Ramie seed vitality evaluation method and system based on multispectral image analysis

Abstract

The invention relates to the technical field of ramie seed vitality evaluation methods, and more specifically, to a ramie seed vitality evaluation method based on multispectral image analysis and a system thereof, comprising the following steps: collecting multispectral images and photosynthetic autotrophic fluorescence images of ramie seeds to be tested; constructing a feature matrix of the multispectral fluorescence image to extract feature clustering centers based on a feature clustering decomposition algorithm to form spectral feature data of the ramie seeds; reducing the dimension of the feature matrix based on singular value decomposition to generate spectral principal components of the ramie seeds; projecting the spectral principal components to a low-dimensional feature matrix based on nonlinear embedding mapping; generating a spectral index matrix based on the low-dimensional feature matrix; inputting the spectral index matrix into a deep inversion prediction model to obtain the vitality index of the ramie seeds, and in the processing stage of the multispectral image data, effectively solving the information loss and overlap problems of the traditional technology in the dimensionality reduction process through feature clustering decomposition and singular value decomposition technology.

Description

Technical Field

[0001] The invention relates to the technical field of ramie seed vitality evaluation methods, and more specifically, to a ramie seed vitality evaluation method based on multispectral image analysis and a system thereof. Background Art

[0002] With the continuous development of agricultural technology, seed vigor assessment has become one of the key means to improve agricultural production efficiency, especially in the field of planting economic crops such as ramie. Accurate seed vigor assessment can help farmers screen out high-quality seeds at an early stage, thereby ensuring crop yield and quality. However, traditional ramie seed vigor assessment methods mainly rely on the detection of physical indicators, usually including simple biochemical analysis such as moisture content and material fullness. Although this type of method can preliminarily reflect the state of the seeds, it is difficult to accurately predict the germination potential and growth ability of the seeds fundamentally due to its relatively single detection method. In addition, the detection results of traditional methods usually rely on surface physical characteristics, which to a certain extent ignores the vitality potential inside the seeds, making the evaluation results lack accuracy.

[0003] At present, spectral analysis has been applied to seed quality analysis as a non-destructive detection method. In particular, multispectral images can provide spectral information of seeds in different bands, which helps to reveal the intrinsic physiological state of seeds. However, existing multispectral analysis methods mainly use linear feature extraction methods or simple cluster analysis. When processing complex spectral data of ramie seeds, they often face the problems of high data dimensions and overlapping features, which leads to certain limitations of multispectral image data in practical applications. In addition, traditional multispectral image analysis technology lacks a mechanism for processing data noise, which is prone to measurement errors due to impurities on the seed surface or uneven illumination, affecting the accuracy of vitality assessment.

[0004] In addition, although some existing technologies attempt to introduce simple machine learning models in the process of spectral feature processing, it is often difficult to fully and accurately capture the spectral characteristics of ramie seeds due to the relatively simple means of feature extraction and dimensionality reduction. For example, conventional linear dimensionality reduction methods such as principal component analysis (PCA) are limited to the processing of low-dimensional data. When applied to the processing of multi-spectral data, due to the high spectral dimension and complex features, the model prediction results often have problems of information loss or over-simplification, making the seed vitality prediction results unstable. In addition, the training data of traditional models are often limited to a single environment or sample set, lacking the ability to adaptively adjust across environments or multiple samples, and thus lacking generalization between different seed samples.

[0005] In this context, how to accurately extract the multispectral features of ramie seeds by introducing multi-level and multi-dimensional data processing and feature extraction algorithms, and effectively solve the problems of noise interference, feature overlap and dimensionality reduction distortion in traditional methods, has become an important technical problem in improving the vitality assessment of ramie seeds. Therefore, the innovation of ramie seed vitality assessment methods and systems based on multispectral image analysis is particularly necessary. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention proposes a ramie seed vitality evaluation method and system based on multispectral image analysis to solve the problems of noise interference, feature overlap and dimensionality reduction distortion existing in the traditional method.

[0007] The present invention provides a ramie seed vitality evaluation method based on multispectral image analysis, 1. The ramie seed vitality evaluation method based on multispectral image analysis is characterized in that it comprises the following steps:

[0008] Collecting multispectral images and photosynthetic autotrophic fluorescence images of ramie seeds to be tested, dividing the ramie seeds into a training set and a test set, and performing splicing and alignment after removing impurities to obtain the multispectral fluorescence images of the ramie seeds;

[0009] Constructing a feature matrix of the multi-spectral fluorescence image to extract feature cluster centers based on a feature clustering decomposition algorithm to form spectral feature data of the ramie seeds;

[0010] Performing dimension reduction on the characteristic matrix based on singular value decomposition to generate the main components of the spectrum of the ramie seeds;

[0011] Projecting the spectral principal components to a low-dimensional feature matrix based on a nonlinear embedding mapping;

[0012] Generate a spectral index matrix based on the low-dimensional feature matrix;

[0013] The spectral index matrix is ​​input into the deep inversion prediction model to obtain the vitality index of ramie seeds.

[0014] Preferably, the feature matrix generates feature cluster centers through a feature clustering decomposition algorithm, and the cluster centers The calculation formula is:

[0015]

[0016] in, For the ramie seeds The spectral channel eigenvalues, For the The center value of the class, is the number of categories.

[0017] Preferably, the feature matrix is ​​reconstructed by singular value decomposition. , the singular value decomposition formula is:

[0018]

[0019] in, is a left singular matrix, is the singular value matrix, is a right singular matrix, Dimensionality reduction is performed before extraction singular values ​​to generate the reconstruction matrix , forming the main component of the spectrum of the ramie seeds.

[0020] Preferably, the reconstruction matrix Projected to a low-dimensional feature matrix through a nonlinear embedding mapping function , the nonlinear embedding mapping formula is:

[0021]

[0022] in, is the activation function, and are the weight matrices of the first and second layers respectively, and is the bias vector.

[0023] Preferably, based on the low-dimensional feature matrix Generate spectral index matrix , the generation formula of the spectral index matrix is:

[0024]

[0025] in, Represents the low-dimensional feature matrix Middle The first principal component item,, and is the number of samples and the dimension of principal components, A fine-tuning term to prevent the denominator from being zero.

[0026] Preferably, the spectral index matrix generates the vitality index of ramie seeds through a deep inversion prediction model, and the vitality index includes relative carotenoid content, dry weight loss rate, vitality and fullness. The formula of the deep inversion prediction model is:

[0027]

[0028] in, is the prediction result vector, which contains various vitality indicators of ramie seeds. is the weight matrix, is the bias term, is an adaptive activation function.

[0029] Preferably, in the depth inversion prediction model, and The prediction accuracy of the vitality index is adaptively optimized by gradually updating according to the inversion error feedback.

[0030] The ramie seed vitality evaluation system based on multispectral image analysis includes: an acquisition module, an image processing module, a feature extraction module, a data processing module and a prediction module; wherein:

[0031] An acquisition module is used to acquire the multispectral image and the photosynthetic autotrophic fluorescence image of the ramie seeds, align and stitch them, and obtain the multispectral fluorescence image of the ramie seeds;

[0032] An image processing module, used for constructing a feature matrix of the multi-spectral fluorescence image, and generating a reconstruction matrix based on feature clustering decomposition and singular value decomposition;

[0033] A feature extraction module, used for generating the low-dimensional feature matrix through nonlinear embedding mapping;

[0034] A data processing module, used for generating a spectral index matrix based on the low-dimensional feature matrix;

[0035] A prediction module is used to output the vitality index of ramie seeds based on the spectral index matrix and the deep inversion prediction model.

[0036] Preferably, the acquisition module captures the multispectral image and photosynthetic autotrophic fluorescence image of the ramie seeds based on a dark box, and removes impurities to ensure image purity.

[0037] Preferably, the data processing module includes an adaptive normalization function for adjusting the fine-tuning term of the spectral index matrix The noise-resistant stability of the spectral index matrix is ​​achieved, and the result is input into the prediction module to ensure the accuracy and consistency of the ramie seed vitality evaluation result.

[0038] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0039] The present invention proposes a ramie seed vitality assessment method and system based on multispectral image analysis, aiming to form a low-dimensional spectral feature matrix of ramie seeds by constructing a multispectral image feature matrix of ramie seeds, combining feature clustering decomposition, singular value decomposition (SVD), multi-layer nonlinear embedding mapping and spectral index generation. Then, a deep inversion prediction model is used with low-dimensional spectral features as input to achieve high-precision prediction of seed carotenoid relative content, dry weight loss rate, vitality and fullness. In particular, the present invention solves a series of problems in the prior art due to high feature extraction dimensionality, noise interference and unstable prediction results, and provides a highly reliable and efficient method for ramie seed vitality assessment.

[0040] The present invention uses innovative algorithm design to form a complementary and synergistic overall method chain for feature extraction, data dimensionality reduction and prediction processes among the various method steps. Specifically, in the processing stage of multispectral image data, feature clustering decomposition and singular value decomposition techniques are used to effectively solve the information loss and overlap problems in the dimensionality reduction process of traditional technologies. Singular value decomposition is used to convert high-dimensional multispectral data into a low-dimensional principal component matrix, reducing redundant data and retaining key information of the multispectral characteristics of ramie seeds.

[0041] After data dimensionality reduction, the nonlinear characteristics of spectral data are further enhanced through nonlinear embedding mapping, so that the spectral index matrix can more accurately reflect the internal vitality state of ramie seeds, solving the limitations of traditional linear models in complex spectral data. In addition, the introduction of the adaptive normalization and deep inversion prediction model of the present invention enables the system to adaptively adjust the structure of the prediction model through real-time error feedback and parameter optimization mechanism, thereby improving the generalization ability in different environments and sample sets. Therefore, the entire data processing flow forms a good connection from input data to vitality index output, and through the complementarity and synergistic superposition of each algorithm step, an innovative ramie seed vitality evaluation system suitable for complex spectral data is constructed.

[0042] Through the above method, the present invention realizes efficient processing of multispectral image data and high-precision prediction of ramie seed vitality assessment. Specifically, the combination of feature clustering decomposition and singular value decomposition enables the spectral characteristics of ramie seeds to be fully extracted, and the noise of spectral data is effectively suppressed, ensuring the integrity and accuracy of the data. Nonlinear embedding mapping provides nonlinear support for the generation of spectral index, making the generation of spectral index matrix more accurate. In addition, the deep inversion prediction model in the present invention realizes the stable output of ramie seed carotenoid content, dry weight loss rate, vitality and fullness through adaptive activation function and weight dynamic adjustment mechanism.

[0043] In summary, the ramie seed vitality evaluation system of the present invention effectively overcomes the problems of noise influence, excessive feature dimension and poor model generalization ability existing in the existing methods, and realizes high-precision evaluation of ramie seed vitality with highly innovative multi-layer embedding and nonlinear optimization algorithms, which provides advanced screening and analysis tools for the cultivation of economic crops such as ramie, and has important practical value in agricultural production and seed quality detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a logic block diagram of the acquisition module and the image processing module of the present invention.

[0045] Figure 2 It is a logic block diagram of the feature matrix construction and feature extraction module of the present invention.

[0046] Figure 3 It is a logic block diagram of the data processing module of the present invention.

[0047] Figure 4 It is a logic block diagram of the prediction module of the present invention.

[0048] Figure 5 It is a logic block diagram of the overall structure of the system of the present invention. DETAILED DESCRIPTION

[0049] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects thereof are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0050] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0051] The following is a clear and complete description of the technical solution of the present invention in conjunction with specific implementation methods. Figure 1-5 The present invention provides a ramie seed vitality assessment method and system based on multispectral image analysis, which are specifically used in the field of agricultural planting to accurately assess the vitality of ramie seeds by extracting and analyzing the features of ramie seed multispectral images.

[0052] First, the present invention proposes a ramie seed vitality assessment method based on multispectral image analysis. The method starts with collecting multispectral images and photosynthetic autotrophic fluorescence images of ramie seeds to be tested, dividing the collected ramie seeds into a training set and a test set, and removing impurities from the samples. Then, the multispectral fluorescence images are obtained by shooting and aligning in a dark box, and a spectral feature data matrix is ​​constructed. This process extracts the feature clustering center of the ramie seed spectral data through a feature clustering decomposition algorithm. The feature matrix is ​​then reduced in dimension by singular value decomposition to form a spectral principal component, which is projected to a low-dimensional feature matrix using a nonlinear embedding mapping function, and a spectral index matrix is ​​generated based on the low-dimensional feature matrix. Finally, the spectral index matrix is ​​input into a deep inversion prediction model to obtain the vitality index of the ramie seeds. The specific operation process and the functional modules are interconnected in the present method to form a complete closed loop from data input to result output.

[0053] As a preferred embodiment, during the operation of the ramie seed vitality assessment method, an image data processing module 1, a feature matrix extraction module 2, a low-dimensional mapping module 3, a spectral index generation module 4 and a depth inversion prediction module 5 are involved. The synergistic effect of each module ensures the efficiency and accuracy of the vitality assessment.

[0054] In an embodiment of the present invention, a feature clustering decomposition algorithm is used to extract the spectral features of ramie seed images, and a feature matrix is ​​generated through the feature clustering decomposition algorithm to generate feature cluster centers. The calculation formula is:

[0055]

[0056] in, For the Spectral channel The cluster center of the eigenvalues, Indicates the first The first eigenvalues, For the The center value of the class, Represents the number of categories. In the process of generating the feature matrix, the present invention preferably clusters the categories The value range is 3 to 10 to ensure the adaptability and accuracy of feature clustering. The specific value can be flexibly adjusted according to the actual spectral differences of the multispectral images of ramie seeds. This algorithm helps to improve the classification accuracy of the model and the accuracy of seed vitality judgment by extracting the spectral features of each channel of ramie seeds.

[0057] As a preferred embodiment, the reconstruction matrix is ​​further constructed by performing singular value decomposition (SVD) on the feature matrix , so that the main components of the spectrum of ramie seeds can be extracted. The singular value decomposition formula is:

[0058]

[0059] in, is a left singular matrix, is the singular value matrix, is a right singular matrix, representing different matrix decompositions. After dimensionality reduction, it is preferred to extract the first 3-5 singular values ​​to form a reconstructed matrix after dimensionality reduction , in order to preserve the main information of multispectral data and avoid redundancy. This method not only retains the spectral information of the image, but also simplifies the complexity of subsequent calculations, which helps to improve the efficiency of data analysis.

[0060] As a preferred embodiment, the nonlinear embedding mapping reconstructs the matrix Projection to low-dimensional feature matrix , the mapping formula is as follows:

[0061]

[0062] in, is the activation function, preferably the ReLU activation function; and is the weight matrix, representing the weight coefficient of each layer; and are bias vectors, which are used to adjust the mapping relationship between each layer. Through the nonlinear embedding operation of the low-dimensional mapping function, the spectral feature dimension of ramie seeds can be effectively reduced, and the stability and robustness of the vitality index evaluation can be improved.

[0063] In an embodiment of the present invention, it is proposed to use a low-dimensional feature matrix Generate spectral index matrix , the formula is as follows:

[0064]

[0065] in, Represents the low-dimensional feature matrix Middle The first principal component item, To prevent the fine-tuning item from having a denominator of zero, the preferred value range is to The spectral characteristic index of ramie seeds was generated through the spectral index matrix, which ensured the accuracy of subsequent data analysis and the index matrix could maintain high stability under noisy data, thus improving the overall reliability of the evaluation.

[0066] As a preferred embodiment, it is proposed to convert the spectral index matrix The deep inversion prediction model was input to generate the vitality index of ramie seeds, including relative carotenoid content, dry weight loss rate, vitality and fullness. The model structure is defined as:

[0067]

[0068] in, is an output result vector, representing the vitality index of the ramie seeds; is the weight matrix of the prediction model, is the bias term, It is an adaptive activation function to achieve nonlinear characteristic response of vitality evaluation. The weight adjustment ensures that the model can accurately output the required indicators and has dynamic response capabilities.

[0069] As a preferred embodiment, the optimization method of the depth inversion prediction model is to gradually update the inversion error feedback. and , thus having adaptive adjustment function when predicting the vitality index of ramie seeds, improving the prediction accuracy of the model under different seed vitality conditions, especially in the data analysis of diversified seed characteristics.

[0070] The present invention also includes a ramie seed vitality evaluation system based on multispectral image analysis, and the system specifically includes an acquisition module 1, an image processing module 2, a feature extraction module 3, a data processing module 4, and a prediction module 5. The acquisition module 1 is used to acquire the multispectral image and the photosynthetic autotrophic fluorescence image of the ramie seeds, align and splice them, and obtain the multispectral fluorescence image of the ramie seeds; the image processing module 2 is used to construct the feature matrix of the multispectral fluorescence image, and generate a reconstruction matrix based on feature clustering decomposition and singular value decomposition; the feature extraction module 3 is used to generate the low-dimensional feature matrix through nonlinear embedding mapping; the data processing module 4 is used to generate a spectral index matrix based on the low-dimensional feature matrix, and input it into the prediction module 5 to obtain the vitality index of the ramie seeds; the prediction module 5 is used to output the vitality index of the ramie seeds based on the spectral index matrix and the deep inversion prediction model.

[0071] As a preferred embodiment, the acquisition module 1 takes the multispectral image and the photosynthetic autotrophic fluorescence image of the ramie seeds based on a dark box, and removes impurities to ensure the purity of the image.

[0072] As a preferred embodiment, the data processing module includes an adaptive normalization function for adjusting the fine-tuning term of the spectral index matrix The noise-resistant stability of the spectral index matrix is ​​achieved, and the result is input into the prediction module to ensure the accuracy and consistency of the ramie seed vitality evaluation result.

[0073] The embodiments and comparative examples of the present invention are based on multispectral image data of ramie seeds, and a professional data set containing a large number of ramie seed samples is specifically selected. The superiority of the multispectral image analysis method of the present invention is verified by detecting the vitality index of ramie seeds. The detection methods and results of the embodiments and comparative examples are described in detail below, and the technical advantages of each result are analyzed through the evaluation of multiple vitality indexes.

[0074] The embodiment of the present invention is based on multispectral image acquisition and feature analysis of ramie seeds, and utilizes feature clustering decomposition, singular value decomposition, nonlinear embedding mapping and spectral index generation methods to input the extracted spectral index matrix into a deep inversion prediction model to generate vitality indicators, specifically including the relative content of carotenoids, dry weight loss rate, vitality and fullness of ramie seeds.

[0075] The comparative example selects the traditional ramie seed vitality analysis method, performs simple clustering and regression analysis with conventional spectral data, does not use the algorithms such as multi-layer embedding mapping and adaptive activation function in the present invention, and only calculates the vitality index of ramie seeds based on basic features and statistical methods. The data processing flow of the comparative example is relatively simple, lacking complex low-dimensional feature extraction and spectral index matrix generation methods.

[0076] In this comparative test, each vitality index is tested using a standardized testing method to more clearly show the difference between the embodiment and the comparative example:

[0077] 1. Relative content of carotenoids: The carotenoid content was measured by high performance liquid chromatography and expressed as a percentage of the total dry weight of seeds.

[0078] 2. Dry weight loss rate: The life activity of the seeds can be reflected by accurately measuring the dry weight loss rate of the ramie seeds after treatment.

[0079] 3. Vitality: Using the seed germination experiment, the vitality is expressed as a percentage of the germination rate, and the average is taken after three replicates.

[0080] 4. Fullness: This is measured by physically measuring the fullness of the seeds to characterize their inherent vitality potential, and is calculated as a ratio to the overall density of the seeds.

[0081] These indicators can fully demonstrate the vitality of ramie seeds and verify the effectiveness and superiority of the algorithm used in the embodiment.

[0082] The test results of the embodiments of the present invention and the comparative examples are summarized in the following table:

[0083]

[0084] From the above test results, it can be seen that the embodiments of the present invention are superior to the comparative examples in various ramie seed vitality indicators. The specific analysis is as follows:

[0085] 1. Relative content of carotenoids: 8.2% in the embodiment, while only 6.7% in the comparative example. The present invention accurately extracts the spectral characteristics of ramie seeds through a deep inversion prediction model, and uses nonlinear embedding mapping to make the spectral characteristics more complete and accurate, reflecting the higher relative content of carotenoids in ramie seeds. High carotenoid content indicates that the seeds have stronger vitality and photosynthesis potential.

[0086] 2. Dry weight loss rate: The example is significantly lower than the comparative example, which are 12.1% and 16.3% respectively. The low dry weight loss rate indicates that the seeds have higher water retention capacity and structural stability. The adaptive normalization operation and spectral index matrix generation algorithm of the present invention effectively reduce the impact of data noise, making the reflection of the structural state of ramie seeds more realistic. This index reflects the advantages of the present invention in seed dryness tolerance and activity retention.

[0087] 3. Vitality (germination rate): The vitality of the embodiment reached 89.4%, which is much higher than 74.6% of the comparative example. The present invention uses multi-layer embedding mapping to accurately extract the low-dimensional spectral features of the seeds, so that the deep inversion prediction model can make a more accurate estimate of the vitality. The high germination rate shows that the evaluation method of the present invention is more practical and accurate in the analysis of ramie seed vitality.

[0088] 4. Fullness: The fullness of the embodiment is 0.89, while that of the comparative example is 0.76. The spectral index generation method of the present invention accurately analyzes the internal fullness state of the seeds, avoids the omission of the spectral characteristics of the seeds, and enables the true fullness of the seeds to be reflected. This index demonstrates the in-depth analysis capability of the present invention on the internal structure of the seeds during the vitality assessment process.

[0089] Through the detection and comparison of ramie seed vitality indicators, it can be seen that the multispectral image analysis method of the present invention has significant advantages. Especially in terms of carotenoid content, dry weight loss rate, vitality and fullness, compared with the comparative example, the technical solution provided by the present invention can more comprehensively and accurately evaluate the vitality of ramie seeds. These data fully verify the practicality and innovation of the present invention in the detection of ramie seed vitality.

[0090] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for evaluating ramie seed vitality based on multispectral image analysis, characterized in that: The following steps are involved: Collecting multispectral images and photosynthetic autotrophic fluorescence images of ramie seeds to be tested, dividing the ramie seeds into a training set and a test set, and performing splicing and alignment after removing impurities to obtain the multispectral fluorescence images of the ramie seeds; Constructing a feature matrix of the multi-spectral fluorescence image to extract feature cluster centers based on a feature clustering decomposition algorithm to form spectral feature data of the ramie seeds; Performing dimension reduction on the characteristic matrix based on singular value decomposition to generate the main components of the spectrum of the ramie seeds; Projecting the spectral principal components to a low-dimensional feature matrix based on a nonlinear embedding mapping; Generate a spectral index matrix based on the low-dimensional feature matrix; Inputting the spectral index matrix into a deep inversion prediction model to obtain the vitality index of ramie seeds; The feature matrix generates feature cluster centers through the feature clustering decomposition algorithm, and the cluster centers C(f ij ) is calculated as: Among them, f ij is the jth eigenvalue of the ith spectral channel of the ramie seeds, μ k is the center value of the kth class, K is the number of classes; The characteristic matrix generates a reconstruction matrix F' through singular value decomposition, and the singular value decomposition formula is: F=U∑V T Wherein, U is a left singular matrix, Σ is a singular value matrix, V is a right singular matrix, Σ is subjected to dimensionality reduction processing, and the first m1 singular values ​​are extracted to generate a reconstruction matrix F′, thereby forming the main components of the spectrum of the ramie seeds; The reconstruction matrix F′ is projected to the low-dimensional feature matrix Z through a nonlinear embedding mapping function, and the nonlinear embedding mapping formula is: Z=φ(F′)=σ(W2·σ(W1·F′+b1)+b2) Among them, σ is the activation function, W1 and W2 are the weight matrices of the first and second layers respectively, and b1 and b2 are bias vectors; Generate a spectral index matrix I based on the low-dimensional feature matrix Z spec , the generation formula of the spectral index matrix is: Among them, Z i1,j1 represents the j1th item of the i1th principal component in the low-dimensional feature matrix Z, m and n are the number of samples and the dimension of the principal component, and ∈ is a fine-tuning item to prevent the denominator from being zero; The spectral index matrix generates the vitality index of ramie seeds through the deep inversion prediction model, and the vitality index includes the relative content of carotenoids, dry weight loss rate, vitality and fullness. The formula of the deep inversion prediction model is: in, is the prediction result vector, including various vitality indicators of ramie seeds, W y is the weight matrix, b y is the bias term, and σ is the adaptive activation function.

2. The method for evaluating ramie seed vitality according to claim 1, characterized in that: In the depth inversion prediction model, W y and b y′ The prediction accuracy of the vitality index is adaptively optimized by gradually updating according to the inversion error feedback.

3. A ramie seed vitality assessment system based on multispectral image analysis that implements the method according to any one of claims 1 to 2, characterized in that: include: Acquisition module, image processing module, feature extraction module, data processing module and prediction module; wherein: An acquisition module is used to acquire the multispectral image and the photosynthetic autotrophic fluorescence image of the ramie seeds, align and stitch them, and obtain the multispectral fluorescence image of the ramie seeds; An image processing module, used for constructing a feature matrix of the multi-spectral fluorescence image, and generating a reconstruction matrix based on feature clustering decomposition and singular value decomposition; A feature extraction module, used for generating the low-dimensional feature matrix through nonlinear embedding mapping; A data processing module, used for generating a spectral index matrix based on the low-dimensional feature matrix; A prediction module is used to output the vitality index of ramie seeds based on the spectral index matrix and the deep inversion prediction model.

4. The ramie seed vitality evaluation system according to claim 3, characterized in that: The acquisition module takes multispectral images and photosynthetic autotrophic fluorescence images of the ramie seeds based on a dark box, and removes impurities to ensure image purity.

5. The ramie seed vitality evaluation system according to claim 3, characterized in that: The data processing module includes an adaptive normalization function for achieving the anti-noise stability of the spectral index matrix by adjusting the fine-tuning term ∈ of the spectral index matrix, and inputting the result into the prediction module to ensure the accuracy and consistency of the ramie seed vitality evaluation result.

Images