A method for automatic data analysis based on image analysis

Abstract

The present invention discloses an automatic data analysis method based on image analysis, which belongs to the technical field of sample concentration analysis. The method first obtains historical sample images under fixed environmental information and the concentration label values ​​corresponding to the historical sample images, and then studies the correlation between the sample image features in the historical sample images and the concentration label values ​​corresponding to the historical sample images, thereby forming a data mapping rule. Finally, the sample concentration can be analyzed through the data mapping rule. The method combines image feature extraction and image feature analysis technology, can be applied to most samples whose color changes with concentration, improves the sample concentration detection efficiency, and can effectively reduce the sample concentration detection cost, and has a wide range of application scenarios.

Description

Technical Field

[0001] The present invention belongs to the technical field of sample concentration analysis, and in particular relates to an automatic data analysis method based on image analysis. Background Art

[0002] During sample use, concentration measurements are often required. Existing concentration meters are typically integrated, including concentration meters, pH (acidity / alkalinity) meters, ORP (oxidation-reduction potential) meters, conductivity meters, and online corrosion rate meters. However, integrating these meters is expensive and prohibitive for most users. Therefore, a low-cost and efficient concentration analysis method is urgently needed for most sample testing. Summary of the Invention

[0003] The present invention provides an automatic data analysis method based on image analysis, which is used to solve the problems of high cost and low efficiency of sample concentration analysis in the prior art.

[0004] An automatic data analysis method based on image analysis, comprising:

[0005] Obtaining historical sample images under fixed environmental information and concentration label values ​​corresponding to the historical sample images; wherein the fixed environmental information includes a fixed light source and a fixed placement position of the sample;

[0006] Extract the sample image features from each historical sample image, and construct a training data pair based on the sample image features corresponding to the historical sample image and the corresponding concentration label value;

[0007] Based on the training data pairs, the association between the sample image features and the corresponding concentration label values ​​is learned through a deep learning algorithm to obtain data mapping rules;

[0008] After obtaining the sample image to be analyzed under fixed environmental information and extracting the image features to be analyzed corresponding to the sample image to be analyzed, the image features to be analyzed are mapped according to data mapping rules to obtain the sample concentration analysis result.

[0009] Furthermore, obtaining historical sample images under fixed environmental information and concentration label values ​​corresponding to the historical sample images includes obtaining historical sample images with a preset static time under fixed environmental information and concentration label values ​​corresponding to the historical sample images.

[0010] Furthermore, sample image features are extracted from each historical sample image, and a training data pair is constructed based on the sample image features corresponding to the historical sample image and the corresponding concentration label value, including:

[0011] Extract the brightness value features, contrast features and gray value features of the historical sample image, and fuse the brightness value features, contrast features and gray value features to obtain the sample image features;

[0012] Traverse all sample images and use the sample image features as training data. Construct the expected output data through the concentration label values ​​corresponding to the sample image features to obtain the training data pairs.

[0013] Furthermore, the brightness value features, contrast features, and gray value features of the historical sample images are extracted, and after the brightness value features, contrast features, and gray value features are fused, the sample image features are obtained, including:

[0014] The brightness value features of the historical sample images are extracted as follows:

[0015] Bright(i,j)=α×R(i,j)+α×G(i,j)+α×B(i,j)

[0016] Where Bright(i, j) represents the brightness value feature of pixel (i, j), R(i, j) represents the red value of pixel (i, j) in the historical sample image, G(i, j) represents the green value of pixel (i, j) in the historical sample image, B(i, j) represents the blue value of pixel (i, j) in the historical sample image, α represents the brightness adjustment coefficient, and pixel (i, j) represents the pixel in the i-th row and j-th column in the historical sample image;

[0017] The contrast features of the historical sample images are extracted as follows:

[0018] contrast(i,j)=(max(max(R(i,j),G(i,j),B(i,j)))-min(min(R(i,j),G(i,j),B(i,j))) / (max(max(R(i,

[0019] j),G(i,j),B(i,j)))+min(min(R(i,j),G(i,j),B(i,j)))

[0020] Among them, contrast(i, j) represents the contrast feature of pixel (i, j), max() represents the maximum value function, and min() represents the minimum value function;

[0021] The grayscale value features of the historical sample images are extracted as follows:

[0022] Gray(i,j)=(β1×R(i,j)+β2×G(i,j)+β3×B(i,j)) / 3

[0023] Wherein, Gray(i, j) represents the gray value feature of pixel (i, j), β1 represents the first weight coefficient, β2 represents the second weight coefficient, β3 represents the third weight coefficient, β1+β2+β3=1;

[0024] After fusing the brightness value features, contrast features, and grayscale value features, the sample image features are obtained.

[0025] Furthermore, after fusing the brightness value feature, contrast feature, and grayscale value feature, the sample image features are obtained, including:

[0026] The sample code corresponding to the sample corresponding to the historical sample image is obtained, and the sample image feature is obtained according to the sample code, brightness value feature, contrast feature and gray value feature.

[0027] Furthermore, obtaining historical sample images under fixed environmental information and concentration label values ​​corresponding to the historical sample images includes:

[0028] The system obtains fixed environmental information starting from the sample mixing time. When the preset static time is reached, multiple historical sample images corresponding to the same sample are collected at preset time intervals, and the concentration label values ​​corresponding to the historical sample images are obtained by manual marking. The concentration label values ​​corresponding to the multiple historical sample images corresponding to the same sample are the same.

[0029] Furthermore, sample image features are extracted from each historical sample image, and a training data pair is constructed based on the sample image features corresponding to the historical sample image and the corresponding concentration label value, including:

[0030] Extract the brightness value features, contrast features and gray value features of the historical sample image, and fuse the brightness value features, contrast features and gray value features to obtain the sample image features;

[0031] The sample image features corresponding to multiple historical sample images corresponding to the same sample are fused to obtain fused sample features, and the fused sample features are used as training data. The expected output data is constructed by the concentration label values ​​corresponding to the fused sample features to obtain training data pairs.

[0032] Furthermore, based on the training data pairs, the association between the sample image features and the corresponding concentration label values ​​is learned through a deep learning algorithm to obtain data mapping rules, including:

[0033] The initial data mapping rules are constructed through a deep learning algorithm, the training data in the training data pair is used as the input data of the initial data mapping rules, the expected output data in the training data pair is used as the output data of the initial data mapping rules, and the initial data mapping rules are optimized through a heuristic optimization algorithm to obtain the final data mapping rules.

[0034] Furthermore, after each time period, the historical sample images under fixed environmental information and the concentration label values ​​corresponding to the historical sample images are re-acquired, the data mapping rules are updated, and the updated data mapping rules are used for analysis in the subsequent sample concentration analysis process.

[0035] Furthermore, after obtaining the sample image to be analyzed under fixed environmental information and extracting the image features to be analyzed corresponding to the sample image to be analyzed, the image features to be analyzed are mapped using data mapping rules to obtain the sample concentration analysis results, including:

[0036] Obtain the sample image to be analyzed under fixed environmental information, extract the image features to be analyzed corresponding to the sample image to be analyzed, and then determine whether the number of obtained image features to be analyzed is greater than 1. If so, fuse multiple image features to be analyzed to obtain fused features to be analyzed, and analyze the fused features to be analyzed through data mapping rules to obtain sample concentration analysis results. Otherwise, directly analyze the image features to be analyzed through data mapping rules to obtain sample concentration analysis results.

[0037] The present invention provides an automatic data analysis method based on image analysis. The method first obtains historical sample images under fixed environmental information and the concentration label values ​​corresponding to the historical sample images, and then learns the correlation between the sample image features in the historical sample images and the concentration label values ​​corresponding to the historical sample images, thereby forming a data mapping rule. Finally, the sample concentration can be analyzed through the data mapping rule. The method combines image feature extraction and image feature analysis technology, can be applied to most samples whose color changes with concentration, improves the sample concentration detection efficiency, and can effectively reduce the sample concentration detection cost, and has a wide range of application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0039] Figure 1 The present invention provides a flowchart of an automatic data analysis method based on image analysis.

[0040] The above drawings illustrate specific embodiments of the present invention, which will be described in more detail below. These drawings and the accompanying description are not intended to limit the scope of the present invention in any way, but rather to illustrate the concept of the present invention to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0041] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent like or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.

[0042] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0043] like Figure 1 As shown, an embodiment of the present invention provides a method for automatic data analysis based on image analysis, comprising:

[0044] S1. Acquire historical sample images under fixed environment information and concentration label values ​​corresponding to the historical sample images; wherein the fixed environment information includes a fixed light source and a fixed placement position of the sample.

[0045] Fixed environmental information can be further interpreted as: except for changes in sample concentration or sample size, all other conditions remain unchanged, thereby improving the accuracy of sample concentration analysis. The concentration labels corresponding to historical sample images can be manually input. Before analysis begins, the relationship between historical sample images and their corresponding concentration labels is learned. Subsequent analysis can be performed based on the learned data mapping rules, effectively improving analysis efficiency.

[0046] Optionally, after acquiring the historical sample images, the historical sample images may be pre-processed to improve the accuracy of data recognition. For example, filtering may be performed on the historical sample image data.

[0047] S2. Extract the sample image features from each historical sample image, and construct a training data pair based on the sample image features corresponding to the historical sample image and the corresponding concentration label value.

[0048] By first extracting sample image features from historical sample images, constructing training data pairs, and finally learning the data rules contained in the training data pairs, the data recognition accuracy can be effectively improved.

[0049] S3. Based on the training data pair, the association relationship between the sample image features and the corresponding concentration label values ​​is learned through a deep learning algorithm to obtain data mapping rules.

[0050] Learning the association between sample image features and corresponding concentration label values ​​through a deep learning algorithm can include: taking the training data in the training data pair as input data, and taking the expected output data constructed by the concentration label values ​​in the training data pair as a reference, learning data rules through a deep learning algorithm, and thus obtaining data mapping rules.

[0051] S4. Obtaining a sample image to be analyzed under fixed environmental information, extracting the image features to be analyzed corresponding to the sample image to be analyzed, and then mapping the image features to be analyzed using data mapping rules to obtain a sample concentration analysis result.

[0052] Once the data mapping rules are established, the image of the sample to be analyzed can be acquired and recognized, allowing for rapid and cost-effective sample concentration analysis results. It is worth noting that to ensure accurate data recognition, the data format and data processing of the sample image to be analyzed are identical to those of the historical sample images.

[0053] The present invention provides an automatic data analysis method based on image analysis. The method first obtains historical sample images under fixed environmental information and the concentration label values ​​corresponding to the historical sample images, and then learns the correlation between the sample image features in the historical sample images and the concentration label values ​​corresponding to the historical sample images, thereby forming a data mapping rule. Finally, the sample concentration can be analyzed through the data mapping rule. The method combines image feature extraction and image feature analysis technology, can be applied to most samples whose color changes with concentration, improves the sample concentration detection efficiency, and can effectively reduce the sample concentration detection cost, and has a wide range of application scenarios.

[0054] In an embodiment of the present invention, obtaining historical sample images under fixed environmental information and concentration label values ​​corresponding to the historical sample images includes: obtaining historical sample images with a preset static time under fixed environmental information and concentration label values ​​corresponding to the historical sample images.

[0055] In an embodiment of the present invention, sample image features are extracted from each historical sample image, and a training data pair is constructed based on the sample image features corresponding to the historical sample image and the corresponding concentration label value, including:

[0056] Extract the brightness value features, contrast features and gray value features of the historical sample image, and fuse the brightness value features, contrast features and gray value features to obtain the sample image features;

[0057] Traverse all sample images and use the sample image features as training data. Construct the expected output data through the concentration label values ​​corresponding to the sample image features to obtain the training data pairs.

[0058] In an embodiment of the present invention, the brightness value features, contrast features, and gray value features of the historical sample image are extracted, and the brightness value features, contrast features, and gray value features are fused to obtain the sample image features, including:

[0059] The brightness value features of the historical sample images are extracted as follows:

[0060] Bright(i,j)=α×R(i,j)+α×G(i,j)+α×B(i,j)

[0061] Where Bright(i, j) represents the brightness value feature of pixel (i, j), R(i, j) represents the red value of pixel (i, j) in the historical sample image, G(i, j) represents the green value of pixel (i, j) in the historical sample image, B(i, j) represents the blue value of pixel (i, j) in the historical sample image, α represents the brightness adjustment coefficient, and pixel (i, j) represents the pixel in the i-th row and j-th column in the historical sample image;

[0062] The contrast features of the historical sample images are extracted as follows:

[0063] contrast(i,j)=(max(max(R(i,j),G(i,j),B(i,j)))-min(min(R(i,j),G(i,j),B(i,j))) / (max(max(R(i,

[0064] j),G(i,j),B(i,j)))+min(min(R(i,j),G(i,j),B(i,j)))

[0065] Among them, contrast(i, j) represents the contrast feature of pixel (i, j), max() represents the maximum value function, and min() represents the minimum value function;

[0066] The grayscale value features of the historical sample images are extracted as follows:

[0067] Gray(i,j)=(β1×R(i,j)+β2×G(i,j)+β3×B(i,j)) / 3

[0068] Wherein, Gray(i, j) represents the gray value feature of pixel (i, j), β1 represents the first weight coefficient, β2 represents the second weight coefficient, β3 represents the third weight coefficient, β1+β2+β3=1;

[0069] After fusing the brightness value features, contrast features, and grayscale value features, the sample image features are obtained.

[0070] Optionally, feature fusion can be performed by directly concatenating the three feature matrices to obtain a matrix with three times the original dimension. Alternatively, the three feature matrices can be transformed into a single feature matrix by weighted summation, while maintaining the original dimension. In this embodiment of the present invention, the three feature matrices are preferably concatenated into a single feature matrix.

[0071] By extracting brightness, contrast, and grayscale features, the amount of data processing can be greatly reduced, and the image can be effectively expressed, thereby achieving accurate image analysis. It is worth noting that normalization methods can be used to further reduce the amount of data.

[0072] In the embodiment of the present invention, after the brightness value feature, the contrast feature, and the gray value feature are fused, the sample image feature is obtained, including:

[0073] The sample code corresponding to the sample corresponding to the historical sample image is obtained, and the sample image feature is obtained according to the sample code, brightness value feature, contrast feature and gray value feature.

[0074] Because different samples or samples may affect the results of image analysis, the embodiment of the present invention adds the sample code as a feature value, adds it to the feature matrix obtained by feature fusion, and fills in the missing parts with zeros. It is worth noting that the sample referred to in the embodiment of the present invention is a liquid and colored sample (such as a liquid and colored medicine). When the sample image to be analyzed is subsequently processed, it is necessary to follow the same process as the historical sample image to ensure accurate data recognition.

[0075] In an embodiment of the present invention, obtaining historical sample images under fixed environmental information and concentration label values ​​corresponding to the historical sample images includes:

[0076] The system obtains fixed environmental information starting from the sample mixing time. When the preset static time is reached, multiple historical sample images corresponding to the same sample are collected at preset time intervals, and the concentration label values ​​corresponding to the historical sample images are obtained by manual marking. The concentration label values ​​corresponding to the multiple historical sample images corresponding to the same sample are the same.

[0077] In an embodiment of the present invention, sample image features are extracted from each historical sample image, and a training data pair is constructed based on the sample image features corresponding to the historical sample image and the corresponding concentration label value, including:

[0078] Extract the brightness value features, contrast features and gray value features of the historical sample image, and fuse the brightness value features, contrast features and gray value features to obtain the sample image features;

[0079] The sample image features corresponding to multiple historical sample images corresponding to the same sample are fused to obtain fused sample features, and the fused sample features are used as training data. The expected output data is constructed by the concentration label values ​​corresponding to the fused sample features to obtain training data pairs.

[0080] In an embodiment of the present invention, based on the training data pair, a deep learning algorithm is used to learn the association between the sample image features and the corresponding concentration label values ​​to obtain data mapping rules, including:

[0081] The initial data mapping rules are constructed through a deep learning algorithm, the training data in the training data pair is used as the input data of the initial data mapping rules, the expected output data in the training data pair is used as the output data of the initial data mapping rules, and the initial data mapping rules are optimized through a heuristic optimization algorithm to obtain the final data mapping rules.

[0082] For the same concentration of the same sample, if only one historical sample image is obtained after the sample stabilizes (that is, the static time reaches the preset static time), the initial data mapping rules can be constructed through the VGG16 neural network, and then the heuristic optimization algorithm can be used to learn the correlation between the training data and the expected output data in the training data pair to achieve the optimization of the initial data mapping rules.

[0083] For the same concentration of the same sample, if historical sample images are obtained at preset time intervals starting from sample mixing or dilution until the sample stabilizes, and multiple historical sample images are obtained, a VGG16 neural network for learning spatial features and an AlexNet neural network for learning temporal features can be constructed. After learning the correlation between the training data and the expected output data in the training data pair through a heuristic optimization algorithm, the output of the VGG16 neural network and the output of the AlexNet neural network are merged using a weighted summation method to obtain the final data mapping rule.

[0084] Multiple different images at different time points have not only spatial features but also temporal features. The embodiment of the present invention can effectively improve the sample concentration analysis effect by integrating the recognition of these two features.

[0085] Optionally, an embodiment of the present invention provides a heuristic optimization algorithm to optimize the initial data mapping rule, which may include:

[0086] The initial data mapping rule is constructed through a deep learning algorithm, and the parameters in the initial data mapping rule are randomly initialized to obtain a parameter vector. Multiple parameter vectors are repeatedly obtained to form a data solution space.

[0087] Obtain the loss function value corresponding to each parameter vector, and take the parameter vector with the smallest loss function value as the optimal parameter vector; there are many loss functions for neural networks (such as root mean square loss), and the existing technology is relatively complete, so the embodiments of the present invention will not be introduced in detail.

[0088] For each parameter vector, the parameter vector is updated by the optimal parameter vector guided update method to achieve local optimization, which can effectively explore the data solution space;

[0089] Updating the parameter vector by guiding the update through the optimal parameter vector may include:

[0090] The updated second optimized population includes:

[0091]

[0092] in, Indicates the d-th dimension parameter of the i-th parameter vector in the t-th training process, i=1,2,…I, I represents the total number of parameter vectors, d=1,2,…D, D represents the total dimension of the parameters in the parameter vector, rand represents a random number between (0,1), Indicates updated Represents the d-th dimension parameter in the optimal parameter vector.

[0093] For the parameter vector after guidance update, the hybrid search update method is used to update the parameter vector, which can effectively improve the speed and accuracy of data search.

[0094] Updating the parameter vector using a hybrid search update method may include:

[0095]

[0096]

[0097] in, represents the parameter vector after the j-th guided update, j = 1, 2, ... I, Indicates after update δ represents the mixing factor, represents the d-th dimension parameter of the target individual with the farthest Euclidean distance from the parameter vector after the j-th boot update, Represents the d-th dimension parameter in the average vector corresponding to all guided updated parameter vectors, δmax Indicates the preset maximum value of the mixing factor, δ min It represents the preset minimum value of the mixing factor, e represents the natural constant, and b represents the constant control factor.

[0098] By selecting the target individual with the farthest Euclidean distance for the j-th guided updated parameter vector, excessive aggregation of parameter vectors in the solution space can be avoided. Combined with the mixing factor that changes with the number of training times, the probability of global search can be increased in the early stage of the algorithm to avoid falling into local optimality. As the algorithm iterates, all parameter vectors are highly aggregated in the data solution space, which can effectively increase the search accuracy and thus improve the parameter search effect.

[0099] For the parameter vector after hybrid search, the global search method is used to update the parameter vector, which can effectively get rid of the local optimum and achieve better parameter search effect.

[0100] Updating the parameter vector using a global search method may include:

[0101]

[0102] in, Represents the d-th dimension parameter in the parameter vector after the n-th hybrid search, n=1,2,…I, Indicates updated Levy(λ) represents the Levy flight path, Represents the d-th dimension parameter in the optimal parameter vector.

[0103] Optionally, after each update of the parameter vector, the optimal parameter vector may be updated to improve the convergence speed of the algorithm.

[0104] When the number of training times reaches the maximum number of training times, the above updating steps are repeated until the maximum number of training times is reached, and the optimal parameter vector is output as the final parameter of the data mapping rule to obtain the final data mapping rule.

[0105] In an embodiment of the present invention, after each time period, the historical sample images under fixed environmental information and the concentration label values ​​corresponding to the historical sample images are re-acquired, the data mapping rules are updated, and the updated data mapping rules are used for analysis in the subsequent sample concentration analysis process.

[0106] In an embodiment of the present invention, after obtaining a sample image to be analyzed under fixed environmental information and extracting the image features to be analyzed corresponding to the sample image to be analyzed, the image features to be analyzed are mapped using data mapping rules to obtain a sample concentration analysis result, including:

[0107] Obtain the sample image to be analyzed under fixed environmental information, extract the image features to be analyzed corresponding to the sample image to be analyzed, and then determine whether the number of obtained image features to be analyzed is greater than 1. If so, fuse multiple image features to be analyzed to obtain fused features to be analyzed, and analyze the fused features to be analyzed through data mapping rules to obtain sample concentration analysis results. Otherwise, directly analyze the image features to be analyzed through data mapping rules to obtain sample concentration analysis results.

[0108] Optionally, a data mapping rule for identifying a single image feature and a data mapping rule for identifying multiple image features may be constructed. After identification using the two rules, the predicted probabilities are weighted and summed, thereby further increasing the recognition rate.

[0109] It will be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention may be implemented in various computer languages, for example, the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0110] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0111] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0112] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0113] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0114] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A data automatic analysis method based on image analysis, characterized in that: include: Obtain historical sample images under fixed environmental information and the concentration label values ​​corresponding to the historical sample images, including: Under fixed environmental information, starting from the sample mixing time, when the preset static time is reached, multiple historical sample images corresponding to the same sample are collected at preset time intervals, and the concentration label values ​​corresponding to the historical sample images are obtained by manual labeling. The concentration label values ​​corresponding to the multiple historical sample images corresponding to the same sample are the same; Wherein, the fixed environment information includes a fixed light source and a fixed placement position of the sample; Extract the sample image features from each historical sample image, and construct a training data pair based on the sample image features corresponding to the historical sample image and the corresponding concentration label value, including: Extract the brightness value features, contrast features and gray value features of the historical sample image, and fuse the brightness value features, contrast features and gray value features to obtain the sample image features; The sample image features corresponding to multiple historical sample images corresponding to the same sample are fused to obtain fused sample features, and the fused sample features are used as training data. The expected output data is constructed by the concentration label values ​​corresponding to the fused sample features to obtain training data pairs; Based on the training data pairs, the association between the sample image features and the corresponding concentration label values ​​is learned through a deep learning algorithm to obtain data mapping rules; Obtaining a sample image to be analyzed under fixed environmental information, extracting the image features to be analyzed corresponding to the sample image to be analyzed, and then mapping the image features to be analyzed using data mapping rules to obtain a sample concentration analysis result; The brightness value features, contrast features, and gray value features of the historical sample images are extracted, and after the brightness value features, contrast features, and gray value features are fused, the sample image features are obtained, including: The brightness value features of the historical sample images are extracted as follows: Bright ( i , j )= α ×R( i , j )+ α ×G( i , j )+ α ×B( i , j ) in, Bright ( i , j ) represents pixel points ( i , j ) brightness value features, R( i , j ) represents the pixel point in the historical sample image ( i , j ) red value, G( i , j ) represents the pixel point in the historical sample image ( i , j )’s green value, B( i , j ) represents the pixel point in the historical sample image ( i , j ), α Indicates the brightness adjustment coefficient, pixel point ( i , j ) represents the first i Rank j Column pixels; The contrast features of the historical sample images are extracted as follows: contrast ( i ， j )=( max ( max ( R ( i ， j ), G ( i ， j ), B ( i ， j )))- min ( min ( R ( i ， j ), G ( i ， j ), B ( i ， j ))) / ( max ( max ( R ( i ， j ), G ( i ， j ), B ( i ， j )))+ min ( min ( R ( i ， j ), G ( i ， j ), B ( i ， j ))) in, contrast ( i , j ) represents pixel points ( i , j )’s contrast characteristics, max () represents the maximum value function, min () represents the minimum value function; The grayscale value features of the historical sample images are extracted as follows: Gray( i , j )=( β 1*R( i , j )+ β 2*G( i , j )+ β 3*B( i , j )) / 3 Among them, Gray ( i , j ) represents pixel points ( i , j )’s grayscale value features, β 1 represents the first weight coefficient, β 2 represents the second weight coefficient, β 3 represents the third weight coefficient, β 1+ β 2+ β 3=1; After fusing the brightness value feature, contrast feature and gray value feature, the sample image feature is obtained; After fusing the brightness value features, contrast features, and grayscale value features, the sample image features are obtained, including: Obtaining a sample code corresponding to a sample corresponding to a historical sample image, and obtaining sample image features based on the sample code, brightness value features, contrast features, and grayscale value features; Based on the training data pairs, the association between the sample image features and the corresponding concentration label values ​​is learned through a deep learning algorithm to obtain data mapping rules, including: An initial data mapping rule is constructed through a deep learning algorithm. The training data in the training data pair is used as the input data of the initial data mapping rule, and the expected output data in the training data pair is used as the output data of the initial data mapping rule. The initial data mapping rule is optimized through a heuristic optimization algorithm to obtain the final data mapping rule. The initial data mapping rules are optimized through a heuristic optimization algorithm to obtain the final data mapping rules, including: Randomly initialize each parameter in the initial data mapping rule to obtain a parameter vector, and repeatedly obtain multiple parameter vectors to form a data solution space; The training data in the training data pair is used as the input data of the initial data mapping rule, the expected output data in the training data pair is used as the output data of the initial data mapping rule, the loss function value corresponding to each parameter vector is obtained, and the parameter vector with the smallest loss function value is used as the optimal parameter vector; For each parameter vector, the parameter vector is updated by the optimal parameter vector guided update method as follows: ， in, Indicates in t During the training i The parameter vector d dimensional parameters, i =1,2,…I, I represents the total number of parameter vectors, d =1,2,…D, where D represents the total dimension of the parameters in the parameter vector, rand Represents a random number between (0,1), Indicates updated , represents the optimal parameter vector d Dimensional parameters; For the parameter vector after the guidance update, the hybrid search update method is used to update the parameter vector as follows: ， in, Indicates the j The parameter vector after bootstrapping, j =1,2,…I, Indicates after update , represents the mixing factor, Indicates the j The target individual with the largest Euclidean distance between the parameter vectors after the guided update d dimensional parameters, Represents the average vector corresponding to all the parameter vectors after the bootstrap update. d dimensional parameters, Indicates the preset maximum value of the mixing factor, represents the preset minimum value of the mixing factor, e represents a natural constant, b represents the constant control factor; For the parameter vector after hybrid search, the global search method is used to update the parameter vector as follows: ， in, Indicates the n The parameter vector after hybrid search d dimensional parameters, n =1,2,…I, Indicates updated , represents the Levy flight path, represents the optimal parameter vector d Dimensional parameters; When the number of training times reaches the maximum number of training times, the above three updating steps are repeated until the maximum number of training times is reached, and the optimal parameter vector is output as the final parameter of the data mapping rule to obtain the final data mapping rule.

2. The automatic data analysis method based on image analysis according to claim 1, characterized in that: After each time period, the historical sample images under fixed environmental information and the concentration label values ​​corresponding to the historical sample images are re-acquired, the data mapping rules are updated, and the updated data mapping rules are used for analysis in the subsequent sample concentration analysis process.

3. The automatic data analysis method based on image analysis according to claim 2, characterized in that: After obtaining the sample image to be analyzed under fixed environmental information and extracting the image features to be analyzed corresponding to the sample image to be analyzed, the image features to be analyzed are mapped using data mapping rules to obtain the sample concentration analysis results, including: Obtain the sample image to be analyzed under fixed environmental information, extract the image features to be analyzed corresponding to the sample image to be analyzed, and then determine whether the number of obtained image features to be analyzed is greater than 1. If so, fuse multiple image features to be analyzed to obtain fused features to be analyzed, and analyze the fused features to be analyzed through data mapping rules to obtain sample concentration analysis results. Otherwise, directly analyze the image features to be analyzed through data mapping rules to obtain sample concentration analysis results.

Images