each connector and the second local region corresponding to each sign from the surveillance images in the exhibition hall. Specifically, this includes:

[0053] The target detection model detects the Region of Interest (ROI) of the hanging rope (i.e., the first local region) from the surveillance image based on a preset detection confidence level; simultaneously, it detects the ROI of the sign (i.e., the second local region) from the surveillance image based on the same preset detection confidence level. The preset detection confidence level can be flexibly set according to detection requirements and actual conditions. For example, the preset detection confidence level can be set to 0.7, 0.8, etc.; this embodiment uses 0.7 as an example. The actual detection confidence levels of both the first and second local regions should be greater than or equal to the preset detection confidence level.

[0054] Then, in this embodiment, the detection box coordinates of the first local region and the second local region are normalized to pixel coordinates.

[0055] S2. Determine a first valid pixel based on a first local region, and determine a second valid pixel based on a second local region. Specifically, this includes:

[0056] 1. Perform denoising and color space conversion on the first local region to obtain the first local image. Perform denoising and color space conversion on the second local region to obtain the second local image.

[0057] 1) Noise reduction processing.

[0058] This embodiment uses Gaussian blur denoising as an example. To address the problem of low image quality in the first / second local regions due to low resolution and reflection interference in the surveillance images, this embodiment employs a 3×3 Gaussian convolution kernel for denoising. The mathematical model for denoising is as follows:

[0059] Gaussian convolution kernel definition:

[0060] (1)

[0061] In formula (1), Use a Gaussian convolution kernel; This represents the offset of a pixel relative to the kernel center (kernel neighborhood coordinates). , ; Parameters for controlling the degree of ambiguity, (Adapted to 3×3 convolution kernel size).

[0062] Image convolution operation:

[0063] (2)

[0064] In formula (2), These are the pixel values ​​after noise reduction; These are the pixel values ​​of the original image (first local region / second local region); The index for traversing the Gaussian convolution kernel; For The index is a Gaussian convolution kernel.

[0065] 2) Color space conversion.

[0066] This embodiment aims to eliminate the influence of lighting on color judgment by performing a color space conversion on the first / second local region, converting it from the RGB color space to the HSV color space. The conversion process is as follows:

[0067] First, normalize the RGB values ​​of the first local region / second local region to... .

[0068] Then, define three intermediate variables: maximum value, minimum value, and difference:

[0069] (3)

[0070] (4)

[0071] (5)

[0072] In formulas (3) to (5), These are the normalized RGB values ​​corresponding to the red, green, and blue component values, respectively. , , These represent the maximum value, minimum value, and difference, respectively.

[0073] Calculate lightness:

[0074] (6)

[0075] In formula (6), For brightness, ; This is the maximum value.

[0076] Calculate saturation:

[0077] (7)

[0078] In formula (7), For saturation, ; This is the maximum value.

[0079] Calculate hue:

[0080] (8)

[0081] In formula (8), For hue, ; It is the maximum value; These are the normalized RGB values ​​corresponding to the red, green, and blue component values.

[0082] 2. Determine the first effective pixel that meets the first preset condition based on the first partial image, and determine the second effective pixel that meets the second preset condition based on the second partial image.

[0083] The first preset condition includes saturation greater than or equal to preset saturation, and the second preset condition includes saturation greater than or equal to preset saturation and brightness less than or equal to preset brightness.

[0084] Specifically, in this embodiment, the saturation of each pixel in the first local image is determined, and the mask of pixels with saturation greater than or equal to a preset saturation is set to 1, otherwise it is set to 0. Then, the pixels with a mask of 1 in the first local image are determined as the first valid pixels. The mathematical model of this process is as follows:

[0085] (9)

[0086] In formula (9), The coordinates in the first local image are... The pixel mask, where a value of 1 indicates a pixel. The first valid pixel, where 0 indicates a pixel. Not the first valid pixel; For pixels saturation; The preset saturation level can be 0.2, 0.3, etc. This embodiment will use 0.2 as an example for explanation.

[0087] Similarly, in this embodiment, the saturation and brightness of each pixel in the second local image are first determined, and the mask of pixels with saturation greater than or equal to a preset saturation and brightness less than or equal to a preset brightness is set to 1, otherwise it is set to 0. Then, the pixels with a mask of 1 in the second local image are determined as the second effective pixels. The mathematical model of this process is as follows:

[0088] (10)

[0089] In formula (10), The coordinates in the second local image are... The pixel mask, where a value of 1 indicates a pixel. The second valid pixel, where 0 indicates a pixel. Not the second valid pixel; For pixels saturation; Preset saturation; For pixels Brightness; To preset the brightness, in this embodiment, .

[0090] S3. Determine the combined color tone for each exhibition main logo based on the first and second effective pixels. Specifically, this includes:

[0091] S31. Determine the first primary color of each connector based on all first valid pixels of each connector, and determine the second primary color of each sign based on all second valid pixels of each sign. Specifically, this includes:

[0092] 1. Divide the hue range.

[0093] In this embodiment, the color wheel is divided into 12 hue intervals, each defined as a 30° interval. Each hue interval corresponds to a class of hues (such as red, orange, yellow, green, etc.), and each hue interval can be represented as:

[0094] (11)

[0095] In formula (11), For the first Each hue range is represented by a corresponding angle range on the color wheel.

[0096] 2. Determine the percentage of pixels with different hues in the first effective pixel / second effective pixel.

[0097] Specifically, for each lanyard, this embodiment first determines the total number of its first effective pixels, and then determines the number of interval pixels belonging to each hue interval among its first effective pixels. The quotient obtained by dividing the number of interval pixels by the total number of first effective pixels is used as the pixel ratio of the corresponding hue.

[0098] Similarly, for each sign, this embodiment first determines the total number of its second effective pixels, and then determines the number of interval pixels belonging to each hue interval among its second effective pixels. The quotient obtained by dividing the number of interval pixels by the total number of second effective pixels is used as the pixel proportion of the corresponding hue.

[0099] The mathematical model for this process is as follows:

[0100] (12)

[0101] In formula (12), The first effective pixel / the second effective pixel The percentage of pixels in the same color tone; First effective pixel / second effective pixel The mask; First effective pixel / second effective pixel The hue range it belongs to; For indicator functions, when hour, It is 1 if it is true, otherwise it is 0.

[0102] 3. Determine the primary and secondary color schemes for the hanging ropes / signs.

[0103] In this embodiment, the hue with the largest pixel proportion among all first effective pixels of each lanyard is determined as the first primary hue of the corresponding lanyard, and the hue with the largest pixel proportion among all second effective pixels of each sign is determined as the second primary hue of the corresponding sign. The mathematical model for this process is as follows:

[0104] (13)

[0105] In formula (13), The primary / secondary primary color scheme for the hanging rope / signboard; The first effective pixel / the second effective pixel The percentage of pixels in the color tone category.

[0106] S32. Determine the combined color scheme of each exhibition main logo based on the first and second primary colors corresponding to each exhibition main logo.

[0107] This embodiment first determines the codes for each first primary color and each second primary color. For each exhibition main logo, its first primary color code is multiplied by the total number of all first primary colors to obtain the product. Then, the sum of this product and the codes of the second primary colors of the exhibition main logo is used as the code for the combined color of the exhibition main logo. This yields different combined color schemes. The mathematical model for this process is as follows:

[0108] If there are 12 primary colors and 12 secondary colors, the code for the primary color of the lanyard can be set as follows: The secondary primary color of the sign is coded as follows: The code for the combined hue is:

[0109] (14)

[0110] In formula (14), The coding of the combined color scheme for the main logo of the exhibition; The primary color code for the lanyard; The code for the second primary color of the sign.

[0111] Since this embodiment has 12 primary color tones and 12 secondary color tones, there are a total of 12 × 12 = 144 color tones possible combinations. .

[0112] S4. Classify all exhibition and convention main logos according to the combined color scheme in order to count the number of exhibition and convention main logos.

[0113] In S4, all exhibition and convention main logos are categorized based on their combined color scheme. Specifically, this includes:

[0114] Based on the combined color scheme of all exhibition main logos, a clustering algorithm is used to classify all exhibition main logos.

[0115] Clustering algorithms can include adaptive K-Means clustering, K-medoids clustering, or K-modes clustering, etc. This embodiment uses adaptive K-Means clustering as an example for illustration. The mathematical model of the clustering process is as follows:

[0116] 1. Cluster center initialization:

[0117] Based on the combined color scheme of all exhibition main logos, the elbow rule is used to calculate the clustering inertia value, and the value is selected when the rate of decrease in clustering inertia value changes abruptly. Value as the optimal cluster number Based on the optimal number of clusters Initialize cluster centers .

[0118] 2. Sample allocation:

[0119] (15)

[0120] In formula (15), For the first Clustering; The absolute distance in a one-dimensional feature scene; The coding of the combined color scheme for the main logo of the exhibition; and The first Cluster centers and the first Cluster centers.

[0121] 3. Cluster center update:

[0122] (16)

[0123] In formula (16), This represents the number of iterations. For the first The number of samples in each cluster; for The iteration of the ... Cluster centers of clusters; For the sample; For the first Clustering.

[0124] 4. Iteration termination condition:

[0125] (17)

[0126] In formula (17), This represents the number of iterations. for The iteration of the ... Cluster centers of clusters; for The iteration of the ... Cluster centers of clusters; The convergence threshold, .

[0127] 5. Clustering effectiveness assessment:

[0128] (18)

[0129] In formula (18), For the sample Average distance to similar samples; For the sample The average distance to the nearest outlier sample; The number of all samples; This is a clustering effectiveness index. This indicates excellent clustering performance.

[0130] S5. Verify classification accuracy using multi-dimensional confidence scores. Specifically, this includes:

[0131] 1. Determine the first confidence level of each exhibition main body logo based on the first local area and the second local area corresponding to each exhibition main body logo.

[0132] Specifically, for each exhibition entity identifier, this embodiment first determines the average of the cross-union ratio (CUI) of the detection frames in its first local region and the second local region, as well as the minimum of the actual detection confidence scores of the first and second local regions. Then, the product of this average and the minimum is used as the first confidence score of the exhibition entity identifier. The mathematical model for this process is as follows:

[0133] (19)

[0134] In formula (19), The crossover ratio (CLORD) of the detection frames for the first local area of ​​the hanging rope is ≥0.5. The crossover ratio (CLORD) of the detection frames for the second local region of the sign is ≥0.5. The actual detection confidence level (≥0.7) for the first local area of ​​the hanging rope. The actual detection confidence level (≥0.7) for the second local area of ​​the sign. The first confidence level of the main exhibition logo reflects the target detection confidence level of the target detection model. .

[0135] 2. Determine the second confidence level of each exhibition entity identifier based on the first and second effective pixels of each exhibition entity identifier.

[0136] Specifically, for each exhibition entity identifier, this embodiment first determines the proportion of its first effective pixel in its first partial image and the proportion of its second effective pixel in its second partial image. Then, the average of the first pixel proportion and the second pixel proportion is used as the second confidence level of the exhibition entity identifier. The mathematical model for this process is as follows:

[0137] (20)

[0138] (twenty one) (twenty two)

[0139] In formulas (20) to (22), The second confidence level for the main logo of the exhibition reflects the confidence level of the combined color scheme. ; The percentage of the first effective pixel of the main identifier of the exhibition in its first partial image; The proportion of the second effective pixel of the main identifier of the exhibition in its second local image.

[0140] 3. Determine the third confidence level based on the color scheme of all exhibition main logos.

[0141] Specifically, this embodiment determines the proportion of samples with the same combined hue in each cluster obtained by S4 classification, and then divides the sum of the proportions of samples corresponding to all clusters by the total number of clusters to obtain the third confidence level. The mathematical model for this process is as follows:

[0142] (twenty three)

[0143] In formula (23), For the first The number of samples with the same combined hue in a cluster; The optimal number of clusters; For the first The number of samples in each cluster; The third confidence level reflects the degree of confidence in the cluster purity. .

[0144] 4. Determine the classification confidence level of each exhibition entity identifier based on the first confidence level, second confidence level, and third confidence level.

[0145] This embodiment performs a weighted sum of the first, second, and third confidence levels for each exhibition entity identifier to obtain the corresponding classification confidence level for the exhibition entity identifier, i.e.:

[0146] (twenty four)

[0147] In formula (24), The classification confidence level of the main exhibition logo reflects the overall confidence level of the classification results. ; The first level of confidence for the main logo of the exhibition; The second level of confidence for the main logo of the exhibition; The third confidence level; , , They are respectively , , The weighting coefficients, .

[0148] in, , , It can be flexibly configured according to specific scenarios. For example, Adjustable in complex lighting scenarios Adjustable in occluded scenes .

[0149] 5. Determine whether the classification results are valid based on the classification confidence level.

[0150] Define a first preset interval, a second preset interval, and a third preset interval corresponding to the classification confidence value, with the first preset interval, the second preset interval, and the third preset interval decreasing sequentially. For example, the first preset interval is [0.8, 1], the second preset interval is [0.7, 0.8], and the third preset interval is [0, 0.7].

[0151] When the classification confidence level meets the first preset interval, the color type of the corresponding exhibition main logo is determined to be valid;

[0152] When the classification confidence level meets the second preset interval, S1 to S3 are repeated for the corresponding exhibition main body logo to update the combined color tone of the corresponding exhibition main body logo, and the color tone type of the corresponding exhibition main body logo is re-determined based on the updated combined color tone.

[0153] When the classification confidence level meets the third preset interval, the corresponding exhibition main identifier is determined to be a interference sample, and its color type is invalid.

[0154] Finally, this embodiment counts the number of color types that are determined to be valid. Since the nameplates used by the same exhibition entity (exhibitor) are usually of the same color type, the number of color types that are determined to be valid can be used as the number of exhibition entities, thereby realizing the counting of the number of exhibition entities.

[0155] Another embodiment of the present invention provides a system for classifying exhibition main body logos based on the above-mentioned fusion partition color features, the system comprising:

[0156] The local extraction module is used to extract the first local region corresponding to each connector and the second local region corresponding to each sign from the surveillance images in the exhibition hall using the target detection model.

[0157] A pixel determination module is used to determine a first effective pixel that meets a first preset condition based on a first local region, and to determine a second effective pixel that meets a second preset condition based on a second local region; the first preset condition includes saturation greater than or equal to a preset saturation, and the second preset condition includes saturation greater than or equal to a preset saturation and brightness less than or equal to a preset brightness.

[0158] The color tone determination module is used to determine the combined color tone of each exhibition main logo based on the first effective pixel and the second effective pixel.

[0159] The classification module is used to categorize all exhibition entity logos based on their color combinations, in order to count the number of exhibition entities.

[0160] This invention extracts local areas of connectors and signs from surveillance images, determines the main color tone of these areas, and then determines the combined color tone of the main exhibition signage. Based on this combined color tone, the main exhibition signage is classified to facilitate the counting of exhibition entities. Because this invention integrates color features from different areas of the main exhibition signage, it not only makes the features of each main exhibition signage clearer but also increases the distinguishability between different main exhibition signages. Therefore, compared to classification based on the color features of a single area, it significantly improves the accuracy of classification. Simultaneously, this invention utilizes methods such as noise reduction and color space conversion to optimize the image quality of local areas, effectively eliminating interference from lighting, obstructions, etc., on the classification results. Furthermore, this invention evaluates, verifies, and corrects the effectiveness of the classification results based on multi-dimensional confidence levels, reducing the risk of misclassification and improving the reliability of the classification results, thereby achieving accurate and efficient classification of main exhibition signage.

[0161] The above description is merely a few embodiments of this application and is not intended to limit this application in any way. Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any changes or modifications made by those skilled in the art without departing from the scope of the technical solution of this application using the disclosed technical content are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. A method for classifying exhibition main signage by integrating zoning color features, wherein the exhibition main signage includes a connector and a sign suspended below the connector, characterized in that, The method includes: S1. Use the target detection model to extract the first local region corresponding to each connector and the second local region corresponding to each sign from the surveillance images in the exhibition hall; S2. Determine a first effective pixel that meets the first preset condition based on the first local region, and determine a second effective pixel that meets the second preset condition based on the second local region; the first preset condition includes saturation greater than or equal to preset saturation, and the second preset condition includes saturation greater than or equal to the preset saturation, and brightness less than or equal to preset brightness. S3. Determine the combined color tone of each exhibition main logo based on the first effective pixel and the second effective pixel; S4. Classify all exhibition entity logos according to the combined color scheme in order to count the number of exhibition entities.

2. The method according to claim 1, characterized in that, S2 specifically includes: The first local region is denoised and its color space is converted to obtain a first local image. The second local region is denoised and its color space is converted to obtain a second local image. The first effective pixel is determined based on the first local image, and the second effective pixel is determined based on the second local image.

3. The method according to claim 1, characterized in that, S3 specifically includes: S31. Determine the first primary color of each connector based on all first valid pixels of each connector, and determine the second primary color of each sign based on all second valid pixels of each sign. S32. Determine the combined color scheme of each exhibition main logo based on the first and second primary colors corresponding to each exhibition main logo.

4. The method according to claim 2, characterized in that, Following S4, the method further includes: The classification confidence level of each exhibition main identifier is determined based on the first local region, the second local region, the first effective pixel, the second effective pixel, and the color tone type; When the classification confidence level meets the first preset interval, the color type of the corresponding exhibition main body logo is determined to be valid; when the classification confidence level meets the second preset interval, the color type of the corresponding exhibition main body logo is re-determined; the first preset interval is greater than the second preset interval.

5. The method according to claim 4, characterized in that, The classification confidence level of each exhibition main identifier is determined based on the first local region, the second local region, the first effective pixel, the second effective pixel, and the color tone type, specifically including: The first confidence level of each exhibition main logo is determined based on the first local area and the second local area corresponding to each exhibition main logo. The second confidence level of each exhibition main logo is determined based on the first effective pixel and the second effective pixel. The third confidence level is determined based on the color tone type of all exhibition main logos. The classification confidence level of each exhibition entity identifier is determined based on the first confidence level, the second confidence level, and the third confidence level.

6. The method according to claim 5, characterized in that, The first confidence level of each exhibition entity identifier is determined based on the first and second local regions corresponding to each entity identifier, specifically including: Determine the average value of the cross-union ratio of the detection frames of the first local region and the second local region corresponding to each exhibition main identifier, as well as the minimum value between the actual detection confidence of the first local region and the actual detection confidence of the second local region corresponding to each exhibition main identifier; The product of the average value and the minimum value corresponding to each exhibition entity identifier is used as the first confidence level of the corresponding exhibition entity identifier.

7. The method according to claim 5, characterized in that, The second confidence level of each exhibition entity identifier is determined based on the first and second valid pixels of each identifier, specifically including: Determine the percentage of the first effective pixel of each exhibition main identifier in its first local image, and the percentage of the second effective pixel in its second local image; The average of the first pixel percentage and the second pixel percentage of each exhibition main identifier is used as the second confidence level of the corresponding exhibition main identifier.

8. The method according to claim 5, characterized in that, When the classification confidence level meets the second preset interval, the color type of the corresponding exhibition main logo is redefined, specifically including: When the classification confidence level meets the second preset range, S1 to S3 are repeated for the corresponding exhibition main body logo to update the combined color tone of the corresponding exhibition main body logo; The color scheme of the corresponding main exhibition logo will be redefined based on the updated color scheme.

9. The method according to claim 3, characterized in that, S31 specifically includes: The color with the largest pixel proportion among all first effective pixels of each connector is determined as the first primary color of the corresponding connector, and the color with the largest pixel proportion among all second effective pixels of each sign is determined as the second primary color of the corresponding sign.

10. A system for classifying exhibition main body logos based on the fusion of partition color features as described in claim 1, the system comprising: The local extraction module is used to extract the first local region corresponding to each connector and the second local region corresponding to each sign from the surveillance images in the exhibition hall using the target detection model. A pixel determination module is used to determine a first effective pixel that meets a first preset condition based on a first local region, and to determine a second effective pixel that meets a second preset condition based on a second local region; the first preset condition includes saturation greater than or equal to a preset saturation, and the second preset condition includes saturation greater than or equal to a preset saturation and brightness less than or equal to a preset brightness. The color tone determination module is used to determine the combined color tone of each exhibition main logo based on the first effective pixel and the second effective pixel. The classification module is used to categorize all exhibition entity logos based on their color combinations, in order to count the number of exhibition entities.

Images