A stroke population meal nutrition component detection system based on spectral analysis

Abstract

This invention relates to the field of health monitoring technology, specifically a spectral analysis-based system for detecting the nutritional components of meals in stroke patients. The system comprises: a detection device that collects spectral data cubes of the meal via an optical detection module and obtains mass data via a weighing module; a user terminal that collects image data and establishes a personal health record; a data processing center that integrates multimodal data, performs spectral preprocessing, feature extraction, image recognition and segmentation, and uses a chemometric model to invert nutritional components; and a risk assessment module that calculates a vascular burden index and performs risk grading based on the results and the individual's health record. This invention solves the problem of existing technologies being unable to accurately and quickly monitor the hidden salt and fat content in meals in a home environment, and thus unable to effectively assess the dietary risk of stroke recurrence. It enables non-destructive and rapid quantitative analysis of nutritional components, provides immediate risk warnings and personalized interventions, and effectively assists users in dietary risk management.

Description

Technical Field

[0001] This invention relates to the field of health monitoring technology, specifically to a system for detecting the nutritional components of meals in stroke patients based on spectral analysis. Background Technology

[0002] Currently, stroke patients require dietary management during their rehabilitation period, but there are many risk factors in daily family diets that are difficult to monitor accurately. Traditional dietary record-keeping methods mainly rely on manual estimation and recording by patients or their families, which is not only cumbersome, but also cannot effectively quantify the specific salt (sodium) and fat content in meals, especially the hidden salt and fat in condiments, sauces, and processed foods, resulting in poor actual effects of salt and fat control.

[0003] Existing nutritional analysis technologies, such as food image recognition applications, can only provide a rough assessment of food type. Their nutritional data largely comes from general databases and cannot reflect the actual differences in composition caused by specific cooking methods and ingredient ratios. They are particularly inadequate for accurately capturing key indicators such as sodium content and fatty acid composition, which require sophisticated chemical analysis. This results in a lack of accurate data foundation for risk assessment.

[0004] Therefore, for people with a history of stroke who need to strictly control their diet to prevent recurrence, the lack of an effective tool to accurately analyze the key nutrients (especially sodium and various types of fat) in each meal in a home setting makes it difficult for them to obtain risk warnings and professional and personalized dietary guidance tailored to their individual health conditions. Summary of the Invention

[0005] The purpose of this invention is to provide a spectral analysis-based system for detecting nutritional components in the diet of stroke patients. The detection device acquires spectral data cubes of the meal through an optical detection module and obtains mass data through a weighing module. The user terminal acquires image data and establishes a personal health record. The data processing center integrates multimodal data, performs spectral preprocessing, feature extraction, image recognition and segmentation, and uses a chemometric model to invert nutritional components. The risk assessment module calculates the vascular burden index based on the results and the individual's health record for risk classification. This system enables non-destructive and rapid quantitative analysis of nutritional components, provides immediate risk warnings and personalized interventions, effectively assists users in dietary risk management, and solves the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A system for detecting nutritional components in the diet of stroke patients based on spectral analysis, characterized in that it includes:

[0008] The testing equipment includes a housing, an optical detection module, a weighing module, and a control module disposed within the housing. The optical detection module is used to collect spectral data of the food, and the weighing module is used to acquire the quality data of the food.

[0009] The user terminal runs an application that collects image data of the meal, creates and stores the user's personal health record, and enables human-computer interaction.

[0010] The data processing center is configured to receive and process spectral and mass data from the detection equipment and image data from the user terminal, and to obtain the nutritional composition results of the meal through inversion analysis using a pre-stored chemometric model.

[0011] The risk assessment module is configured to calculate the vascular burden index and determine the risk level based on the nutritional component results and the physiological and pathological indicators in the user's personal health record.

[0012] The suggestion generation module is configured to automatically generate personalized dietary intake suggestions based on a pre-stored rule base and feed them back to the user through the user terminal when the risk assessment module determines that there is a risk.

[0013] Preferably, the optical detection module includes a light source unit and a spectral sensing unit; the light source unit is configured to emit near-infrared light with a wavelength range of 900 nm to 1700 nm; the spectral sensing unit is configured as a hyperspectral imager or a miniature near-infrared spectrometer, used to receive the spectral signals reflected by the food and generate a spectral data cube containing spatial and spectral dimensions.

[0014] Preferably, the data sources for the user's personal health record include importing from an authorized medical information system via an encrypted interface or manual input by the user. The data content includes at least recent resting blood pressure, blood lipid levels, fasting blood glucose levels, age, height, weight, and current medication information. The application is also configured to provide viewfinder-assisted focusing when acquiring image data and to perform preliminary processing such as autofocus, white balance correction, and color space conversion.

[0015] Preferably, the data processing center processes the spectral data by preprocessing and feature extraction; the preprocessing includes dark current noise subtraction, reflectance correction using a standard calibration plate, and smoothing and denoising using the SG convolutional filtering algorithm; the feature extraction includes calculating the absorption depth, absorption peak area, and spectral derivative features of a specific band, or obtaining principal component scores through principal component analysis.

[0016] Preferably, the data processing center processes the image data by: using a food image recognition model based on a deep learning convolutional neural network to identify the types of food in the meal, using an image segmentation algorithm to perform pixel-level segmentation of the identified food regions, and combining the quality data to estimate the quality of each type of food.

[0017] Preferably, the chemometric model includes:

[0018] The macronutrient quantitative correction model used to invert the content of calories, protein, fat and carbohydrates was trained using the partial least squares regression method.

[0019] The multivariate calibration model used for sodium content inversion is a deep neural network that uses spectral feature parameters of spectral bands corresponding to moisture state and ionic environment, combined with food type information, to perform nonlinear analysis.

[0020] The fatty acid classification model used to invert fatty acid composition is constructed based on the spectral differences in the characteristic absorption bands of fat molecules through principal component analysis or linear discriminant analysis, and combined with the PLSR model to output the percentage and content of saturated fatty acids, monounsaturated fatty acids and polyunsaturated fatty acids.

[0021] Preferably, the vascular burden index is calculated by calculating the risk sub-scores of sodium content and saturated fat content separately, and then synthesizing them using a weighted formula; the risk sub-score is calculated based on the ratio of the intake of the nutrient to the user's recommended daily intake, and its weighting factor is dynamically adjusted according to the severity of pathological indicators in the user's personal health record.

[0022] Preferably, the rule base built into the suggestion generation module is constructed based on the DASH dietary principles and includes food replacement and intake adjustment strategies for different risk factors; the generated suggestions specifically indicate the nutritional components that need to be adjusted, the risk level, and specific operational guidelines.

[0023] Preferably, the optimization learning module is configured to collect detection data anonymously with user authorization, and periodically use the newly accumulated dataset to perform incremental learning or retraining on the chemometrics model and the food image recognition model to improve the prediction accuracy and adaptability of the model.

[0024] Compared with the prior art, the beneficial effects of the present invention are:

[0025] 1. This invention deeply integrates the spectral detection and weighing technology of desktop testing equipment with image recognition and personal health record management of mobile phones to construct a system that combines multimodal data acquisition and nutritional component inversion analysis. It enables the accurate quantification of key risk components such as sodium content and saturated fat in meals in a home environment, breaking through the limitations of traditional methods that rely on manual estimation and general databases. This provides a conceptual and technical foundation for solving the problem of hidden salt and fat monitoring.

[0026] 2. By integrating near-infrared spectroscopy analysis, deep learning image recognition, and personalized health data, this invention can automatically complete the process from ingredient detection to risk warning, providing high-risk individuals for stroke with quantitative and real-time dietary risk indicators that are directly related to their current physiological and pathological state. This allows users to receive clear risk warnings and specific adjustment suggestions before eating, achieving precise intervention. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the system operation of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] To address the limitations of existing technologies in accurately and quickly monitoring the hidden salt and fat content in meals at home, and thus ineffectively assessing the dietary risk of stroke recurrence, please refer to [link to relevant documentation]. Figure 1 This embodiment provides the following technical solution:

[0030] When a user uses the system for the first time, they complete the system initialization setup through the application installed on their smartphone.

[0031] Specifically, the application guides users to connect their mobile phones to the testing device via Bluetooth or Wi-Fi to ensure a stable data communication link. Users create personal health records within the application. These records are imported from an authorized medical information system through an encrypted interface or are manually entered by the user. The data includes at least recent resting blood pressure, total cholesterol, LDL cholesterol, HDL cholesterol, triglycerides, and fasting blood glucose levels. Users can also selectively enter information such as age, height, weight, and current medications.

[0032] The detection equipment in this system is a benchtop device. Its internal casing mainly houses an optical detection module, a weighing module, and a control module. Specifically:

[0033] The optical detection module includes a light source unit and a spectral sensing unit; the light source unit is configured to emit light covering the near-infrared band, preferably in the wavelength range of 900 nm to 1700 nm; the spectral sensing unit is configured as a hyperspectral imager or a miniature near-infrared spectrometer to receive the spectral signals reflected by the food.

[0034] The weighing module is located on the surface of the device housing and is configured as a high-precision pressure sensor to obtain the weight of the meal;

[0035] The control module prompts the user to place the detection device on a plane with stable light before use, and prompts the user to place the standard reflectivity calibration plate on the weighing module. Then, it issues a command to make the detection device perform a light source stability check. The spectral sensing unit simultaneously performs background spectral data acquisition to complete the calibration and establish a benchmark for subsequent measurements.

[0036] The user places the food container steadily on the weighing module and sends a start-up command to the control module via the application.

[0037] The control module first drives the weighing module to perform the following: measure and obtain the total mass data of the food and utensils. The utensils are for the testing equipment only, and the weighing module has pre-set the mass value of the utensils and removes it from the total mass.

[0038] The control module drives the light source unit to perform the following: illuminate the food surface with a predetermined intensity and angle to reduce the adverse effects of shadows and specular reflections.

[0039] The control module drives the spectral sensing unit to perform the following: scan the surface of the food, capture its spectral data, and generate a spectral data cube containing dozens to hundreds of continuous bands. The spectral data cube contains spatial and spectral dimension information, that is, each pixel corresponds to a series of continuous wavelength reflection intensity values.

[0040] While the spectral scan is being performed, the user is guided to take one or more color images from directly above the food using the phone's camera. During this process, the application provides a viewfinder to assist the user in focusing and shooting, ensuring that the image clearly covers the entire food. The application performs preliminary processing on the captured images, including autofocus, white balance correction, and color space conversion.

[0041] Through the above process, the system obtains three sets of raw data: spectral data provided by the spectral sensing unit, image data provided by the mobile phone camera, and mass data provided by the weighing module. This data is temporarily cached in preparation for subsequent transmission and processing.

[0042] The collected multimodal data is transmitted to a data processing center, which can be deployed on a cloud server or integrated into a control module, depending on the computational complexity and real-time requirements.

[0043] First, the data processing center preprocesses the spectral data. The preprocessing process includes: dark current noise subtraction, which is to subtract the background signal of the spectral sensing instrument under no-light conditions; reflectance correction, which is to convert the original spectral signal into relative reflectance using the pre-acquired standard calibration plate spectral data; and smoothing and denoising, preferably using the SG convolution filtering algorithm to reduce random noise.

[0044] Next, key spectral features are extracted from the preprocessed spectral data, including but not limited to: absorption depth in specific bands, absorption peak area, first or second derivative characteristics of the spectrum, and principal component scores obtained through principal component analysis. These spectral features are indirect indicators related to nutrient concentrations.

[0045] Simultaneously, the image data is processed. First, a food image recognition model based on deep learning convolutional neural networks is used to analyze the images. This network model can preferably adopt the ResNet-50 architecture pre-trained on a large food image dataset. It has been optimized for Chinese cuisine through transfer learning. Its output results are the names and confidence probabilities of various food items present in the meal.

[0046] Furthermore, an image segmentation algorithm, preferably the U-Net semantic segmentation model, is used to segment the identified food regions at the pixel level, determine the pixel area ratio of each type of food in the image, outline the region contour of different foods in the plate, and estimate the quality of each type of food by combining the quality data obtained from the weighing module and the type of food.

[0047] The inversion of nutrient composition is achieved through a series of pre-trained chemometric models, specifically:

[0048] For macronutrients, including at least calories, protein, fat and carbohydrates, a quantitative calibration model is established between them and spectral characteristics. This model is preferably trained using partial least squares regression. The sample set used to train the model consists of spectral data and spectral characteristics of a large number of standard food samples with known precise nutrient content. The mass of each macronutrient in the meal is output through the mapping rules formed within the quantitative calibration model.

[0049] For sodium content, since the characteristic signal of sodium ions in the near-infrared region is relatively weak, an indirect correlation analysis method is adopted. Specifically, because sodium ions, after dissolving in the aqueous phase of food, significantly change the structure and dynamic properties of water molecules, and these changes can be keenly captured by near-infrared spectroscopy, a nonlinear model is trained using spectral characteristic parameters corresponding to the water state and ionic environment, combined with food type information, and using a large amount of standard sample data with known sodium content. A deep neural network is preferred to establish and form a multivariate calibration model. Through the mapping rules formed by the iteration of the multivariate calibration model itself, that is, using the nonlinear relationship between the spectral shape changes of the OH bonding frequency and the overtone region (approximately 1450 nm and 1940 nm bands) and the sodium content, a nonlinear transformation is performed to finally output the sodium content in the meal.

[0050] For fatty acids, a fatty acid classification model is trained and constructed using principal component analysis or linear discriminant analysis. Based on the differences in the absorption characteristics of the stretching vibration fundamental frequencies (approximately 1200 nm, 1390 nm, 1720 nm, 1760 nm, and 2306 nm bands) and combination frequencies (approximately 2140 nm, 2180 nm, 2310 nm, and 2340 nm bands) of the CH bonds in fat molecules, the characteristic spectra of saturated fatty acids, monounsaturated fatty acids, and polyunsaturated fatty acids are distinguished and substituted into the trained PLSR model. After the model completes the calculation, it directly outputs the predicted percentage of saturated fatty acids (SFA%), monounsaturated fatty acids (MUFA%), and polyunsaturated fatty acids (PUFA%). Combined with the fat mass output by the quantitative correction model, the content of each fatty acid is finally obtained.

[0051] Finally, by combining the spectral characteristics of the nutritional components, the food type information provided by image recognition, and the quality and proportion of each food region, a detailed list of nutritional components including calories, protein, fat, carbohydrates, sodium content, and fatty acid composition is output.

[0052] After obtaining the nutritional data, the system activates its built-in stroke dietary risk assessment model, primarily used to calculate a quantified vascular burden index. The calculation process is as follows:

[0053] First, the latest physiological and pathological indicators are retrieved from the individual's health record. Risk sub-scores are then calculated for the sodium and saturated fat content of the current meal. Each risk sub-score represents the degree of health threat posed by the sodium / saturated fat content to the current user; a higher score indicates a greater immediate risk. The risk sub-score is calculated based on the ratio of the nutrient intake to the user's recommended daily intake, and then weighted. The weighting factors are dynamically adjusted according to the severity of the user's pathological indicators. Finally, all risk sub-scores are synthesized into a vascular burden index using a specific weighting formula. For example, the specific calculation of the vascular burden index can be expressed as:

[0054] Vascular burden index = [W Na * (Na meal / Na max )^2 + W SF * (SF meal / SF max )^2]^(1 / 2)

[0055] Among them, W Na and W SF The weighting factors for sodium and saturated fat are respectively. For patients with hypertension, the weighting factor for sodium is set higher, while the weighting factor for saturated fat is relatively lower. For patients with dyslipidemia but normal blood pressure, the weighting factor for saturated fat is higher, while the weighting factor for sodium is lower.

[0056] Na meal and SF meal The measured amounts of sodium and saturated fat in this meal;

[0057] Na max and SF max The recommended daily intake limit for each user is set based on their personal health record and with reference to the DASH dietary principles or clinical guidelines.

[0058] The calculated vascular burden index is compared with preset multi-level risk thresholds, which can be divided into two levels: a concern threshold and a warning threshold. If the vascular burden index exceeds either threshold, an early warning signal is generated, and a prompt message is sent to the user via the application, clearly indicating the nutritional components that should be adjusted and their risk levels, thus enabling immediate intervention.

[0059] After the warning is issued, the system's built-in intake recommendation generator is triggered, automatically generating personalized intake recommendations that conform to the DASH dietary principles based on the risk assessment results.

[0060] The intake recommendation generator operates based on a rules engine. The rule base includes the DASH dietary principles and food substitution and intake adjustment strategies for different risk factors. For example, for meals with a high risk of sodium intake, the recommendation could be: "The sodium content of this meal was detected to be approximately A mg, exceeding your recommended single-meal intake by B%. It is recommended to reduce the intake of the sauce by approximately C grams."

[0061] All suggestions are displayed on the main interface of the mobile application, accompanied by simple illustrations. The interface also shows a detailed list of nutrients from this scan, the level and specific value of the vascular burden index, and a trend chart of historical data, allowing users to fully understand their dietary status. Users can adjust their diet on the spot according to the suggestions, and the record is archived for long-term tracking and follow-up visits.

[0062] With user authorization, the system anonymizes and collects detection data for model retraining. It periodically uses newly accumulated datasets to incrementally learn or retrain the nutritional component inversion model and the food image recognition model to continuously improve the adaptability and prediction accuracy to new food types and different cooking methods.

[0063] Working Principle: Upon first use, the user initializes the system via a mobile application, establishing a personal health record and connecting to the testing device. The testing device incorporates an optical detection module, a weighing module, and a control module, which are calibrated during the preparation phase to ensure accurate measurement standards. During data acquisition, the user places the food on the weighing module, and the system automatically acquires the mass data. The optical detection module collects the near-infrared spectral data of the food, forming a data cube containing spatial and spectral dimensions. Simultaneously, the user captures an image of the food using their mobile phone camera, with the application assisting in ensuring image quality.

[0064] In the data processing stage, spectral data undergoes preprocessing such as dark current noise subtraction, reflectance correction, and smoothing denoising to extract spectral features related to nutritional components. Image data is used to identify food types through a deep learning model, and the mass proportions of various food types are estimated using a semantic segmentation model. In the inversion analysis stage, a pre-trained chemometric model is used to map spectral features and food information to nutritional components, including calories, protein, fat, carbohydrates, sodium content, and fatty acid composition. Sodium content and fatty acid analysis are indirectly derived through spectral features in specific bands, overcoming the limitations of direct detection.

[0065] During the risk assessment phase, the system calculates the risk sub-scores of sodium and saturated fat in meals based on physiological indicators in the user's health record and synthesizes a vascular burden index. This index is compared with preset thresholds, and a warning signal is generated if it exceeds the limit. In the suggestion generation phase, the system automatically outputs personalized intake suggestions based on the DASH dietary principles, such as adjusting sauce intake, and displays a list of nutritional components and risk trends through the application.

[0066] It should 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. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0067] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A system for detecting nutritional components in the diet of stroke patients based on spectral analysis, characterized in that, include: The testing equipment includes an optical detection module, a weighing module, and a control module. The optical detection module is used to emit near-infrared light of a specific wavelength and collect spectral data of the food to form a spectral data cube. The weighing module is used to acquire the quality data of the food. The control module coordinates the work of each module and manages data communication. The user terminal is used to collect image data of meals, establish and store personal health records containing users' physiological and pathological indicators, and provide a human-computer interaction interface. The data processing center receives and fuses spectral data, mass data, and image data, and outputs nutritional composition results by performing the following processes: preprocessing and feature extraction of spectral data, performing food type identification and region segmentation on image data, and performing nutritional composition inversion analysis using pre-stored chemometric models. The risk assessment module is used to assess the stroke risk level of the current meal by calculating a quantified vascular burden index based on nutritional composition results and personal health records. The suggestion generation module is used to automatically generate personalized dietary intake suggestions based on a pre-stored dietary rule library when the risk exceeds the limit, and then provide feedback through the user terminal.

2. The system for detecting nutritional components of meals in stroke patients based on spectral analysis according to claim 1, characterized in that, The optical detection module includes a light source unit and a spectral sensing unit; The light source unit is configured to emit near-infrared light with a wavelength range of 900 nanometers to 1700 nanometers; The spectral sensing unit is configured as a hyperspectral imager or a miniature near-infrared spectrometer to receive spectral signals reflected from the food and generate a spectral data cube containing spatial and spectral dimensions.

3. The system for detecting nutritional components of meals in stroke patients based on spectral analysis according to claim 1, characterized in that, The data sources for the user's personal health record include imports from an authorized medical information system via an encrypted interface or manual input by the user. The data content includes at least recent resting blood pressure, blood lipid levels, fasting blood glucose levels, age, height, weight, and current medication information. The application is also configured to provide viewfinder-assisted focusing when acquiring image data, and to perform preliminary processing such as autofocus, white balance correction, and color space conversion.

4. The system for detecting nutritional components of meals in stroke patients based on spectral analysis according to claim 1, characterized in that, The data processing center processes spectral data including preprocessing and feature extraction. The preprocessing includes dark current noise subtraction, reflectivity correction using a standard calibration plate, and smoothing and denoising using the SG convolutional filtering algorithm. The feature extraction includes calculating the absorption depth, absorption peak area, and spectral derivative characteristics of a specific spectral band, or obtaining principal component scores through principal component analysis.

5. The system for detecting nutritional components of meals in stroke patients based on spectral analysis according to claim 1, characterized in that, The data processing center processes image data including: A food image recognition model based on deep learning convolutional neural networks is used to identify the types of food in a meal. Image segmentation algorithms are then used to segment the identified food regions at the pixel level, and the quality of each type of food is estimated by combining quality data.

6. The system for detecting nutritional components of meals in stroke patients based on spectral analysis according to claim 1, characterized in that, The chemometric model includes: A macronutrient quantitative calibration model is used to invert the content of calories, protein, fat and carbohydrates. This model is trained using partial least squares regression. A multivariate calibration model was used to invert sodium content. This model is a deep neural network that uses spectral feature parameters of spectral bands corresponding to moisture state and ionic environment, combined with food type information, to perform nonlinear analysis. A fatty acid classification model is used to invert fatty acid composition. It is constructed based on the spectral differences of the characteristic absorption bands of fat molecules through principal component analysis or linear discriminant analysis, and combined with the PLSR model to output the percentage and content of saturated fatty acids, monounsaturated fatty acids and polyunsaturated fatty acids.

7. The system for detecting nutritional components of meals in stroke patients based on spectral analysis according to claim 1, characterized in that, The vascular burden index is calculated as follows: The risk fractions for sodium content and saturated fat content were calculated separately and then synthesized using a weighted formula. The risk sub-score is calculated based on the ratio of the intake of this nutrient to the user's recommended daily intake, and its weighting factor is dynamically adjusted according to the severity of pathological indicators in the user's personal health record.

8. A system for detecting nutritional components of meals in stroke patients based on spectral analysis according to claim 7, characterized in that, The specific formula for calculating the vascular burden index is as follows: Vascular burden index = [W Na * (Na meal / Na max )^2 + W SF * (SF meal / SF max )^2]^(1 / 2); where W Na and W SF These are the weighting factors for sodium and saturated fat, respectively. Na meal and SF meal The measured amounts of sodium and saturated fat in this meal; Na max and SF max This represents the upper limit of the recommended daily intake for each individual user.

9. A system for detecting nutritional components in the diet of stroke patients based on spectral analysis according to claim 1, characterized in that, The rule base built into the suggestion generation module is based on the DASH dietary principles and includes food replacement and intake adjustment strategies for different risk factors. The generated suggestions specifically indicate the nutritional components that need to be adjusted, the risk level, and specific operational guidelines.

10. A system for detecting nutritional components in the diet of stroke patients based on spectral analysis according to claim 1, characterized in that, The system also includes an optimization learning module, which is configured as follows: With user authorization, detection data is collected anonymously, and the newly accumulated dataset is used periodically to perform incremental learning or retraining on the chemometrics model and food image recognition model.

Images