AI-supported system for quality assessment and detection of adulteration in clarified butter

Abstract

A system for AI-driven quality assessment and adulteration detection in clarified butter, the system includes: a data acquisition and processing unit comprising a multispectral imaging unit configured to capture high-resolution images of ghee samples under controlled illumination conditions, the module also comprising a temperature-stabilized sample chamber with optically transparent walls to ensure uniform light scattering and minimize shadow artifacts; a chemical measurement subsystem capable of performing at least one conventional analysis, including gravimetric moisture determination, Baudouin reaction or spectroscopic profiling, wherein the subsystem is operationally connected to a microfluidic cartridge for sample handling and reagent dispensing; An annotated data repository configured to store captured images along with chemical and physical adulteration measurements. The repository includes a structured relational database encoding metadata, including batch number, sample origin, illumination parameters, and adulteration metrics. a deep learning inference machine incorporating convolutional neural networks (CNNs), where the machine is trained on the annotated data store to detect forgery signatures, including variations in color gradients, crystalline microstructures, and spectral reflectance profiles; an automated Baudouin test analysis pipeline in which sequential image frames of the reaction chamber are captured and processed using a color segmentation technique in conjunction with AI-driven classification to quantify the adulteration of clarified butter with reduced subjectivity; and A user interface integrated into a mobile device, an industrial console, or a cloud-accessible platform, configured to generate non-destructive, real-time reports of counterfeiting, with these reports including predicted counterfeit levels, confidence metrics, and regulatory compliance thresholds.

Description

AREA OF INVENTION

[0001] The present invention relates to systems and devices for assessing food quality. In particular, it relates to an artificial intelligence-based framework for the non-destructive and automated detection of adulteration in clarified butter using computer vision, image processing, and deep learning models. BACKGROUND OF THE INVENTION

[0002] Purified butter is a nutrient-rich dairy product valued for its flavor, health benefits, and cultural significance. However, adulteration with vegetable oils, clarified butter, and excess moisture is widespread due to economic incentives, jeopardizing consumer safety and market confidence. Conventional detection methods, including chemical tests such as the Baudouin test, gravimetric analyses, and advanced techniques like GC-MS or spectroscopy, are either labor-intensive, expensive, or require highly skilled personnel. Furthermore, these techniques are not suitable for real-time or on-site testing, particularly for small-scale producers and end consumers.

[0003] Recent efforts to develop electronic nasal systems or mobile applications have failed to achieve reliable accuracy due to their reliance on calibration, sensor drift, and limited use of AI. Therefore, there is an urgent need for a fast, scalable, and cost-effective system that combines the accuracy of laboratory testing with the accessibility of mobile or industrial devices.

[0004] Purified butter has long held a firm place in culinary, medicinal, and cultural practices, particularly in South Asia, where it is considered a staple food and an object of ritual significance. Its high nutritional content, combined with attractive sensory qualities such as taste and aroma, makes it one of the most prized dairy products. However, the very qualities that give ghee its premium value also make it susceptible to adulteration. Adulteration practices, typically involving the addition of cheaper oils, hydrogenated fats, and water or moisture, are driven by the economic incentive to increase margins at the expense of quality. This poses a significant food safety concern, not only because adulterants diminish ghee's nutritional and organoleptic qualities, but also because they present health risks ranging from gastrointestinal issues to long-term metabolic disorders.The increasing demand for ghee, particularly in global markets where traditional dairy products are gaining popularity, has further exacerbated the problem. It has become crucial to ensure robust, reliable, and scalable mechanisms for detecting adulteration and guaranteeing purity throughout the entire supply chain.

[0005] Historically, the detection of adulteration in ghee relied largely on chemical tests. The most common of these was the Baudouin test, used to determine the presence of clarified butter, a hydrogenated fat frequently mixed with ghee. In the Baudouin test, ghee is mixed with hydrochloric acid and furfural solution. The presence of clarified butter results in a distinct pink or red color. While this method is simple and has been used for decades, it is subjective because it relies on human interpretation of the color change. Furthermore, the test's sensitivity decreases at very low levels of adulteration, making it unsuitable for detecting marginal impurities that may nevertheless have adverse health effects. In addition, its reliance on expert handling and controlled laboratory conditions limits its practicality in the field.Similarly, gravimetric methods have been used to determine moisture content, in which samples are subjected to controlled drying to ascertain the water content. While this method is reliable, it is time-consuming and requires specialized equipment, making it unsuitable for rapid or on-site testing.

[0006] Beyond these basic chemical approaches, advanced analytical tools have been developed to achieve greater accuracy in adulteration detection. Techniques such as gas chromatography-mass spectrometry (GC-MS) allow for the precise identification of adulterants, including specific vegetable oils and fatty acid profiles. However, GC-MS requires expensive equipment, highly skilled technicians, and significant sample preparation time. Its usefulness is therefore limited to well-equipped laboratories and is often unaffordable for small-scale producers and regulatory bodies in resource-constrained environments. Similarly, spectroscopic techniques, including ATR-FTIR (attenuated total reflection Fourier transform infrared spectroscopy) and UV-VIS spectrophotometry, have been employed to detect subtle differences between pure and adulterated ghee.These methods have proven promising in distinguishing between different adulterants based on their chemical fingerprints. However, they have drawbacks such as overlapping absorption bands that complicate interpretation, dependence on calibration using reference standards, and the need for expensive instruments. These characteristics, in turn, limit them to controlled laboratory environments and prevent widespread use in field scenarios.

[0007] Another class of detection methods is based on measuring physical properties such as refractive index, specific gravity, melting point, and viscosity. These tests are inexpensive and relatively easy to perform, even in rudimentary environments. For example, an unusually high refractive index or a low melting point may indicate adulteration with vegetable oils. However, these methods lack precision because natural variations in ghee composition due to factors such as cow breed, feed, and seasonal changes can overlap with adulterated profiles. This results in a significant error rate and renders these tests unreliable for regulatory compliance or consumer safety. Such physical property-based methods are therefore better suited for preliminary screening but cannot serve as confirmatory tools.

[0008] With the advent of digital technologies, attempts have been made to introduce more portable and consumer-friendly solutions. One example is mobile applications that purportedly help consumers assess food adulteration through simple visual guidelines, manual input, or rudimentary image analysis. While these applications provide rough estimates or guidelines, they do not offer scientifically validated or quantitative results. Because they rely on user-input data or crude visual cues, they are prone to error and cannot match the precision of laboratory tests. Consequently, their use is limited to informal consumer awareness rather than regulatory or industrial practice. Similarly, consumer kits for detecting food adulteration have also appeared on the market.These often include colorimetric reagents or disposable strips that can detect the presence of certain adulterants. While convenient and providing rapid results, they have limitations such as restricted detection ranges, lack of quantification, and reagent instability, which reduces their reliability and shelf life.

[0009] Recent research has also explored the use of electronic noses, or e-nose systems, for adulteration detection. These systems use gas sensor arrays to detect volatile organic compounds emitted by ghee and analyze their unique patterns using chemometric models. The appeal of e-nose systems lies in their potential portability and their ability to deliver rapid results without complex sample preparation. However, their performance is highly dependent on sensor calibration, and issues such as sensor drift, sensitivity to environmental factors, and cross-sensitivity to unrelated compounds have hindered their widespread adoption. Furthermore, interpreting the multivariate data generated by e-nose systems requires expertise in chemometrics, making them impractical for widespread, decentralized use.Although promising, e-nose technologies still need to reach a level of maturity that enables consistent accuracy in real-world supply chain environments.

[0010] Each of the methods mentioned above illustrates the trade-offs between accessibility, accuracy, cost, and scalability. Chemical methods, while simple and inexpensive, suffer from subjectivity and low sensitivity. Modern analytical techniques offer high precision but remain inaccessible to many due to their cost and infrastructure requirements. Approaches based on physical properties are user-friendly but unreliable. New digital and e-nose systems, while innovative, face challenges regarding accuracy, calibration, and scalability. This fragmented landscape highlights a key challenge: there is no single solution that is fast, non-destructive, accurate, cost-effective, and scalable enough to detect adulteration at all levels of the ghee supply chain—from producers and processors to regulators and consumers.

[0011] The consequences of this technological gap are far-reaching. Regulators struggle to enforce food safety standards without large-scale, real-time testing. Manufacturers, particularly small and medium-sized enterprises, cannot guarantee product integrity or comply with export regulations due to a lack of access to reliable testing. Consumers, on the other hand, lose confidence in food safety due to the absence of accessible and trustworthy tools, especially in markets where ghee consumption holds strong cultural and symbolic significance. As adulteration practices evolve with the introduction of new adulterants and sophisticated blending techniques, static detection methods quickly become obsolete.This dynamic nature of the distortion further exacerbates the inadequacy of existing solutions and requires a system that is not only precise but also adaptable and scalable.

[0012] Another aspect of the problem lies in integrating testing into supply chain transparency. Existing methods often operate in isolation, delivering results limited to laboratory reports or consumer-level observations. They cannot be seamlessly integrated into digital supply chain platforms or regulatory databases. With the increasing globalization of food systems, there is a growing demand for solutions that offer not only accurate testing but also data connectivity, enabling real-time traceability and regulatory compliance. Current methods do not meet this requirement because they are not designed for digital interoperability or data exchange.

[0013] The technical background of adulteration detection in clarified butter thus reveals a complex challenge characterized by diverse but inadequate solutions. Chemical, physical, and analytical methods, despite their strengths, are subject to limitations regarding subjectivity, cost, infrastructure, and scalability. Consumer-friendly kits and mobile applications, while accessible, do not meet the required scientific rigor. Emerging technologies such as e-noses and chemometric analyses remain experimental and difficult to implement on a large scale. Against this backdrop, a critical gap exists for an innovative solution that combines the accuracy of advanced analytical methods with the accessibility and scalability of modern digital systems, while remaining adaptable to evolving adulteration practices.The lack of such a comprehensive solution continues to hinder efforts to ensure the safety, authenticity and marketability of clarified butter in both local and international contexts. Summary of the invention

[0014] The invention provides a modular system that integrates hardware and software for assessing ghee purity. A high-resolution imaging module captures images of ghee samples from multiple angles under controlled illumination. These images are preprocessed and stored along with physical measurements of moisture, clarified butter, and vegetable oil content. The data are annotated and fed into a deep learning engine trained on convolutional neural networks (CNNs) such as ResNet, VGG, or EfficientNet. The AI ​​system predicts the degree of adulteration by identifying subtle differences in the texture, color, and crystal structure of the ghee.

[0015] A remarkable innovation is the automation of the Baudouin test. The system captures sequential images of the chemical reaction and interprets color changes using artificial intelligence, thus eliminating human subjectivity. The results are displayed in real time via a mobile or industrial platform interface, enabling rapid decision-making throughout the entire supply chain.

[0016] The present invention pursues several objectives: It aims to solve the long-standing challenges in detecting adulteration in clarified butter and to make quality assurance fast, reliable, and accessible. One of the main objectives is to provide a comprehensive AI-driven system that enables non-destructive quality assessment of clarified butter using computer vision, image analysis, and deep learning techniques. This ensures that adulteration can be identified and quantified without destructive chemical laboratory procedures. This preserves the integrity of the sample and allows for repeated analyses if necessary.Another important objective of the invention is to overcome the limitations of existing methods, such as the subjectivity in color interpretation in chemical tests like the Baudouin reaction, by introducing an automated, image-based analysis that increases precision and eliminates human error.

[0017] Another objective of the invention is to create a unique and structured dataset of high-resolution images of ghee samples, systematically linked to annotated physical and chemical parameters such as moisture content, ghee content, and vegetable oil composition. This robust database provides the foundation for training and refining deep learning models capable of detecting even subtle adulteration patterns. Furthermore, the invention aims to automate conventional chemical tests, such as the Baudouin test, through advanced image acquisition and AI interpretation, transforming a once time-consuming and expert-led process into an efficient and widely accessible analytical pipeline.

[0018] Another objective of the invention is the development of a user-friendly platform that can be used in various environments – from mobile applications for private use to industrial-grade devices for integration into production lines. This provides the system with scalability and adaptability, making it equally useful for regulators, manufacturers, retailers, and end users. The invention also aims to provide real-time reporting and connectivity, enabling stakeholders to make immediate quality control decisions and, if necessary, submit validated results to regulatory databases to improve compliance and traceability throughout the supply chain.

[0019] A further objective is to ensure the system's future viability through the integration of an adaptive learning framework that self-trains itself when new adulterations or deviations occur in ghee samples. This adaptability guarantees that the invention remains effective even in dynamic market conditions with constantly evolving adulteration practices. Furthermore, the invention aims to bridge the gap between high-precision laboratory methods and field-deployable solutions by providing a cost-effective, scalable, and portable technology that maintains scientific accuracy while being practical for everyday use.By combining these objectives, the invention aims to provide a groundbreaking solution that not only improves the detection and quantification of adulteration in clarified butter, but also promotes transparency, strengthens food safety standards, and restores consumer confidence in one of the most culturally significant dairy products. BRIEF DESCRIPTION OF THE FIGURE

[0020] These and other features, aspects, and advantages of the present invention will be better understood if the following detailed description is read with reference to the accompanying drawing, in which the same symbols consistently represent the same parts. The following applies: Fig. Figure 1 shows a block diagram of an AI-controlled system and device for quality assessment and adulteration detection in clarified butter.

[0021] Experts will also recognize that the elements in the drawing are shown for the sake of simplicity and are not necessarily to scale. For example, the flowcharts illustrate the process by highlighting the main steps to enhance understanding of the aspects of this disclosure. Furthermore, with regard to the design of the device, one or more components of the device may be represented in the drawing by conventional symbols, and the drawing may show only the specific details relevant to understanding the embodiments of this disclosure, so as not to clutter the drawing with details that are readily apparent to those skilled in the art after reading this description. Detailed description of the invention

[0022] For a better understanding of the inventive principles, reference is made below to the embodiment shown in the drawing, which is described in specific terminology. However, this does not limit the scope of the invention. Changes and further modifications of the illustrated system, as well as further applications of the inventive principles, are possible, as would normally occur to a person skilled in the art in the field of invention.

[0023] It is clear to the person skilled in the art that the preceding general description and the following detailed description are exemplary and explanatory of the invention and are not intended as a limitation of it.

[0024] References in this specification to “an aspect”, “another aspect”, or similar expressions mean that a particular feature, structure, or property described in connection with the embodiment is included in at least one embodiment of the present disclosure. Therefore, occurrences of the expressions “in one embodiment”, “in another embodiment”, and similar expressions in this specification may all refer to the same embodiment, but need not.

[0025] The terms "includes," "include," or other variations thereof are intended to cover non-exclusive inclusion, such that a process or method that includes a list of steps may not only contain those steps but may also include other steps not expressly listed or inherent in such process or method. Likewise, the statement "includes..." in the case of one or more devices, subsystems, elements, structures, or components does not, without further limitations, preclude the existence of other devices, subsystems, elements, structures, components, or additional devices, subsystems, elements, structures, or components.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by a person skilled in the art in the field of the invention. The systems, methods, and examples provided herein serve only for illustration and are not to be construed as limitations.

[0027] Embodiments of the present disclosure are described in detail below with reference to the attached drawing.

[0028] In Fig.Figure 100 is a block diagram of an AI-driven system and device for quality assessment and adulteration detection in clarified butter. The system comprises: a data acquisition and processing unit (102) with a multispectral imaging unit (102a) configured to acquire high-resolution images of ghee samples under controlled illumination conditions. The module also includes a temperature-stabilized sample chamber with optically transparent walls to ensure uniform light scattering and minimize shadow artifacts; a chemical measurement subsystem (104) capable of performing at least one conventional test, including gravimetric moisture determination, Baudouin reaction, or spectroscopic profiling, with the subsystem operationally connected to a microfluidic cartridge (104a) for sample handling and reagent delivery;an annotated data archive (106) configured to store captured images along with chemical and physical adulteration measurements, the archive comprising a structured relational database encoding metadata such as batch number, sample origin, illumination parameters, and adulteration metrics; a deep learning inference machine (108) comprising convolutional neural networks (CNNs) (108a) selected from the group consisting of ResNet, EfficientNet, or VGG, the machine being trained on the annotated data store to detect adulteration signatures, including variations in color gradients, crystalline microstructures, and spectral reflectance profiles;an automated Baudouin test analysis pipeline (110) in which sequential image frames of the reaction chamber are captured and processed using a color segmentation technique in conjunction with AI-driven classification to quantify adulteration of clarified butter with reduced subjectivity; and a user interface (112) integrated into a mobile device, industrial console, or cloud-accessible platform and configured to generate non-destructive, real-time adulteration reports, the reports including predicted levels of adulteration, confidence metrics, and regulatory compliance thresholds.

[0029] In one embodiment, the multispectral imaging unit (102a) comprises a CCD or CMOS sensor with a sensitivity covering visible and near-infrared wavelengths between 400 and 1000 nm. The imaging unit is surrounded by a programmable LED illumination array arranged in a ring geometry to provide adjustable spectral intensity profiles, thus enabling amplification of the spectral bands associated with distortions during acquisition.

[0030] In one embodiment, the temperature-stabilized sample chamber maintains a uniform temperature in the range of 20-25 °C by means of a thermoelectric Peltier module coupled with a closed feedback sensor, thus minimizing crystallization artifacts and viscosity-related texture changes in ghee samples that could otherwise distort image analysis.

[0031] In one embodiment, the microfluidic cartridge (104a) for performing the Baudouin reaction comprises a serpentine channel architecture that enables controlled mixing of ghee, hydrochloric acid, and furfural reagent. The cartridge also includes integrated optical windows aligned with the imaging unit to capture sequential reaction images without manual intervention.

[0032] In one embodiment, the annotated data repository (106) is configured such that each sample entry is encoded with synchronized timestamps that link image data and results of chemical analyses. The repository is optimized for training datasets by implementing feature extraction pipelines that store spatial frequency maps, grayscale coincidence matrices, and Fourier transform-based spectral signatures corresponding to the degrees of distortion.

[0033] In one embodiment, the deep learning engine based on a convolutional neural network (108) is implemented on a hardware accelerator selected from GPU, TPU, or FPGA. The engine is trained using supervised backpropagation with annotated ghee datasets and further optimized by applying transfer learning techniques from pretrained food image recognition models, thereby reducing the convergence time and improving the recognition accuracy for minor adulterations.

[0034] In one embodiment, the automated Baudouin test analysis pipeline (110) comprises a chromaticity-based segmentation technique configured to map hue shifts in the CIE Lab color space, coupled with a recurrent neural network that processes the temporal evolution of color changes across successive frames, thus enabling precise quantification of adulteration in clarified butter even at concentrations below 2%.

[0035] In one embodiment, the user interface (112) includes an adaptive reporting framework that not only displays levels of falsification but also generates compliance alerts. The interface is configured to export encrypted result files via a wireless communication protocol such as Wi-Fi, Bluetooth, or LTE, thus enabling direct integration into supply chain traceability systems and regulatory audit databases.

[0036] In one embodiment, the system is also configured with an incremental learning protocol that allows the CNN engine to be retrained after the acquisition of new forgery datasets. The protocol incorporates a federated learning model that distributes training across multiple decentralized devices without exchanging raw image data, thus ensuring both model adaptability and data privacy.

[0037] In one embodiment, the system is designed as a portable handheld device, housing the imaging unit, illumination array, microfluidic cartridge, and integrated GPU accelerator in a single enclosure. The device is powered by a rechargeable lithium-ion battery pack that enables at least four hours of continuous testing and also features a touchscreen interface for on-site counterfeit detection by regulators and consumers.

[0038] The present invention provides a comprehensive system for AI-driven quality assessment and adulteration detection in clarified butter. The detailed description of the invention focuses on its architecture, operational modules, and underlying techniques, which together enable precise, rapid, and non-destructive analysis. The system is based on an integrated framework that combines optical imaging hardware, microfluidic chemical tests, annotated data management, and advanced artificial intelligence techniques, thus ensuring a robust solution that overcomes the shortcomings of conventional methods for detecting ghee adulteration.

[0039] The imaging module forms a crucial basis of the invention. It is based on a multispectral imaging unit with a highly sensitive CCD or CMOS sensor capable of capturing information in the visible and near-infrared range between 400 and 1000 nanometers. This spectral range was selected because adulterants such as clarified butter and vegetable oils exhibit slight deviations in their absorption and reflection profiles compared to pure clarified butter. To ensure consistent image acquisition, the imaging unit is embedded in a temperature-stabilized chamber that maintains sample conditions between 20 and 25 °C using a thermoelectric Peltier module. This feature prevents fluctuations caused by crystallization or viscosity changes in the ghee, phenomena that could otherwise complicate image-based adulteration analysis.The chamber walls are optically transparent and surrounded by a circularly arranged, programmable LED lighting system. This lighting system offers controlled light diffusion and spectral tunability, allowing specific falsification signatures to be emphasized by increasing spectral contrast in targeted wavelength ranges.

[0040] In parallel with image acquisition, the system integrates a chemical analysis subsystem that performs tests such as gravimetric moisture determination, spectroscopy for determining fatty acid composition, and the Baudouin reaction for identifying clarified butter. A microfluidic cartridge was developed to miniaturize and automate the Baudouin test. This cartridge utilizes a serpentine channel architecture that ensures precise mixing of ghee, hydrochloric acid, and furfural reagents under controlled flow conditions. Optical windows aligned with the imaging unit allow for sequential acquisition of the reaction progress without manual intervention, thus eliminating subjective interpretation by human operators. The imaging subsystem records color transitions in real time, and these image frames are forwarded to the engine for automatic interpretation.

[0041] The annotated data repository serves as the basis for training and validating the AI ​​models. It is structured as a relational database in which each entry links high-resolution images with experimentally determined adulteration values ​​such as moisture content, ghee concentration, and vegetable oil content. Each entry also encodes metadata, including the geographic origin of the sample, batch number, lighting configuration, and environmental parameters during testing. In addition to storing the raw images, the repository is enriched with pre-extracted features, such as grayscale coexistence matrices (GLCM) for texture analysis, Fourier transform coefficients for frequency domain representation, and spatial frequency maps. This combination of raw and feature data generates a multidimensional dataset that is crucial for training robust machine learning techniques.

[0042] The deep learning inference engine forms the analytical core of the invention. It is based on convolutional neural networks, including architectures such as ResNet, EfficientNet, and VGG, implemented on hardware accelerators like GPUs, TPUs, or FPGAs to support high-throughput computations. The technique works by first applying preprocessing routines to normalize lighting variations and remove background noise. The images are then passed through convolutional layers that automatically learn low-level features such as edges, gradients, and color intensity distributions. Higher-level layers extract abstract, distortion-specific representations, including crystallization patterns, microstructural irregularities, and fine-grained chromatic aberrations.The CNN models are trained using supervised learning protocols, with annotated adulteration labels guiding backpropagation and weight optimization. Transfer learning strategies are employed, leveraging pre-trained models on large food or natural image datasets. This accelerates convergence and improves accuracy when the number of adulterated datasets is limited. Performance metrics such as precision, recall, and F1 score are continuously monitored during training to ensure the model's generalizability.

[0043] A unique pipeline is dedicated to automating the Baudouin test. In this pipeline, sequential reaction images are mapped into the CIE Lab color space, which is perceptually uniform and therefore responds to subtle hue variations. Chromaticity-based segmentation techniques isolate the area of ​​interest where the color reaction occurs, and the temporal evolution of hue intensity is modeled using a recurrent neural network (RNN). Particularly well-suited for sequential data, the RNN analyzes time-dependent color transitions and classifies the degree of adulteration of clarified butter with an accuracy that surpasses that of manual testing.The combination of spatial CNN-based feature extraction and temporal RNN-based analysis creates a hybrid model that can detect adulteration even at concentrations as low as 1-2% - concentrations that conventional, human-performed Baudouin tests often miss.

[0044] The user interface module consolidates the results into a clear reporting framework. Falsification levels are displayed along with confidence levels, and the interface can trigger alerts if the results exceed legal limits. The reporting engine can be adapted to various deployment environments, whether as a mobile application for consumers, an industrial console integrated into production lines, or a cloud platform for regulators. Connectivity is enabled via wireless protocols such as Wi-Fi, LTE, or Bluetooth, and all result files are encrypted to ensure data security when integrated into supply chain transparency systems. This digital integration allows for seamless data export to regulators, thereby improving traceability and compliance.

[0045] The system's adaptability is ensured by its incremental learning protocol. As new forgery samples are added to the repository, the deep learning engine can undergo training cycles to enhance its detection capabilities. A federated learning framework can also be implemented, allowing distributed devices to contribute to model training without centralizing raw image data. This improves data privacy and scalability, ensuring the system can evolve in parallel with new forgery practices. By supporting incremental updates, the invention remains resilient to evolving threats while minimizing downtime and training costs.

[0046] The portable version of the invention integrates all functional modules into a handheld device. It houses the imaging unit, LED illumination array, microfluidic cartridge, and GPU accelerator in a single enclosure. A rechargeable lithium-ion battery ensures mobility and enables multiple field tests. An integrated touchscreen allows for user interaction, calibration, and immediate reporting, making it extremely convenient for regulators, small-scale producers, and even informed consumers. In industrial configurations, the same system can be integrated as an inline inspection machine connected to automated sampling arms, enabling batch-by-batch adulteration testing without production downtime.

[0047] The described system achieves a technological advancement through the integration of optics, microfluidics, annotated databases, and AI techniques into a unified pipeline. The innovations, particularly the hybrid CNN-RNN approach for sequential test analysis and the use of multispectral imaging for detecting subtle adulterations, create a system that surpasses conventional laboratory methods in accuracy and scalability. By combining precision with portability and adaptability, the invention establishes itself as a transformative platform for ensuring the authenticity, safety, and regulatory compliance of clarified butter across all levels of the supply chain.

[0048] The drawing and the preceding description show examples of embodiments. Those skilled in the art will recognize that one or more of the described elements can be combined to form a single functional element. Alternatively, certain elements can be divided into several functional elements. Elements of one embodiment can be added to another embodiment. For example, the sequence of the processes described here can be changed and is not limited to the manner described here. Furthermore, the actions of a flowchart need not be implemented in the sequence shown; nor does it necessarily have to be performed by all actions. Actions that are not dependent on other actions can also be performed in parallel with the other actions. The scope of the embodiments is in no way limited by these specific examples.Numerous variations are possible, whether explicitly stated in the specification or not, such as differences in structure, dimensions, and material use. The range of embodiments is at least as broad as specified in the following claims.

[0049] Advantages, further benefits, and problem solutions have been described above with reference to specific embodiments. However, the advantages, benefits, problem solutions, and all components that can lead to an advantage, benefit, or solution occurring or becoming more apparent are not to be construed as critical, necessary, or essential features or components of individual or all claims. REFERENCES 100 An AI-supported system and device for quality assessment and detection of adulteration in clarified butter. 102 Data acquisition and processing unit 102a Multispectral Imaging Unit 104 Subsystem for Chemical Measurement 104a Microfluidic cartridge 106 Annotated Data Repository 108 deep learning inference engine 108a Convolutional neural networks (CNNs) 110 Automated Baudouin Test Analysis Pipeline 112 User interface

Claims

[1] A system for AI-driven quality assessment and adulteration detection in clarified butter, the system includes: a data acquisition and processing unit comprising a multispectral imaging unit configured to capture high-resolution images of ghee samples under controlled illumination conditions, the module also comprising a temperature-stabilized sample chamber with optically transparent walls to ensure uniform light scattering and minimize shadow artifacts; a chemical measurement subsystem capable of performing at least one conventional analysis, including gravimetric moisture determination, Baudouin reaction or spectroscopic profiling, wherein the subsystem is operationally connected to a microfluidic cartridge for sample handling and reagent dispensing; An annotated data repository configured to store captured images along with chemical and physical adulteration measurements. The repository includes a structured relational database encoding metadata, including batch number, sample origin, illumination parameters, and adulteration metrics. a deep learning inference machine incorporating convolutional neural networks (CNNs), where the machine is trained on the annotated data store to detect forgery signatures, including variations in color gradients, crystalline microstructures, and spectral reflectance profiles; an automated Baudouin test analysis pipeline in which sequential image frames of the reaction chamber are captured and processed using a color segmentation technique in conjunction with AI-driven classification to quantify the adulteration of clarified butter with reduced subjectivity; and A user interface integrated into a mobile device, an industrial console, or a cloud-accessible platform, configured to generate non-destructive, real-time reports of counterfeiting, with these reports including predicted counterfeit levels, confidence metrics, and regulatory compliance thresholds. [2] System according to claim 1, wherein the multispectral imaging unit comprises a CCD or CMOS sensor with a sensitivity in the visible and near-infrared range between 400 and 1000 nm, wherein the imaging unit is surrounded by a programmable LED illumination array arranged in a ring geometry to provide adjustable spectral intensity profiles. [3] System according to claim 1, wherein the temperature-stabilized sample chamber maintains a uniform temperature in the range of 20-25 °C by means of a thermoelectric Peltier module coupled with a closed feedback sensor, thereby minimizing crystallization artifacts and viscosity-induced texture changes in ghee samples which could otherwise distort image analysis. [4] System according to claim 1, wherein the microfluidic cartridge for carrying out the Baudouin reaction comprises a serpentine channel architecture enabling controlled mixing of ghee, hydrochloric acid and furfural reagent, the cartridge further comprising integrated optical windows aligned with the imaging unit to capture sequential reaction images without manual intervention [5] System according to claim 1, wherein the annotated data repository is configured such that each sample entry is encoded with synchronized timestamps linking image data and results of chemical analyses, wherein the repository is optimized for training datasets by implementing feature extraction pipelines that store spatial frequency maps, grayscale coincidence matrices and Fourier transform-based spectral signatures corresponding to the degrees of distortion. [6] System according to claim 1, wherein the deep learning engine based on a convolutional neural network is implemented on a hardware accelerator selected from GPU, TPU, or FPGA. The engine is trained using supervised backpropagation with annotated ghee datasets and further optimized by applying transfer learning techniques from pretrained food image recognition models, thereby reducing convergence time and improving recognition accuracy in the presence of minor adulterations. [7] System according to claim 1, wherein the automated Baudouin test analysis pipeline comprises a chromaticity-based segmentation technique configured to map hue shifts in the CIE Lab color space, coupled with a recurrent neural network that processes the temporal evolution of color changes across successive frames, thus enabling precise quantification of adulteration by clarified butter even at concentrations below 2%. [8] System according to claim 1, wherein the user interface comprises an adaptive reporting framework that not only displays levels of adulteration but also generates compliance alerts, wherein the interface is configured to export encrypted result files via a wireless communication protocol such as Wi-Fi, Bluetooth or LTE, thus enabling direct integration into supply chain traceability systems and databases for regulatory audits. [9] System according to claim 1, wherein the system is further configured with an incremental learning protocol that enables retraining of the CNN engine upon acquisition of new fakery datasets, wherein the protocol comprises a federated learning model that distributes the training across multiple decentralized devices without exchanging raw image data, thereby ensuring both model adaptability and data privacy.

Images