Image recognition method and system based on multi-source data fusion

Abstract

The invention discloses an image recognition method and system based on multi-source data fusion, relates to the technical field of image recognition, and aims to solve the problems of poor image quality and low recognition efficiency in the image recognition process of multivariate data fusion. According to the method, high-dimensional standard image data can be mapped to a low-dimensional subspace through feature dimensionality reduction, so that the dimensionality of the data is greatly reduced, the calculation complexity in a subsequent image recognition task is reduced, the integral style of the image can be unified and attractive by selecting the integrated background image as a container of the target image data, and the image recognition efficiency is improved. The position of target image data is determined on the integrated background image, the position and layout of each image can be flexibly adjusted according to needs, key point detection is performed by using an accelerated robust feature algorithm, feature points in the images can be rapidly extracted, good stability and robustness are maintained, and the method is suitable for processing of a large-scale image data set.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to an image recognition method and system based on multi-source data fusion. Background Art

[0002] Image recognition refers to the technology of using computers to process, analyze and understand images in order to identify targets and objects of various different patterns.

[0003] The Chinese patent with publication number CN116912141A discloses a deep neural network target recognition method based on multi-source SAR image fusion. It mainly adopts a multi-source SAR image fusion method to provide a feature-rich and clear input image for the deep neural network target recognition method. Compared with the existing image recognition technology based on images taken in a single direction, it has a higher image recognition accuracy. Through the image fusion of multi-source SAR images based on the image feature level, it solves the problem that the coherent speckle noise, azimuth angle, pitch angle and other factors cannot obtain feature-rich and clear SAR images. Although the above patent solves the problem of image recognition, there are still the following problems in actual operation:

[0004] 1. After the image data fusion is completed, no final image evaluation and plan formulation are performed, resulting in reduced image data processing efficiency.

[0005] 2. The processed image data is not further optimized, resulting in poor image quality.

[0006] 3. The image data is not effectively integrated and key features are not confirmed, resulting in poor stability of the image data.

[0007] 4. The data is not stored and processed effectively, resulting in poor quality of the original image data. Summary of the invention

[0008] The purpose of the present invention is to provide an image recognition method and system based on multi-source data fusion. Through feature dimensionality reduction, high-dimensional standard image data can be mapped to a low-dimensional subspace, which greatly reduces the dimension of the data, thereby reducing the computational complexity in subsequent image recognition tasks. By selecting an integrated background image as a container for target image data, the overall style of the image can be unified and beautiful. The position of the target image data is confirmed on the integrated background image, and the position and layout of each image can be flexibly adjusted as needed. The accelerated robust feature algorithm is used for key point detection, which can quickly extract feature points in the image, maintain good stability and robustness, is suitable for processing large-scale image data sets, and can solve the problems in the prior art.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] An image recognition method based on multi-source data fusion includes the following steps:

[0011] S1: Image data acquisition and classification: image data is acquired from different data sources, and the acquired image data is uniquely coded and labeled;

[0012] S2: Obtain image preprocessing: perform data preprocessing on each image data with a unique coding label, and mark the image data after data preprocessing as target image data;

[0013] S3: Processing image fusion: integrating the target image data and extracting key features from the integrated image to obtain standard image data;

[0014] S4: Fusion image processing and recognition: optimizing the standard image data, and performing image recognition on the optimized standard image data, and obtaining evaluation image data after image recognition;

[0015] S5: Identify image evaluation: The image data is evaluated for image quality, and an adjustment plan is formulated based on the quality evaluation results.

[0016] Preferably, the step of acquiring image data from different data sources in S1 and uniquely coding and labeling the acquired image data respectively includes:

[0017] Acquire image data of different illumination and angles from different terminals and sensors;

[0018] The image data is uniquely coded and labeled using the database's self-incrementing ID and timestamp;

[0019] The acquired image data is stored in a database, and after storage is completed, a unique ID is assigned through the database's self-incrementing ID;

[0020] Timestamping the image data stored in the database and assigned with a unique ID;

[0021] Generate a time label for the image data to which the unique ID has been assigned in the database according to the timestamp;

[0022] The unique coding label of the image data is obtained based on the time label and the assigned unique ID.

[0023] Preferably, data preprocessing is performed on each image data with a unique coding number in S2, including:

[0024] Perform data preprocessing on the image data with unique coding labels;

[0025] The data preprocessing steps are as follows: first, the image data is loaded through the image processing library, and after the image loading is completed, the image data is unified in format, type, and size;

[0026] The unified image data is then normalized, wherein the normalization is to scale the pixel values ​​of the image data to within a preset range, and the preset range is 0 to 1 or -1 to 1.

[0027] Preferably, the data preprocessing of each image data with a unique coding number in S2 further includes:

[0028] The normalized image data is denoised through a filter;

[0029] The denoised image data is adjusted for contrast through histogram equalization;

[0030] Perform data enhancement on the image data after contrast adjustment;

[0031] After data enhancement, the preprocessed image data is obtained;

[0032] And mark the image data after preprocessing as target image data.

[0033] Preferably, the step of integrating the target image data in S3 includes:

[0034] First, an integrated background image is selected from the data, wherein the integrated background image is a container of the target image data, and the size of the integrated background image is larger than the size of the target image data, and the margin size of the integrated background image is consistent;

[0035] Confirm the position of the target image data on the integrated background image;

[0036] After the position of the target image data is confirmed, the location and size attributes of each target image data are confirmed;

[0037] Create image datasets of target image data respectively according to positioning and size attributes.

[0038] Preferably, performing key feature extraction on the image to be integrated in S3 includes:

[0039] Each image in the image data set is subjected to key point detection by an accelerated robust feature algorithm, wherein key point detection is to extract feature points in the image, and the feature points include corner points and edge points of the image;

[0040] Generate a descriptor for each detected key point, where the descriptor is the local image information around the key point;

[0041] Perform feature matching on the descriptors generated by the key points, wherein feature matching is to calculate the similarity measure between two descriptors and determine the matching key points according to the similarity threshold;

[0042] After the key points of the image data set are matched, the key features of the target image data are obtained and marked as standard image data.

[0043] Preferably, the image optimization of the standard image data in S4 includes:

[0044] Optimizing the standard image data, wherein the image optimization includes processing the skewness, clarity, distortion, denoising, rendering and sharpening of the standard image data;

[0045] First, the standard image data is processed in order of skewness, clarity, and distortion;

[0046] Among them, the skewness threshold, clarity threshold, and distortion threshold of the standard image data are confirmed in turn;

[0047] If the skewness threshold, clarity threshold, and distortion threshold are not within the standard threshold range, the threshold is adjusted until it is within the standard threshold range;

[0048] After the skewness threshold, clarity threshold, and distortion threshold of the standard image data are adjusted, denoising, rendering, and sharpening are performed;

[0049] The standard image data after threshold adjustment is filtered and denoised by using a Gaussian filter model until the denoising effect reaches a set range;

[0050] Then, the denoised standard image data is rendered, the brightness value and grayscale level of the denoised standard image data are adjusted, and when the brightness value and grayscale level reach the white balance standard, the standard image data is color rendered;

[0051] Weakening the color-rendered standard image data, confirming the foreground area and the background area of ​​the color-rendered standard image data, and then calculating the contrast between the foreground area and the background area;

[0052] If the contrast is less than the set contrast threshold, the foreground area is enhanced; the edge features of the critical position between the foreground area and the background area are extracted, and the critical position width is identified based on the edge features. If the critical position width exceeds the set width threshold, the critical position width is symmetrically reduced on both sides to be equal to the width threshold, so as to obtain the standard image data after sharpening adjustment.

[0053] Preferably, performing image recognition on the optimized standard image data in S4 includes:

[0054] Performing feature dimensionality reduction on the standard image data, wherein the feature dimensionality reduction is performing feature analysis on the standard image data;

[0055] The weight value of each analyzed feature is evaluated according to statistical tests, information gain and model weight indicators, and the features with the top three weight values ​​are selected to map the original features to a new low-dimensional subspace through mathematical transformation, and retain the key information in the standard image data;

[0056] The standard image data in the low-dimensional subspace is the data after feature dimensionality reduction;

[0057] The standard image data after dimensionality reduction is trained through a neural network model, and label mapping is performed after the training is completed. After label mapping, the standard image data after image recognition is obtained and marked as evaluation image data.

[0058] An image recognition method based on multi-source data fusion, comprising:

[0059] Image contrast evaluation module for:

[0060] Acquiring parameters for evaluating the image data, wherein the parameters for evaluating the image data include a contrast parameter, a brightness parameter, a saturation parameter, a hue parameter, an entropy parameter, and a clarity parameter;

[0061] Comparing parameters of the evaluation image data with parameters of qualified image data, wherein the qualified image data and the parameters of the qualified image data are retrieved from a database;

[0062] Judging and evaluating the quality level of the image data according to the comparison result;

[0063] Among them, the quality levels are divided into first-level quality, second-level quality and third-level quality.

[0064] Preferably, said:

[0065] Evaluation and adjustment plan decision module, used to:

[0066] Formulate an adjustment plan for the evaluation image data of each quality level according to the quality level of the evaluation image data;

[0067] The adjustment plan is sent to the display terminal for display.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] 1. The present invention provides an image recognition method and system based on multi-source data fusion, which uses a database self-increment ID combined with a timestamp to uniquely encode and label image data, thereby ensuring that each image has a unique identifier in the system. Data preprocessing of the uniquely coded image data helps to reduce errors caused by inconsistent data formats and improve the accuracy and stability of image recognition.

[0070] 2. The present invention provides an image recognition method and system based on multi-source data fusion. By selecting an integrated background image as a container for target image data, the overall style of the image can be unified and beautiful. The position of the target image data is confirmed on the integrated background image. The position and layout of each image can be flexibly adjusted as needed. The accelerated robust feature algorithm is used for key point detection, which can quickly extract feature points in the image, maintain good stability and robustness, and is suitable for processing large-scale image data sets.

[0071] 3. The present invention provides an image recognition method and system based on multi-source data fusion, which can significantly improve the overall quality of the image by processing the skewness, clarity and distortion of standard image data. The high-dimensional standard image data can be mapped to a low-dimensional subspace through feature dimensionality reduction, which greatly reduces the dimension of the data and thus reduces the computational complexity in subsequent image recognition tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 A schematic diagram of the image recognition steps of multi-source data fusion of the present invention;

[0073] Figure 2 It is a schematic diagram of the image recognition process of multi-source data fusion of the present invention. DETAILED DESCRIPTION

[0074] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0075] In order to solve the problem in the prior art that after the image data is acquired, the data is not effectively stored and processed, resulting in poor quality of the original image data, please refer to Figure 1 and Figure 2 , this embodiment provides the following technical solutions:

[0076] An image recognition method based on multi-source data fusion includes the following steps:

[0077] S1: Image data acquisition and classification: image data is acquired from different data sources, and the acquired image data is uniquely coded and labeled;

[0078] Among them, the image data is uniquely coded and labeled by combining the database's self-increment ID with the timestamp, which can ensure that each image has a unique identification in the system;

[0079] S2: Obtain image preprocessing: perform data preprocessing on each image data with a unique coding label, and mark the image data after data preprocessing as target image data;

[0080] Among them, by unifying the format, type and size of image data, it ensures that all image data are consistent and comparable in subsequent processing and analysis;

[0081] S3: Processing image fusion: integrating the target image data and extracting key features from the integrated image to obtain standard image data;

[0082] Among them, the use of accelerated robust feature algorithm for key point detection can quickly extract feature points in the image and maintain good stability and robustness;

[0083] S4: Fusion image processing and recognition: optimizing the standard image data, and performing image recognition on the optimized standard image data, and obtaining evaluation image data after image recognition;

[0084] Among them, by processing the skewness, clarity and distortion of standard image data, the overall quality of the image can be significantly improved;

[0085] S5: Identification image evaluation: image quality evaluation is performed on the image data, and adjustment plans are formulated based on the quality evaluation results;

[0086] Among them, an adjustment plan is formulated according to the quality level of the evaluated image data and sent to the display terminal for display, which not only improves the processing speed but also ensures the pertinence and effectiveness of the adjustment plan.

[0087] In S1, image data is obtained from different data sources, and the obtained image data are uniquely coded and labeled, including:

[0088] Acquire image data of different illumination and angles from different terminals and sensors;

[0089] The image data is uniquely coded and labeled using the database's self-incrementing ID and timestamp;

[0090] The acquired image data is stored in a database, and after storage is completed, a unique ID is assigned through the database's self-incrementing ID;

[0091] Timestamping the image data stored in the database and assigned with a unique ID;

[0092] Generate a time label for the image data to which the unique ID has been assigned in the database according to the timestamp;

[0093] The unique coding label of the image data is obtained based on the time label and the assigned unique ID.

[0094] Specifically, acquiring image data from different terminals and sensors can ensure the diversity and richness of data. This includes images with different lighting conditions and shooting angles, which helps the image recognition or analysis system better adapt to various practical scenarios and improve the generalization and accuracy of the system. The image data is uniquely encoded and labeled using a combination of a database auto-increment ID and a timestamp to ensure that each image has a unique identifier in the system. This helps avoid data confusion and improves the accuracy and efficiency of data processing. The use of unique coding labels can enhance data security. Since each image has a unique identifier, it is easier to detect and prevent data tampering or forgery during data transmission, storage, and processing.

[0095] Data preprocessing is performed on each image data with a unique coding number in S2, including:

[0096] Perform data preprocessing on the image data with unique coding labels;

[0097] The data preprocessing steps are as follows: first, the image data is loaded through the image processing library, and after the image loading is completed, the image data is unified in format, type, and size;

[0098] The unified image data is then normalized, wherein the normalization is to scale the pixel values ​​of the image data to within a preset range, and the preset range is 0 to 1 or -1 to 1.

[0099] The normalized image data is denoised through a filter;

[0100] The denoised image data is adjusted for contrast through histogram equalization;

[0101] Perform data enhancement on the image data after contrast adjustment;

[0102] After data enhancement, the preprocessed image data is obtained;

[0103] And mark the image data after preprocessing as target image data.

[0104] Specifically, by unifying the format, type, and size of image data, it ensures that all image data are consistent and comparable in subsequent processing and analysis. This helps reduce errors caused by inconsistent data formats and improves the accuracy and stability of image recognition. The normalization process scales pixel values ​​to a preset range, which helps machine learning models converge faster and improves prediction accuracy. Normalization can also eliminate deviations caused by different image brightness and contrast, allowing the model to focus more on the essential features of the image content. Using filters for denoising can eliminate noise and interference in the image and improve the quality of the image. Histogram equalization can adjust the contrast of the image so that both dark and bright details of the image can be better displayed. By performing data enhancement on the image data, more training samples can be generated and the amount of training data for the model can be increased.

[0105] In order to solve the problem that the existing technology does not effectively integrate image data and confirm key features, which leads to poor stability of image data, please refer to Figure 1 and Figure 2 , this embodiment provides the following technical solutions:

[0106] Perform image integration on the target image data in S3, including:

[0107] First, an integrated background image is selected from the data, wherein the integrated background image is a container of the target image data, and the size of the integrated background image is larger than the size of the target image data, and the margin size of the integrated background image is consistent;

[0108] Confirm the position of the target image data on the integrated background image;

[0109] After the position of the target image data is confirmed, the location and size attributes of each target image data are confirmed;

[0110] Create image datasets of target image data respectively according to positioning and size attributes.

[0111] Specifically, by selecting an integrated background image as a container for target image data, the overall style of the image can be unified and beautiful. By confirming the position of the target image data on the integrated background image, the position and layout of each image can be flexibly adjusted as needed. This helps to ensure a harmonious spatial relationship between images, avoid overlapping or blank areas, and improve the overall visual effect. After confirming the positioning and size attributes of each target image data, the size, proportion, position and other attributes of each image can be adjusted according to specific needs. By integrating multiple target image data into a background image, these images can be easily managed and displayed. By integrating image data, the number of requests to the server can be reduced, the total size of the image can be reduced, and the page loading speed can be increased. The integrated image can be saved, transmitted and displayed as a whole, reducing the complexity of processing each image separately.

[0112] Extract key features of the images to be integrated in S3, including:

[0113] Each image in the image data set is subjected to key point detection by an accelerated robust feature algorithm, wherein key point detection is to extract feature points in the image, and the feature points include corner points and edge points of the image;

[0114] Generate a descriptor for each detected key point, where the descriptor is the local image information around the key point;

[0115] Perform feature matching on the descriptors generated by the key points, wherein feature matching is to calculate the similarity measure between two descriptors and determine the matching key points according to the similarity threshold;

[0116] After the key points of the image data set are matched, the key features of the target image data are obtained and marked as standard image data.

[0117] Specifically, using the accelerated robust feature algorithm for key point detection can quickly extract feature points in the image, maintain good stability and robustness, and is suitable for processing large-scale image data sets. By generating a descriptor for each detected key point, the local image information around the key point can be captured. The descriptor not only contains the location information of the key point, but also contains the texture, structure and other features around it, providing a rich information basis for subsequent feature matching. By calculating the similarity measure between descriptors and determining the matching key points according to the preset similarity threshold, accurate feature matching can be achieved. This matching method has a certain robustness to rotation, scale change, illumination change, etc. between images, and can improve the accuracy and stability of image recognition.

[0118] In order to solve the problem in the prior art that the processed image data is not further optimized, resulting in poor image quality, please refer to Figure 1 and Figure 2 , this embodiment provides the following technical solutions:

[0119] Optimize the standard image data in S4, including:

[0120] Optimizing the standard image data, wherein the image optimization includes processing the skewness, clarity, distortion, denoising, rendering and sharpening of the standard image data;

[0121] First, the standard image data is processed in order of skewness, clarity, and distortion;

[0122] Among them, the skewness threshold, clarity threshold, and distortion threshold of the standard image data are confirmed in turn;

[0123] If the skewness threshold, clarity threshold, and distortion threshold are not within the standard threshold range, the threshold is adjusted until it is within the standard threshold range;

[0124] After the skewness threshold, clarity threshold, and distortion threshold of the standard image data are adjusted, denoising, rendering, and sharpening are performed;

[0125] The standard image data after threshold adjustment is filtered and denoised by using a Gaussian filter model until the denoising effect reaches a set range;

[0126] Then, the denoised standard image data is rendered, the brightness value and grayscale level of the denoised standard image data are adjusted, and when the brightness value and grayscale level reach the white balance standard, the standard image data is color rendered;

[0127] Weakening the color-rendered standard image data, confirming the foreground area and the background area of ​​the color-rendered standard image data, and then calculating the contrast between the foreground area and the background area;

[0128] If the contrast is less than the set contrast threshold, the foreground area is enhanced; the edge features of the critical position between the foreground area and the background area are extracted, and the critical position width is identified based on the edge features. If the critical position width exceeds the set width threshold, the critical position width is symmetrically reduced on both sides to be equal to the width threshold, so as to obtain the standard image data after sharpening adjustment.

[0129] Specifically, by processing the skewness, clarity, and distortion of standard image data, the overall quality of the image can be significantly improved. Skewness adjustment can ensure the correct direction of the image, clarity adjustment can make the image details clearer, and distortion processing can reduce the distortion of the image during transmission or processing. The use of the Gaussian filter model for denoising can effectively remove noise from the image and improve the purity and readability of the image. By rendering and color adjustment of the image, the color of the image can be made more vivid and natural. Contrast adjustment can make the light and dark contrast of the image more obvious, which helps to highlight the important information in the image. The processing of edge features and the adjustment of the critical position width reflect the refined processing of image details. This processing can not only improve the visual effect of the image, but also provide more accurate information for subsequent image analysis and recognition.

[0130] Image recognition is performed on the optimized standard image data in S4, including:

[0131] Performing feature dimensionality reduction on the standard image data, wherein the feature dimensionality reduction is performing feature analysis on the standard image data;

[0132] The weight value of each analyzed feature is evaluated according to statistical tests, information gain and model weight indicators, and the features with the top three weight values ​​are selected to map the original features to a new low-dimensional subspace through mathematical transformation, and retain the key information in the standard image data;

[0133] The standard image data in the low-dimensional subspace is the data after feature dimensionality reduction;

[0134] The standard image data after dimensionality reduction is trained through a neural network model, and label mapping is performed after the training is completed. After label mapping, the standard image data after image recognition is obtained and marked as evaluation image data.

[0135] Specifically, through feature dimensionality reduction, high-dimensional standard image data can be mapped to a low-dimensional subspace, which greatly reduces the dimension of the data, thereby reducing the computational complexity in subsequent image recognition tasks. This can not only improve the processing speed, but also reduce the consumption of computing resources. In the process of feature dimensionality reduction, by selecting features with higher weight values ​​and performing mathematical transformations, key information can be retained. In this way, while reducing the dimension, the impact on the accuracy of image recognition can be minimized. Using a neural network model to train the reduced-dimensional standard image data can improve the accuracy of image recognition by learning the inherent laws and characteristics of the data. Through label mapping, the recognition results can be matched with specific labels, which is convenient for subsequent data analysis and application. At the same time, marking the recognized image data as evaluation image data helps to quantitatively evaluate the recognition results and further improve the quality and reliability of image recognition.

[0136] In order to solve the problem in the prior art that the final image evaluation and solution formulation are not performed after the image data fusion is completed, which leads to the reduction of image data processing efficiency, please refer to Figure 1 and Figure 2 , this embodiment provides the following technical solutions:

[0137] An image recognition method based on multi-source data fusion, comprising:

[0138] Image contrast evaluation module for:

[0139] Acquiring parameters for evaluating the image data, wherein the parameters for evaluating the image data include a contrast parameter, a brightness parameter, a saturation parameter, a hue parameter, an entropy parameter, and a clarity parameter;

[0140] Comparing parameters of the evaluation image data with parameters of qualified image data, wherein the qualified image data and the parameters of the qualified image data are retrieved from a database;

[0141] Judging and evaluating the quality level of the image data according to the comparison result;

[0142] Among them, the quality levels are divided into first-level quality, second-level quality and third-level quality.

[0143] Evaluation and adjustment plan decision module, used to:

[0144] Formulate an adjustment plan for the evaluation image data of each quality level according to the quality level of the evaluation image data;

[0145] The adjustment plan is sent to the display terminal for display.

[0146] Specifically, by obtaining multiple parameters of the evaluation image data, a comprehensive and objective evaluation of the image quality is achieved. This parameterized evaluation method is more accurate and reliable than the traditional subjective evaluation, and can reflect various aspects of image quality more finely. By comparing with the parameters of qualified image data, the difference between the evaluation image data and the standard can be clearly understood, and the image quality can be divided into different levels, which is helpful for hierarchical management and optimization of image quality. Different processing strategies can be adopted for images of different quality levels, thereby improving processing efficiency and effect. The evaluation adjustment plan decision module can formulate an adjustment plan according to the quality level of the evaluation image data and send it to the display terminal for display. This intelligent decision-making method not only improves the processing speed, but also ensures the pertinence and effectiveness of the adjustment plan. By displaying the adjustment plan on the display terminal, users can intuitively understand how to improve image quality. This helps to enhance the user experience and improve user satisfaction with the image processing system.

[0147] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0148] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An image recognition method based on multi-source data fusion, characterized in that: The steps include: S1: Image data acquisition and classification: image data is acquired from different data sources, and the acquired image data is uniquely coded and labeled; S2: Obtain image preprocessing: perform data preprocessing on each image data with a unique coding label, and mark the image data after data preprocessing as target image data; S3: Processing image fusion: integrating the target image data and extracting key features from the integrated image to obtain standard image data; S4: Fusion image processing and recognition: optimizing the standard image data, and performing image recognition on the optimized standard image data, and obtaining evaluation image data after image recognition; S5: Identification image evaluation: The image data is evaluated for image quality, and an adjustment plan is formulated based on the quality evaluation results.

2. The image recognition method based on multi-source data fusion according to claim 1, characterized in that: In S1, image data is obtained from different data sources, and the obtained image data are uniquely coded and labeled, including: Acquire image data of different illumination and angles from different terminals and sensors; The image data is uniquely coded and labeled using the database's self-incrementing ID and timestamp; The acquired image data is stored in a database, and after storage is completed, a unique ID is assigned through the database's self-incrementing ID; Timestamping the image data stored in the database and assigned with a unique ID; Generate a time label for the image data to which the unique ID has been assigned in the database according to the timestamp; The unique coding label of the image data is obtained based on the time label and the assigned unique ID.

3. The image recognition method based on multi-source data fusion according to claim 2, characterized in that: Data preprocessing is performed on each image data with a unique coding number in S2, including: Perform data preprocessing on the image data with unique coding labels; The data preprocessing steps are as follows: first, the image data is loaded through the image processing library, and after the image loading is completed, the image data is unified in format, type, and size; The unified image data is then normalized, wherein the normalization is to scale the pixel values ​​of the image data to within a preset range, and the preset range is 0 to 1 or -1 to 1.

4. The image recognition method based on multi-source data fusion according to claim 3 is characterized in that: The data preprocessing is performed on each image data with a unique coding number in S2, and further includes: The normalized image data is denoised through a filter; The denoised image data is adjusted for contrast through histogram equalization; Perform data enhancement on the image data after contrast adjustment; After data enhancement, the preprocessed image data is obtained; And mark the image data after preprocessing as target image data.

5. The image recognition method based on multi-source data fusion according to claim 4 is characterized in that: Perform image integration on the target image data in S3, including: First, an integrated background image is selected from the data, wherein the integrated background image is a container of the target image data, and the size of the integrated background image is larger than the size of the target image data, and the margin size of the integrated background image is consistent; Confirm the position of the target image data on the integrated background image; After the position of the target image data is confirmed, the location and size attributes of each target image data are confirmed; Create image datasets of target image data respectively according to positioning and size attributes.

6. The image recognition method based on multi-source data fusion according to claim 5, characterized in that: Extract key features of the images to be integrated in S3, including: Each image in the image data set is subjected to key point detection by an accelerated robust feature algorithm, wherein key point detection is to extract feature points in the image, and the feature points include corner points and edge points of the image; Generate a descriptor for each detected key point, where the descriptor is the local image information around the key point; Perform feature matching on the descriptors generated by the key points, wherein feature matching is to calculate the similarity measure between two descriptors and determine the matching key points according to the similarity threshold; After the key points of the image data set are matched, the key features of the target image data are obtained and marked as standard image data.

7. The image recognition method based on multi-source data fusion according to claim 6, characterized in that: Optimize the standard image data in S4, including: Optimizing the standard image data, wherein the image optimization includes processing the skewness, clarity, distortion, denoising, rendering and sharpening of the standard image data; First, the standard image data is processed in order of skewness, clarity, and distortion; Among them, the skewness threshold, clarity threshold, and distortion threshold of the standard image data are confirmed in turn; If the skewness threshold, clarity threshold, and distortion threshold are not within the standard threshold range, the threshold is adjusted until it is within the standard threshold range; After the skewness threshold, clarity threshold, and distortion threshold of the standard image data are adjusted, denoising, rendering, and sharpening are performed; The standard image data after threshold adjustment is filtered and denoised by using a Gaussian filter model until the denoising effect reaches a set range; Then, the denoised standard image data is rendered, the brightness value and grayscale level of the denoised standard image data are adjusted, and when the brightness value and grayscale level reach the white balance standard, the standard image data is color rendered; Weakening the color-rendered standard image data, confirming the foreground area and the background area of ​​the color-rendered standard image data, and then calculating the contrast between the foreground area and the background area; If the contrast is less than the set contrast threshold, the foreground area is enhanced; the edge features of the critical position between the foreground area and the background area are extracted, and the critical position width is identified based on the edge features. If the critical position width exceeds the set width threshold, the critical position width is symmetrically reduced on both sides to be equal to the width threshold, so as to obtain the standard image data after sharpening adjustment.

8. The image recognition method based on multi-source data fusion according to claim 7, characterized in that: Image recognition is performed on the optimized standard image data in S4, including: Performing feature dimensionality reduction on the standard image data, wherein the feature dimensionality reduction is performing feature analysis on the standard image data; The weight value of each analyzed feature is evaluated according to statistical tests, information gain and model weight indicators, and the features with the top three weight values ​​are selected to map the original features to a new low-dimensional subspace through mathematical transformation, and retain the key information in the standard image data; The standard image data in the low-dimensional subspace is the data after feature dimensionality reduction; The standard image data after dimensionality reduction is trained through a neural network model, and label mapping is performed after the training is completed. After label mapping, the standard image data after image recognition is obtained and marked as evaluation image data.

9. An image recognition system based on multi-source data fusion, applied in the image recognition method based on multi-source data fusion as claimed in claim 8, characterized in that: include: Image contrast evaluation module for: Acquiring parameters for evaluating the image data, wherein the parameters for evaluating the image data include a contrast parameter, a brightness parameter, a saturation parameter, a hue parameter, an entropy parameter, and a clarity parameter; Comparing parameters of the evaluation image data with parameters of qualified image data, wherein the qualified image data and the parameters of the qualified image data are retrieved from a database; Judging and evaluating the quality level of the image data according to the comparison result; Among them, the quality levels are divided into first-level quality, second-level quality and third-level quality.

10. The image recognition method based on multi-source data fusion according to claim 9, characterized in that: Said: Evaluation and adjustment plan decision module, used to: Formulate an adjustment plan for the evaluation image data of each quality level according to the quality level of the evaluation image data; The adjustment plan is sent to the display terminal for display.

Images