CN-122022545-A - Rhizoma polygonati processing after-forming evaluation system based on multi-scale fusion characteristics

CN122022545ACN 122022545 ACN122022545 ACN 122022545ACN-122022545-A

Abstract

The invention relates to the technical field of rhizoma polygonati processing quality detection, in particular to a rhizoma polygonati processed color forming evaluation system based on multi-scale fusion characteristics, which comprises front-end acquisition equipment, a server and a user terminal, the server comprises an image preprocessing module, a multi-scale feature extraction and fusion module, a deep learning evaluation module, a historical database module and a result output module. The method comprises the steps of collecting a sealwort section chart, carrying out color space calibration and effective area cutting on the sealwort section chart, extracting three types of scale characteristics through a convolution kernel with increasing size by a multi-scale characteristic extraction and fusion module, splicing to form a fusion characteristic vector after independent mapping, normalization and hierarchical coupling calibration, constructing a deep learning model in a deep learning evaluation module, training based on a sample set in a historical database module, outputting a quantized evaluation result, determining a color grade by a result output module, generating a report containing the evaluation result, the grade and the section chart after preprocessing, and pushing the report to a user terminal.

Inventors

LI JINFENG
ZHANG XIAOWEI
WANG XIAODIAN
HU SHIGANG

Assignees

遵义市农业科学研究院

Dates

Publication Date: 20260512
Application Date: 20251226

Claims (5)

1. The color evaluation system after processing of the rhizoma polygonati based on the multi-scale fusion characteristics is characterized by comprising front-end acquisition equipment, a server and a user terminal; the front-end acquisition equipment is used for acquiring a section chart of the rhizoma polygonati after steaming processing; The server comprises an image preprocessing module, a multi-scale feature extraction and fusion module, a history database module, a deep learning evaluation module and a result output module; The image preprocessing module is used for preprocessing the acquired sectional images, and the preprocessing sequentially comprises color space calibration and effective area cutting; The multi-scale feature extraction and fusion module is used for extracting multi-scale features from the preprocessed sectional view, wherein the multi-scale features comprise a first scale feature, a second scale feature and a third scale feature, the multi-scale features are extracted through convolution kernels of different sizes, the convolution kernels of the first scale feature, the second scale feature and the third scale feature are sequentially increased in size, the scale features are respectively unified to the same preset dimension through independent mapping, normalization is respectively carried out, coupling calibration of the first scale feature and the second scale feature is realized through a cross-scale interaction mechanism, and the coupled calibrated first scale feature, the coupled calibrated second scale feature and the original third scale feature are spliced to form a fusion feature vector; The historical database module is used for constructing a sample set, wherein the sample set comprises cross-sectional diagrams after steaming and processing of rhizoma polygonati with different color forming grades, corresponding fusion feature vectors and manually marked evaluation results, and the evaluation results are scores with values of [0,100 ]; The deep learning evaluation module is used for constructing a deep learning model and training based on a sample set, and outputting an evaluation result of the rhizoma polygonati by inputting the fusion feature vector into the trained deep learning model; And the result output module is used for determining the color forming grade according to the evaluation result, generating an evaluation report containing the evaluation result, the color forming grade and the preprocessed section chart, and pushing the evaluation report to the user terminal.
2. The system for evaluating color after processing of rhizoma Polygonati based on multi-scale fusion features according to claim 1, wherein the front-end acquisition device acquires standard color chart images synchronously when acquiring a cross-sectional image after processing of rhizoma Polygonati steaming, and the color space calibration of the image preprocessing module is realized by calculating a deviation matrix of actual color values and theoretical standard values in the standard color chart images and performing pixel-by-pixel color correction on the cross-sectional image based on the deviation matrix.
3. The system for evaluating the color after processing of the rhizoma polygonati based on the multi-scale fusion characteristics according to claim 2, wherein the effective area cutting implementation mode of the image preprocessing module is that an edge detection algorithm is adopted to identify the outline boundary of the section of the rhizoma polygonati, a geometric center is determined according to the outline boundary, and an area with a preset size is taken as an effective area by taking the geometric center as the center, so that a preprocessed sectional view is obtained.
4. The system for evaluating the processed color of Polygonatum sibiricum Red based on multi-scale fusion features according to claim 3, wherein the independent mapping and normalization process of the multi-scale feature extraction and fusion module comprises the steps of respectively accessing the initial feature vectors of each scale into an independent full-connection layer, mapping the initial feature vectors to the same preset dimension through linear transformation, and carrying out L2 normalization on the features of each scale after mapping.
5. The system for evaluating color after processing of rhizoma polygonati based on multi-scale fusion features, which is characterized by comprising the steps of calculating correlation of first scale features and second scale features to obtain a first local weight distribution matrix, multiplying the first scale features by the first local weight distribution matrix to realize coupling calibration of the first scale features, calculating correlation of the second scale features and third scale features to obtain a second local weight distribution matrix, multiplying the second scale features by the second local weight distribution matrix to realize coupling calibration of the second scale features, and sequentially splicing the first scale features after coupling calibration, the second scale features after coupling calibration and the original third scale features to form a fusion feature vector.

Description

Rhizoma polygonati processing after-forming evaluation system based on multi-scale fusion characteristics Technical Field The invention relates to the technical field of rhizoma polygonati processing quality detection, in particular to a rhizoma polygonati post-processing color forming evaluation system based on multi-scale fusion characteristics. Background The quality of the rhizoma polygonati is measured by the core index of the quality of the rhizoma polygonati after steaming, the conversion efficiency of the effective components and the classification of the market value are directly determined, the rhizoma polygonati with different shapes and sizes is usually required to be cut into rhizoma polygonati sections with basically consistent weight before steaming, and the sufficient steaming degree and the conversion effect of the effective components can be reflected on the characteristics of texture compactness, color uniformity and the like of the section after steaming. Under the traditional method, color forming judgment after processing of the rhizoma polygonati mainly depends on manual visual observation, and has the defects that 1, manual evaluation standards are not uniform, different operators have large definition differences of compact textures, colors and the like, cross-batch evaluation consistency is poor, and 2, only fuzzy grades can be given, and quality fine differences in the same grade cannot be distinguished. Therefore, there is a need for a color forming evaluation system after processing of rhizoma Polygonati, which can uniformly and quantitatively evaluate the color forming evaluation system, and solve the problems of subjectivity and poor consistency of manual evaluation. Disclosure of Invention Aiming at the defects of the prior art, the invention aims to provide a color forming evaluation system after processing of rhizoma polygonati based on multi-scale fusion characteristics, solves the problems of non-uniform manual evaluation standard and poor consistency in the existing evaluation system, and can realize accurate grading and quantitative evaluation of color forming after processing of rhizoma polygonati. The invention provides a basic scheme that a color evaluation system after processing of rhizoma polygonati based on multi-scale fusion characteristics comprises front-end acquisition equipment, a server and a user terminal; the front-end acquisition equipment is used for acquiring a section chart of the rhizoma polygonati after steaming processing; The server comprises an image preprocessing module, a multi-scale feature extraction and fusion module, a history database module, a deep learning evaluation module and a result output module; The image preprocessing module is used for preprocessing the acquired sectional images, and the preprocessing sequentially comprises color space calibration and effective area cutting; The multi-scale feature extraction and fusion module is used for extracting multi-scale features from the preprocessed sectional view, wherein the multi-scale features comprise a first scale feature, a second scale feature and a third scale feature, the multi-scale features are extracted through convolution kernels of different sizes, the convolution kernels of the first scale feature, the second scale feature and the third scale feature are sequentially increased in size, the scale features are respectively unified to the same preset dimension through independent mapping, normalization is respectively carried out, coupling calibration of the first scale feature and the second scale feature is realized through a cross-scale interaction mechanism, and the coupled calibrated first scale feature, the coupled calibrated second scale feature and the original third scale feature are spliced to form a fusion feature vector; The historical database module is used for constructing a sample set, wherein the sample set comprises cross-sectional diagrams after steaming and processing of rhizoma polygonati with different color forming grades, corresponding fusion feature vectors and manually marked evaluation results, and the evaluation results are scores with values of [0,100 ]; The deep learning evaluation module is used for constructing a deep learning model and training based on a sample set, and outputting an evaluation result of the rhizoma polygonati by inputting the fusion feature vector into the trained deep learning model; And the result output module is used for determining the color forming grade according to the evaluation result, generating an evaluation report containing the evaluation result, the color forming grade and the preprocessed section chart, and pushing the evaluation report to the user terminal. The method comprises the steps of acquiring a cross-section diagram of a sealwort after steaming processing by front-end acquisition equipment, preprocessing the original cross-section diagram by an image preprocessing module, ensuring real restoration o