Effectiveness prediction method for medicinal and edible food based on model analysis

Abstract

The invention discloses a model analysis-based effectiveness prediction method for medicinal and edible food, which comprises the following steps of: constructing a feature effectiveness association map formed by combining an image feature space and an effectiveness measurement space, the association map comprises a mapping function between the image feature space obtained by pre-training and the validity measurement space; placing image data of a to-be-detected medicine and food homologous food in the image feature space, and obtaining a mapping position of a quantization coordinate of the image data in the image feature space in the validity measurement space through the mapping function; and based on the coordinate value of the mapping position in the validity measurement space, outputting a prediction result of the validity grade of the to-be-detected medicinal and edible food.

Description

Technical Field

[0001] This invention relates to a method for predicting the effectiveness of food and medicine homology based on model analysis. Background Technology

[0002] Traditional methods for assessing the effectiveness of food-medicine homology products mainly rely on physicochemical analysis (such as high-performance liquid chromatography) or sensory evaluation. While physicochemical analysis is accurate, it is time-consuming, labor-intensive, costly, and destructive, making it unsuitable for large-scale, rapid on-site screening. Sensory evaluation, on the other hand, is easily influenced by subjective factors and lacks stability and consistency.

[0003] Chinese invention patent CN119246702A discloses a "method for detecting the active ingredients of a traditional Chinese medicine composition by UPLC-MS / MS". This method quantitatively analyzes 14 active ingredients (such as saikosaponin A, paeoniflorin, and salvianolic acid B) that may be present in a specific compound traditional Chinese medicine, Ganshuang granules. It employs gradient elution and electrospray ionization mass spectrometry to detect these functional components. However, this method is primarily applicable to the specific compound traditional Chinese medicine Ganshuang granules and is difficult to directly apply to other substances that are both medicinal and edible.

[0004] Therefore, there is an urgent need for a non-destructive, rapid, low-cost, and scalable method for predicting the effectiveness of food and medicine homology, in order to meet the practical needs of modern industry for high-throughput, multi-index evaluation of diverse samples. Summary of the Invention

[0005] This invention provides a model-based method for predicting the effectiveness of food and medicine homology. Through image acquisition and analysis technology, it realizes a detection method applicable to food and medicine homology samples of different types, sources and processing methods.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a model-based method for predicting the effectiveness of food-medicine homology foods, comprising: A feature validity association map is constructed, which is composed of an image feature space and a validity metric space. The association map includes a mapping function between the image feature space and the validity metric space obtained through pre-training. The image feature space is a multi-dimensional space composed of quantized coordinates of color gradation parameters, texture scale parameters and morphological structure parameters of the sample image. The validity metric space is a coordinate domain constructed according to the validity level of the sample image. The image data of the food product to be tested is placed in the image feature space, and the mapping position of the quantized coordinates of the image data in the image feature space in the effectiveness measurement space is obtained through the mapping function. Based on the coordinates of the mapped position in the effectiveness measurement space, the predicted effectiveness level of the food-medicine homology product to be tested is output.

[0007] In one optional embodiment, the color gradation parameter includes a chromaticity coordinate distribution for characterizing the content of a specific active ingredient in a food that is both food and medicine; the texture scale parameter includes texture feature values ​​for characterizing the tissue density in a food that is both food and medicine; and the morphological structure parameter includes morphological measurements for characterizing the regularity or integrity of a specific part of a food that is both food and medicine.

[0008] In one optional embodiment, the chromaticity coordinate distribution uses a joint histogram of the a* and b* channels of the CIE Lab color space as quantification data to characterize the color intensity and uniformity of the specific active ingredient.

[0009] In an optional embodiment, the color gradation parameter includes a chromaticity coordinate distribution for characterizing the content of flavonoids or carotenoids; and / or, the texture scale parameter includes texture feature values ​​for characterizing the density of plant tissue or fungal seed tissue; and / or, the morphological structure parameter includes the aspect ratio for root and stem medicinal and edible foods, or the contour symmetry index for flower, fruit, and leaf medicinal and edible foods.

[0010] In one alternative embodiment, the texture feature values ​​are determined by a combination of energy, contrast, and correlation based on a multi-scale gray-level co-occurrence matrix, and are used to characterize the spatial distribution pattern of the tissue density.

[0011] In one optional embodiment, the sample image and the image of the food-medicine homology to be tested are both high-resolution digital images containing preset identification parts, acquired under set standard light source conditions, fixed shooting angle and constant distance.

[0012] In one optional embodiment, the standard light source conditions for acquiring the high-resolution digital image are simulated sunlight with a color temperature range of 5000K to 5500K, and the fixed shooting angle deviates from the surface normal of the sample image and the food-medicine homology food to be tested by no more than 10°.

[0013] In one alternative embodiment, the mapping function is a two-branch deep neural network based on an attention mechanism; The first branch network is used to extract and fuse the multi-dimensional features of the color gradation parameter and texture scale parameter, and the second branch network is used to extract the local subtle features of the preset identification parts included in the morphological structure parameter. After the multi-dimensional features and local subtle features are weighted and fused by the attention mechanism, they are returned to the coordinate values ​​of the validity measurement space.

[0014] In one optional embodiment, the prediction result is overlaid on the original image of the food-medicine homology to be tested in the form of a visual heatmap; wherein the heatmap is generated by the gradient weights in the mapping function and is used to identify image regions that contribute more than a preset threshold to the prediction result.

[0015] The beneficial effects of the technical solution provided by this invention include at least the following: 1. This invention constructs a "feature validity association map" to systematically associate the image feature space with the validity measurement space, rather than simply splicing features. This allows for a deeper exploration and utilization of the inherent complex mapping relationship between image features and validity, significantly improving the accuracy and stability of prediction.

[0016] 2. This invention is based on image analysis and does not require any chemical treatment or damage to the sample. It can complete the prediction in a short time and is particularly suitable for applications such as rapid screening on the production line and on-site sampling inspection of products in the market, which greatly improves efficiency and reduces costs.

[0017] 3. This invention identifies key image regions that contribute most to the prediction results using a visual heatmap, making the model's decision-making process transparent and intuitive. This not only enhances users' trust in the prediction results but also provides clear data support and guidance for improving production processes (such as optimizing harvesting locations and processing parameters).

[0018] 4. By defining standardized image acquisition conditions, employing multi-scale feature extraction, and designing morphological parameters for specific categories (such as root and stem crops, flowers, fruits, and leaves), this invention enables the method to effectively overcome interference caused by differences in lighting, angle, and individual morphology, and maintains good predictive performance for different types and origins of food products that are both food and medicine. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating a method for predicting the effectiveness of food and medicine homology based on model analysis, provided in an embodiment of the present invention. Detailed Implementation

[0021] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The following embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes, but the scope of protection of the present invention is not limited to the following embodiments.

[0022] In this invention, the terms "in one possible embodiment," "exemplary," or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "in one possible embodiment," "exemplary," or "for example" in this invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "in one possible embodiment," "exemplary," or "for example" is intended to present the relevant concepts in a specific manner.

[0023] Food and medicine homologous substances are widely used in the food, health product, and traditional Chinese medicine industries. Due to their combined nutritional and potential therapeutic effects, they hold an important position in the modern health industry. These substances have a wide range of sources and varieties. The types, contents, and activities of their functional components are greatly affected by factors such as variety, origin, storage, and processing. Accurate and comprehensive prediction of their effectiveness is an important foundation for product quality evaluation and application.

[0024] Figure 1 This is a flowchart illustrating a method for predicting the effectiveness of food-medicine homology based on model analysis, provided in an embodiment of the present invention. Figure 1 As shown, a method for predicting the effectiveness of food-medicine homology foods based on model analysis includes: S101, construct a feature validity association map composed of an image feature space and a validity metric space, wherein the association map includes a pre-trained mapping function between the image feature space and the validity metric space.

[0025] The image feature space is a multi-dimensional space composed of quantized coordinates of color gradation parameters, texture scale parameters, and morphological structure parameters of the sample image, and the validity measurement space is a coordinate domain constructed according to the validity level of the sample image.

[0026] In one optional embodiment, the color gradation parameter includes a chromaticity coordinate distribution for characterizing the content of a specific active ingredient in a food that is both food and medicine; the texture scale parameter includes texture feature values ​​for characterizing the tissue density in a food that is both food and medicine; and the morphological structure parameter includes morphological measurements for characterizing the regularity or integrity of a specific part of a food that is both food and medicine.

[0027] Examples include specific active ingredients such as chlorogenic acid in honeysuckle. Chromaticity coordinate distribution, such as a specific yellow-green hue in petals. Tissue density, for example, indirectly reflects cell compactness by analyzing the texture roughness of petal images, and is related to the accumulation of active ingredients. Morphological structure, for example, is assessed by calculating the proportion and integrity of unopened buds, as excessive opening leads to the loss of active ingredients.

[0028] In one optional embodiment, the chromaticity coordinate distribution uses a joint histogram of the a* and b* channels of the CIE Lab color space as quantification data to characterize the color intensity and uniformity of the specific active ingredient.

[0029] For example, hawthorn is rich in anthocyanins, which are mainly represented in the a (red-green) and b (yellow-blue) channels in the CIE Lab color space. The two-dimensional joint histogram of its a and b channels is calculated. The centroid and dispersion of this histogram can quantify the color intensity of anthocyanins and the uniformity of their distribution across the entire fruit surface, respectively, serving as key color features for predicting its antioxidant activity.

[0030] In an optional embodiment, the color gradation parameter includes a chromaticity coordinate distribution for characterizing the content of flavonoids or carotenoids; and / or, the texture scale parameter includes texture feature values ​​for characterizing the density of plant tissue or fungal seed tissue; and / or, the morphological structure parameter includes the aspect ratio for root and stem medicinal and edible foods, or the contour symmetry index for flower, fruit, and leaf medicinal and edible foods.

[0031] For example, different combinations of characteristics can be used for different categories of medicinal materials. For instance, for carrots (roots and rhizomes), the orange hue corresponding to their carotenoid content is extracted, and their aspect ratio (length to maximum diameter) is calculated; those with regular shapes are generally of better quality. For chrysanthemums (flowers, fruits, and leaves), the outline symmetry index of their flower discs is calculated; chrysanthemums with symmetrical shapes are generally healthier and have a more complete accumulation of active ingredients.

[0032] In one alternative embodiment, the texture feature values ​​are determined by a combination of energy, contrast, and correlation based on a multi-scale gray-level co-occurrence matrix, and are used to characterize the spatial distribution pattern of the tissue density.

[0033] For example, when analyzing microscopic images of Ganoderma lucidum spore powder, a multi-scale gray-level co-occurrence matrix was used to quantify the "tissue density" of the spore wall. Energy (uniformity), contrast (sharpness), and correlation (linear dependence) were calculated at multiple scales (e.g., pixel distance d=1,3,5). The combination of high energy, high contrast, and moderate correlation was verified to be highly correlated with the density and thickness of the spore wall, and consequently, with the difficulty of cell wall disruption and the release rate of active ingredients.

[0034] In one optional embodiment, the sample image and the image of the food-medicine homology to be tested are both high-resolution digital images containing preset identification parts, acquired under set standard light source conditions, fixed shooting angle and constant distance.

[0035] For example, to ensure data consistency, an image acquisition dark box can be set up. When acquiring images of Panax notoginseng rhizomes, the sample is placed on a stage, and the camera is fixed directly above the sample at a distance of 30cm (constant distance) by a clamp. The lens optical axis is perpendicular to the main surface of the sample (fixed shooting angle), and the image is illuminated by a surrounding D65 standard light source (standard light source conditions) to ensure that clear images of the main root and lateral roots (preset identification parts) can be captured completely each time.

[0036] In one optional embodiment, the standard light source conditions for acquiring the high-resolution digital image are simulated sunlight with a color temperature range of 5000K to 5500K, and the fixed shooting angle deviates from the surface normal of the sample image and the food-medicine homology food to be tested by no more than 10°.

[0037] For example, in the image acquisition darkroom, an LED flat panel light with a color temperature of 5300K can be selected as a simulated sunlight source. The camera is mounted on a gimbal with an angle scale, and through repeated calibration, the angle between its optical axis and the normal to the sample stage plane is precisely controlled within 5° to minimize feature distortion caused by reflections and shadows.

[0038] In one alternative embodiment, the mapping function is a two-branch deep neural network based on an attention mechanism; The first branch network is used to extract and fuse the multi-dimensional features of the color gradation parameter and texture scale parameter, and the second branch network is used to extract the local subtle features of the preset identification parts included in the morphological structure parameter. After the multi-dimensional features and local subtle features are weighted and fused by the attention mechanism, they are returned to the coordinate values ​​of the validity measurement space.

[0039] For example, taking the prediction of ginseng efficacy as an example, the specific workflow of its "dual-branch deep neural network" is as follows: The first branch is a CNN backbone network (such as ResNet), responsible for extracting and fusing global color (skin color) and texture (root roughness) features from the entire ginseng image. The second branch is a more local-focused CNN, which receives an image of the rhizome (pre-defined identification part) cropped from the original image and specifically extracts local subtle features such as the number and density of rhizome buds. Finally, an attention mechanism module dynamically assigns different weights to the global features and the local features of the rhizome (e.g., determining whether the prediction relies more on the texture of the main root or the features of the rhizome), and after weighted fusion, finally outputs a predicted content value of the effective component.

[0040] S102, the image data of the food-medicine homology food to be tested is placed in the image feature space, and the mapping position of the quantized coordinates of the image data in the image feature space in the validity measurement space is obtained through the mapping function.

[0041] S103, based on the coordinates of the mapped position in the effectiveness measurement space, output the prediction result of the effectiveness level of the food-medicine homology to be tested.

[0042] In one optional embodiment, the prediction result is overlaid on the original image of the food-medicine homology to be tested in the form of a visual heatmap; wherein the heatmap is generated by the gradient weights in the mapping function and is used to identify image regions that contribute more than a preset threshold to the prediction result.

[0043] For example, taking cinnamon as an example, on the user interface, after the system completes a prediction for a cinnamon image, in addition to displaying the text result "Effectiveness Level: High," it also generates a semi-transparent color heatmap on the original image. This heatmap is generated by calculating the gradient weights in a neural network using class activation mapping (Grad-CAM). The red area in the image identifies the part that contributes the most to this "high" level prediction, namely the oil cell layer on the inner side of the cinnamon bark, while the outer cork layer is shown in blue, indicating a lower contribution. This intuitively tells the user that the prediction is mainly based on the quality of its oil-containing parts.

[0044] In this embodiment, Ningxia wolfberry can be used as an example. First, hundreds of wolfberry samples with known polysaccharide content (determined by HPLC) can be collected as a training set. To construct the "image feature space," images of each sample are captured, and their color (e.g., red saturation), texture (e.g., surface wrinkle density), and morphology (e.g., aspect ratio, stem integrity) parameters are quantified and extracted. To construct the "effectiveness measurement space," the polysaccharide content is divided into four levels: "excellent, excellent, good, and qualified," as the coordinate domain. Subsequently, a support vector regression (SVR) model is used to learn the "mapping function" from the feature space to the measurement space. During prediction, the wolfberry image to be tested is placed in this feature space, and the corresponding "excellent" coordinates in the effectiveness measurement space can be found through this function, thereby outputting the predicted level.

[0045] Furthermore, after deployment, incremental model updates are supported. For example, when a batch of Astragalus membranaceus samples from a new origin is introduced, there is no need to retrain the entire model. Only the images of these new samples and their laboratory-determined efficacy data are used as new data, and the existing "mapping function" is incrementally updated through online learning or fine-tuning algorithms. This allows the model to quickly adapt to products from new sources without forgetting previously learned knowledge, ensuring the system's continuous evolution and long-term usability.

[0046] Furthermore, this method can be integrated with automated production equipment. For example, on a jujube grading production line, a camera captures images of each jujube on the conveyor belt and predicts its validity level in real time. When the system continuously detects multiple jujubes predicted to be "low-grade," it can automatically send a control signal to the upstream cleaning or screening equipment, prompting adjustments to water pressure, washing intensity, or sorting thresholds. This achieves closed-loop optimization of production parameters, improving the quality of the entire batch of products from the source.

[0047] To meet the needs of on-site spot checks, the trained lightweight model can be deployed on a smartphone app. Quality inspectors use their phones to photograph commercially available goji berries, and the app instantly displays the predicted grade and heat map. Simultaneously, the app can upload anonymized image data and prediction results to a cloud database. By aggregating massive amounts of data from across the country, the cloud can periodically train a more powerful global model, which is then pushed to all terminal apps via OTA (Over-The-Air) updates, forming a distributed, self-reinforcing ecosystem.

[0048] The beneficial effects of the technical solution provided by this invention include at least the following: By constructing a "feature effectiveness correlation map," the image feature space and effectiveness measurement space are systematically correlated, rather than simply splicing features. This allows for a deeper exploration and utilization of the inherent complex mapping relationship between image features and effectiveness, significantly improving the accuracy and stability of predictions. Based on image analysis, predictions can be completed in a short time without any chemical treatment or damage to the samples, making it particularly suitable for applications such as rapid screening on production lines and on-site sampling of market products, greatly improving efficiency and reducing costs. Visualized heatmaps identify the key image regions that contribute most to the prediction results, making the model's decision-making process transparent and intuitive. This not only enhances users' trust in the prediction results but also provides clear data support and guidance for improving production processes (such as optimizing harvesting parts and processing parameters). By defining standardized image acquisition conditions, employing multi-scale feature extraction, and designing morphological parameters for specific categories (such as root vegetables, flowers, fruits, and leaves), this method can effectively overcome interference caused by differences in lighting, angle, and individual morphology, and maintain good predictive performance for different types and origins of food products that are both food and medicine.

[0049] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0050] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations 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, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0051] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0052] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0053] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A method for predicting the effectiveness of food-medicine homology foods based on model analysis, characterized in that, include: A feature validity association map is constructed, which is composed of an image feature space and a validity metric space. The association map includes a mapping function between the image feature space and the validity metric space obtained through pre-training. The image feature space is a multi-dimensional space composed of quantized coordinates of color gradation parameters, texture scale parameters and morphological structure parameters of the sample image. The validity metric space is a coordinate domain constructed according to the validity level of the sample image. The image data of the food product to be tested is placed in the image feature space, and the mapping position of the quantized coordinates of the image data in the image feature space in the effectiveness measurement space is obtained through the mapping function. Based on the coordinates of the mapped position in the effectiveness measurement space, the predicted effectiveness level of the food-medicine homology product to be tested is output.

2. The method for predicting the effectiveness of food-medicine homology foods based on model analysis as described in claim 1, characterized in that, The color gradation parameter includes the color coordinate distribution used to characterize the content of specific active ingredients in food products that are both food and medicine; the texture scale parameter includes texture feature values ​​used to characterize the tissue density in food products that are both food and medicine; the morphological structure parameter includes morphological measurements used to characterize the regularity or integrity of specific parts of food products that are both food and medicine.

3. The method for predicting the effectiveness of food-medicine homology foods based on model analysis as described in claim 2, characterized in that, The chromaticity coordinate distribution uses the joint histogram of the a* and b* channels of the CIE Lab color space as quantification data to characterize the color intensity and uniformity of the specific active ingredient.

4. The method for predicting the effectiveness of food-medicine homology foods based on model analysis as described in claim 2, characterized in that, The color gradation parameter includes a chromaticity coordinate distribution used to characterize the content of flavonoids or carotenoids; and / or, the texture scale parameter includes texture feature values ​​used to characterize the density of plant tissue or fungal seed tissue; and / or, the morphological structure parameter includes the aspect ratio for root and stem medicinal and edible foods, or the outline symmetry index for flower, fruit, and leaf medicinal and edible foods.

5. The method for predicting the effectiveness of food-medicine homology foods based on model analysis as described in claim 2, characterized in that, The texture feature values ​​are determined by a combination of energy, contrast, and correlation based on a multi-scale gray-level co-occurrence matrix, and are used to characterize the spatial distribution pattern of the tissue density.

6. The method for predicting the effectiveness of food-medicine homology foods based on model analysis as described in claim 1, characterized in that, The sample images and the images of the food products to be tested are all high-resolution digital images containing preset identification parts, acquired under set standard light source conditions, fixed shooting angles, and constant distances.

7. The method for predicting the effectiveness of food-medicine homology foods based on model analysis as described in claim 6, characterized in that, The standard light source conditions for acquiring the high-resolution digital image are simulated sunlight with a color temperature range of 5000K to 5500K, and the fixed shooting angle deviates from the surface normal of the sample image and the food-medicine homology food to be tested by no more than 10°.

8. The method for predicting the effectiveness of food-medicine homology foods based on model analysis as described in claim 1, characterized in that, The mapping function is a dual-branch deep neural network based on an attention mechanism; The first branch network is used to extract and fuse the multi-dimensional features of the color gradation parameter and texture scale parameter, and the second branch network is used to extract the local subtle features of the preset identification parts included in the morphological structure parameter. After the multi-dimensional features and local subtle features are weighted and fused by the attention mechanism, they are returned to the coordinate values ​​of the validity measurement space.

9. The method for predicting the effectiveness of food-medicine homology foods based on model analysis as described in claim 1, characterized in that, The prediction results are overlaid on the original image of the food-medicine homology to be tested in the form of a visual heatmap; wherein, the heatmap is generated by the gradient weights in the mapping function and is used to identify image regions that contribute more than a preset threshold to the prediction results.

Images