Machine vision-based method for identifying counterfeit coconut oil

Abstract

The application discloses a coconut oil anti-counterfeiting identification method based on machine vision and relates to the technical field of food detection, which comprises the following steps: acquiring and fusing time sequence visual features and near-infrared spectral features by using high-resolution multispectral integrated detection technology, constructing a multi-branch fusion identification model by using a convolutional neural network and a self-attention network, extracting phase change time sequence features and performing global correlation analysis, obtaining multi-dimensional fusion anti-counterfeiting features, performing homologous time sequence binding verification on a phase change standard feature library, constructing a mapping relationship between visual and spectral features and eliminating invalid data, obtaining an effective anti-counterfeiting feature set and performing correlation chain anchoring and encryption processing, generating encrypted anti-counterfeiting feature credentials and standardizing storage, performing local judgment by using an edge cloud collaborative architecture, analyzing adulteration abnormal rules and updating a dynamic anti-counterfeiting feature library. The application improves the anti-counterfeiting identification precision and anti-counterfeiting capability by using high-resolution multispectral detection technology and visual and spectral feature fusion of the whole cycle of solid-liquid phase change of coconut oil.

Description

Technical Field

[0001] This invention relates to the field of food testing technology, and more specifically, to a machine vision-based method for identifying counterfeit coconut oil. Background Technology

[0002] Coconut oil, as a highly sought-after premium edible vegetable oil in recent years, combines nutritional advantages and health benefits with its unique medium-chain fatty acid structure, natural antioxidants, and refreshing flavor. It is favored by health-conscious consumers, baking enthusiasts, and the high-end catering market, resulting in significant market premium potential and continuously rising consumer demand.

[0003] However, existing methods for identifying counterfeit coconut oil are limited in scope, relying mainly on single machine vision appearance detection or single spectral component detection. These methods can only capture the static, single-point-in-time characteristics of coconut oil, making it difficult to cover the dynamic evolution of the entire solid-liquid phase transition cycle. As a result, they are easily circumvented by means of appearance counterfeiting and ingredient blending, resulting in insufficient reliability of anti-counterfeiting measures.

[0004] Furthermore, when it comes to anti-counterfeiting of coconut oil, most point-in-time / time-series feature verification mechanisms are missing. There is no established verification logic for the homogeneous time-series binding of visual features and spectral features, making it difficult to verify the time-series consistency and authenticity of feature data. This poses a risk of feature tampering, data forgery, and interference from invalid data, making it difficult to eliminate counterfeiting at its source.

[0005] In addition, the anti-counterfeiting identification of coconut oil lacks time-point / time-sequence coordination and dynamic adaptability. It mostly adopts offline and local detection modes and has not built a collaborative architecture of edge computing and cloud platform. As a result, it is difficult to achieve the linkage between local rapid judgment and cloud-based in-depth analysis. It is also impossible to continuously monitor the abnormal phase change linkage of adulterated coconut oil and dynamically update the anti-counterfeiting feature library. Consequently, it suffers from slow response and low identification efficiency in the face of new adulteration methods.

[0006] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0007] In response to the problems in related technologies, this invention proposes a machine vision-based method for identifying counterfeit coconut oil, thereby overcoming the aforementioned technical problems in existing related technologies.

[0008] Therefore, the specific technical solution adopted by the present invention is as follows: Firstly, the present invention proposes a machine vision-based anti-counterfeiting identification method for coconut oil, comprising: utilizing high-resolution multispectral integrated detection technology to acquire continuous temporal visual features of the oil's appearance and packaging texture, as well as near-infrared spectral composition features, from the critical temperature range of the solid-liquid phase transition of coconut oil, and fusing the acquired features to obtain a full-cycle anti-counterfeiting dataset of the sample under test; constructing a multi-branch fusion recognition model based on convolutional neural networks and self-attention networks, extracting phase transition temporal features and performing global correlation analysis on the full-cycle anti-counterfeiting dataset of the sample under test to obtain multi-dimensional fused anti-counterfeiting features; and, according to a pre-configured standard feature library of coconut oil phase transition, applying the multi-dimensional fused anti-counterfeiting features... The pseudo-features and the phase transition full-cycle anti-counterfeiting dataset are subjected to homogeneous time-series binding verification. The mapping relationship between the visual features and spectral composition features of temperature nodes is constructed and invalid data is removed to obtain an effective anti-counterfeiting feature set. The effective anti-counterfeiting feature set is anchored and encrypted with phase transition association chain to generate encrypted anti-counterfeiting feature vouchers. The encrypted anti-counterfeiting feature vouchers are then stored in a standardized manner to obtain a standardized encrypted anti-counterfeiting feature library. Using a collaborative architecture of edge computing and cloud platform, the standardized encrypted anti-counterfeiting feature library is used to determine phase transition features. Based on the determination results, key feature data is extracted and uploaded to the cloud. The cloud is used to deeply analyze the abnormal phase transition linkage of adulterated coconut oil, construct and update a dynamic anti-counterfeiting feature library, so as to realize machine vision for multi-dimensional full-cycle anti-counterfeiting identification of coconut oil.

[0009] Furthermore, utilizing high-resolution multispectral integrated detection technology, visual features of the oil's appearance and packaging texture, as well as near-infrared spectral composition features, are obtained from the critical temperature range of the solid-liquid phase transition of coconut oil, capturing continuous temporal data throughout the entire phase transition process. These features are then fused to obtain a full-cycle anti-counterfeiting dataset for the sample under test. This includes: coaxial calibration of the high-resolution multispectral integrated detection technology based on the solid-liquid phase transition characteristics of coconut oil and the preset critical temperature range; placing the coconut oil sample under test within the calibrated temperature-controlled area, with the gradient temperature control terminal controlling the temperature rise and fall of the sample according to the preset temperature gradient step size to obtain feature data for each stage of the phase transition; synchronously acquiring visual features of the oil's appearance and packaging texture, as well as near-infrared spectral composition features, at each preset temperature node; filtering out interference noise and performing image alignment and spectral baseline correction on the visual and spectral composition features; and performing homogeneous registration and fusion of multi-source features at the same temperature node to form a temporally continuous multi-dimensional fused data dataset, thus obtaining the full-cycle anti-counterfeiting dataset for the sample under test.

[0010] Furthermore, a multi-branch fusion recognition model is constructed based on convolutional neural networks and self-attention networks. Phase transition temporal features and global correlation analysis are performed on the phase transition full-cycle anti-counterfeiting dataset of the test sample to obtain multi-dimensional fused anti-counterfeiting features. This includes: constructing a multi-branch fusion recognition model, wherein the model includes convolutional neural network branches and self-attention network branches; inputting the phase transition full-cycle anti-counterfeiting dataset of the test sample into the input layer of the multi-branch fusion recognition model, and performing temporal dimension standardization on the dataset; and using the convolutional neural network branches to perform appearance analysis on the standardized phase transition full-cycle anti-counterfeiting dataset. Temporal feature extraction is performed to obtain the temporal pattern of changes in the visual features of oil appearance and packaging texture with temperature nodes throughout the phase transition process, resulting in an appearance temporal feature vector. Based on the self-attention mechanism in the self-attention network branch, spectral temporal features are extracted from the standardized phase transition full-cycle anti-counterfeiting dataset to obtain the linkage pattern of spectral features with temperature nodes, resulting in a spectral temporal feature vector. The appearance temporal feature vector and the spectral temporal feature vector are input to the feature fusion layer at the output of the multi-branch fusion recognition model. The two-branch features are fused using a preset fusion strategy to obtain multi-dimensional fused anti-counterfeiting features.

[0011] Furthermore, the preset fusion strategy includes: feature weight allocation and vector concatenation fusion; the feature weight allocation is used to assign preset weights to the appearance time-series feature vector and the spectral time-series feature vector respectively; the vector concatenation fusion is used to concatenate the two types of feature vectors after weighting to form a comprehensive recognition vector.

[0012] Furthermore, the pre-configured coconut oil phase change standard feature library includes: standard appearance time-series features, standard spectral features, and linkage matching rules; wherein, the standard appearance time-series features are used as the benchmark features for homogeneous binding verification; the standard spectral features are used to ensure the consistency of spectral data during the phase change process; and the linkage matching rules are used to clarify the correspondence between appearance and spectral features at each temperature node.

[0013] Furthermore, based on the pre-configured coconut oil phase change standard feature library, the multi-dimensional fused anti-counterfeiting features and the phase change full-cycle anti-counterfeiting dataset are subjected to homogeneous temporal binding verification. A mapping relationship between the visual features and spectral component features of temperature nodes is constructed, and invalid data is removed to obtain an effective anti-counterfeiting feature set. This includes: temporal alignment of the multi-dimensional fused anti-counterfeiting features and the phase change full-cycle anti-counterfeiting dataset based on the pre-configured coconut oil phase change standard feature library; comparison of the multi-dimensional fused anti-counterfeiting features and near-infrared spectral features based on the coconut oil phase change standard feature library to identify abnormal data that does not conform to the phase change law; construction of the mapping relationship between temperature nodes and visual and spectral features, and screening of effective feature data, removing invalid and abnormal data to obtain feature information that conforms to the phase change law; and integration of the screened effective feature information to obtain an effective anti-counterfeiting feature set.

[0014] Furthermore, the effective anti-counterfeiting feature set is anchored and encrypted using phase change association chains to generate encrypted anti-counterfeiting feature credentials. These credentials are then stored in a standardized manner to obtain a standardized encrypted anti-counterfeiting feature library. This process includes: extracting appearance time-series features, spectral time-series features, and linkage correspondences at each temperature node based on the effective anti-counterfeiting feature set; anchoring the extracted features using association chains based on the phase change time sequence to construct phase change feature association chains; encrypting the phase change feature association chains using a constraint encryption mechanism and protecting them against tampering based on identity binding, permission constraints, and integrity verification to generate encrypted anti-counterfeiting feature credentials; standardizing the format and unifying the index of the encrypted anti-counterfeiting feature credentials; and storing the standardized encrypted anti-counterfeiting feature credentials in a structured manner to obtain a standardized encrypted anti-counterfeiting feature library.

[0015] Furthermore, utilizing a collaborative architecture of edge computing and cloud platform, phase transition characteristics are determined in a standardized encrypted anti-counterfeiting feature library. Based on the determination results, key feature data is extracted and uploaded to the cloud. The cloud platform is then used for in-depth analysis of the abnormal phase transition patterns of adulterated coconut oil, constructing and updating a dynamic anti-counterfeiting feature library. This enables machine vision to perform multi-dimensional, full-cycle anti-counterfeiting identification of coconut oil. This includes: constructing a collaborative architecture of edge computing and cloud platform, where the edge is responsible for local data determination, and the cloud is responsible for in-depth analysis and feature library updates; and, based on the edge, retrieving the standardized encrypted anti-counterfeiting feature library and determining phase transition characteristics, verifying the features within the library. The integrity, temporal consistency, and validity of encrypted anti-counterfeiting feature certificates are assessed. Based on local edge-end judgment results, key feature data meeting the judgment criteria are screened and extracted, and the extracted key feature data is standardized and regularized. Using the collaborative transmission channel between edge computing and the cloud platform, the regularized key feature data is securely uploaded to the cloud. In-depth analysis is performed on the key feature data uploaded to the cloud to analyze the abnormal linkage patterns between visual and spectral features of adulterated coconut oil throughout the solid-liquid phase transition cycle. Based on the abnormal phase transition linkage patterns, a dynamic anti-counterfeiting feature library is constructed. Based on the key feature data, the dynamic anti-counterfeiting feature library is dynamically iterated and updated to obtain the updated dynamic anti-counterfeiting feature library.

[0016] Furthermore, by utilizing the collaborative transmission channel between edge computing and the cloud platform, the standardized key feature data is securely uploaded to the cloud. This includes: performing integrity checks on the standardized key feature data before transmission, and activating the secure transmission mode of the collaborative transmission channel between edge computing and the cloud platform; using an encrypted transmission mechanism to encrypt the key feature data at the transmission layer, uploading the encrypted key feature data to the cloud, and monitoring the transmission status in real time to investigate transmission interruptions and data loss anomalies; receiving the key feature data through the cloud, using a data verification mechanism to perform integrity verification on the received data, and temporarily archiving the key feature data in the cloud based on the verification results.

[0017] Secondly, this invention proposes a machine vision-based anti-counterfeiting identification system for coconut oil, comprising: a visual spectral acquisition module, used to acquire, using high-resolution multispectral integrated detection technology, the visual features of the oil's appearance and packaging texture, as well as near-infrared spectral composition features, in a continuous temporal sequence throughout the phase transition of coconut oil from the critical temperature range of the solid-liquid phase transition, and to fuse the acquired features to obtain a phase transition full-cycle anti-counterfeiting dataset of the sample to be tested; a visual feature extraction module, used to construct a multi-branch fusion recognition model based on convolutional neural networks and self-attention networks, to extract phase transition temporal features and perform global correlation analysis on the phase transition full-cycle anti-counterfeiting dataset of the sample to be tested, to obtain multi-dimensional fused anti-counterfeiting features; and an anti-counterfeiting feature verification module, used to verify the multi-dimensional fused anti-counterfeiting features according to a pre-configured coconut oil phase transition standard feature library. The system performs homogeneous time-series binding verification on the phase change full-cycle anti-counterfeiting dataset, constructs a mapping relationship between the visual features of temperature nodes and spectral component features, and removes invalid data to obtain a valid anti-counterfeiting feature set. The anti-counterfeiting certificate encryption module is used to anchor and encrypt the phase change association chain of the valid anti-counterfeiting feature set, generating encrypted anti-counterfeiting feature certificates. These encrypted anti-counterfeiting feature certificates are then stored in a standardized manner to obtain a standardized encrypted anti-counterfeiting feature library. The oil sample anti-counterfeiting identification module utilizes a collaborative architecture of edge computing and a cloud platform to determine the phase change features of the standardized encrypted anti-counterfeiting feature library. Based on the determination results, key feature data is extracted and uploaded to the cloud. The cloud-based deep analysis of the abnormal phase change linkage patterns of adulterated coconut oil is used to construct and update a dynamic anti-counterfeiting feature library, enabling machine vision to perform multi-dimensional, full-cycle anti-counterfeiting identification of coconut oil.

[0018] The beneficial effects of this invention are as follows: 1) This invention utilizes high-resolution multispectral integrated detection technology, combined with the fusion of visual and spectral features throughout the entire solid-liquid phase change cycle of coconut oil, to achieve multi-dimensional anti-counterfeiting of coconut oil throughout the entire process from collection and verification to identification. This breaks through the limitations of traditional single visual and spectral detection, and improves the accuracy of anti-counterfeiting identification and anti-counterfeiting capabilities. 2) This invention solves the pain points of existing anti-counterfeiting technologies, such as easy tampering of features, easy forgery of data, and easy misalignment of timing, by using the same source time sequence binding verification and phase change association chain anchoring encryption mechanism. At the same time, by eliminating invalid data, it ensures the authenticity and effectiveness of anti-counterfeiting features and avoids the risk of adulteration and counterfeiting. 3) This invention achieves the linkage between rapid local judgment and in-depth cloud analysis through the collaborative architecture of edge computing and cloud platform, and builds a dynamically iterative and updated anti-counterfeiting feature library, thereby improving the efficiency of on-site identification and continuously mining the abnormal patterns of adulteration, realizing the intelligent, efficient and traceable whole cycle of coconut oil anti-counterfeiting identification. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the 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 of a machine vision-based anti-counterfeiting identification method for coconut oil according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a machine vision-based anti-counterfeiting identification system for coconut oil according to an embodiment of the present invention. Detailed Implementation

[0021] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0022] According to an embodiment of the present invention, a machine vision-based method for identifying counterfeit coconut oil is proposed.

[0023] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, the machine vision-based anti-counterfeiting identification method for coconut oil according to an embodiment of the present invention includes: Step S1: Using high-resolution multispectral integrated detection technology, the visual features of the oil surface and packaging texture and the near-infrared spectral composition features of the coconut oil in the continuous time sequence of the phase change critical temperature range are obtained. The obtained features are then fused to obtain the phase change full-cycle anti-counterfeiting dataset of the sample to be tested. Step S2: Construct a multi-branch fusion recognition model based on convolutional neural network and self-attention network, extract phase transition time series features and perform global correlation analysis on the phase transition full-cycle anti-counterfeiting dataset of the sample to be tested, and obtain multi-dimensional fusion anti-counterfeiting features; Step S3: Based on the pre-configured coconut oil phase change standard feature library, perform homogeneous time-series binding verification on the multi-dimensional fusion anti-counterfeiting features and the phase change full-cycle anti-counterfeiting dataset, construct the mapping relationship between the visual features of temperature nodes and the spectral component features, and remove invalid data to obtain an effective anti-counterfeiting feature set. Step S4: Perform phase change association chain anchoring and encryption processing on the effective anti-counterfeiting feature set to generate encrypted anti-counterfeiting feature vouchers, and standardize the storage of encrypted anti-counterfeiting feature vouchers to obtain a standardized encrypted anti-counterfeiting feature library. Step S5: Utilize the collaborative architecture of edge computing and cloud platform to determine the phase transition characteristics of the standardized encrypted anti-counterfeiting feature library, extract key feature data based on the determination results and upload it to the cloud, and use the cloud to deeply analyze the abnormal phase transition linkage of adulterated coconut oil, construct and update the dynamic anti-counterfeiting feature library, so as to realize the multi-dimensional and full-cycle anti-counterfeiting identification of coconut oil by machine vision.

[0024] In this optional embodiment, high-resolution multispectral integrated detection technology is used to acquire continuous temporal visual features of the oil's appearance and packaging texture, as well as near-infrared spectral composition features, throughout the entire phase transition process of coconut oil from the critical temperature range of the solid-liquid phase transition. The acquired features are then fused to obtain a phase transition full-cycle anti-counterfeiting dataset for the sample under test. This includes: coaxially calibrating the high-resolution multispectral integrated detection technology based on the solid-liquid phase transition characteristics of coconut oil and a preset critical temperature range for phase transition; placing the coconut oil sample under test within the calibrated temperature control area, and using a gradient temperature control terminal to control the temperature rise and fall of the sample according to a preset temperature gradient step size to obtain feature data for each stage of the phase transition; simultaneously acquiring visual features of the oil's appearance and packaging texture, as well as near-infrared spectral composition features, at each preset temperature node; filtering out interference noise from the visual features and spectral composition features and completing image alignment and spectral baseline correction; and performing homogeneous registration and fusion of multi-source features at the same temperature node to form temporally continuous multi-dimensional fused data, thus obtaining a phase transition full-cycle anti-counterfeiting dataset for the sample under test.

[0025] Specifically, based on the solid-liquid phase transition characteristics of coconut oil, the critical solid-liquid temperature range was set to 18-28℃. The high-resolution multispectral integrated detection equipment was coaxially calibrated to ensure that the detection accuracy reached ±0.5℃ and the spectral resolution ≥1nm. The coconut oil sample to be tested was placed in the calibrated constant temperature detection area. The sample was heated and cooled by the gradient temperature control terminal at a temperature gradient step of 1℃ / min to accurately capture the complete phase transition process from solid to liquid and back to solid.

[0026] At six temperature nodes of 18℃, 20℃, 22℃, 24℃, 26℃, and 28℃, visual features such as the appearance of the oil and the texture of the packaging, as well as the near-infrared spectral composition features at the corresponding temperatures, were collected. After the collection was completed, noise filtering was performed on the two types of feature data to remove environmental interference signals and to complete image alignment and spectral baseline correction. Then, the visual features and spectral features at the same temperature node were registered and fused to obtain the phase transition full-cycle anti-counterfeiting dataset of the sample to be tested.

[0027] In this optional embodiment, a multi-branch fusion recognition model is constructed based on convolutional neural networks and self-attention networks. Phase transition temporal features are extracted and global correlation analysis is performed on the phase transition full-cycle anti-counterfeiting dataset of the test sample to obtain multi-dimensional fusion anti-counterfeiting features. This includes: constructing a multi-branch fusion recognition model, wherein the model includes convolutional neural network branches and self-attention network branches; inputting the phase transition full-cycle anti-counterfeiting dataset of the test sample into the input layer of the multi-branch fusion recognition model, and performing temporal dimension standardization on the dataset; and using the convolutional neural network branches to process the standardized phase transition full-cycle anti-counterfeiting dataset. The appearance time-series feature extraction method obtains the temporal pattern of changes in the visual features of oil appearance and packaging texture with temperature nodes throughout the phase transition process, resulting in an appearance time-series feature vector. Based on the self-attention mechanism in the self-attention network branch, spectral time-series feature extraction is performed on the standardized phase transition full-cycle anti-counterfeiting dataset to obtain the linkage pattern of spectral features with temperature nodes, resulting in a spectral time-series feature vector. The appearance time-series feature vector and the spectral time-series feature vector are input into the feature fusion layer at the output of the multi-branch fusion recognition model. The pre-set fusion strategy is used to fuse the two-branch features to obtain multi-dimensional fused anti-counterfeiting features.

[0028] In this optional embodiment, the preset fusion strategy includes: feature weight allocation and vector concatenation fusion; the feature weight allocation is used to assign preset weights to the appearance time-series feature vector and the spectral time-series feature vector respectively; the vector concatenation fusion is used to concatenate the two types of feature vectors after weighting to form a comprehensive recognition vector.

[0029] Specifically, a multi-branch fusion recognition model is constructed, which includes a convolutional neural network branch and a self-attention network branch. The phase transition full-cycle anti-counterfeiting dataset of the sample to be tested is input into the model input layer. The dataset is processed by time-series dimension standardization, with the time length unified to 100 frames and the feature dimension normalized to the [0,1] interval. Using the convolutional neural network branch, the appearance time-series features are extracted from the standardized dataset. A structure of 3 convolutional layers and 1 pooling layer is adopted to obtain the time-series pattern of the visual features of oil appearance and packaging texture changing with temperature nodes (e.g., 18-28℃, 1℃ interval) during the entire phase transition process, resulting in an appearance time-series feature vector with a dimension of 1×256.

[0030] Based on the self-attention mechanism in the self-attention network branch, with 4 attention heads, spectral temporal features are extracted from the standardized dataset to obtain the linkage pattern of near-infrared spectral features (e.g., 400-1700nm band) with temperature nodes, resulting in a spectral temporal feature vector of dimension 1×256. The appearance temporal feature vector and the spectral temporal feature vector are input to the feature fusion layer at the model output. A fusion strategy is adopted, assigning preset weights to the two types of feature vectors respectively, for example, a weight of 0.6 for the appearance temporal feature vector and a weight of 0.4 for the spectral temporal feature vector. The two weighted feature vectors are then concatenated to form a comprehensive recognition vector of dimension 1×512, resulting in a multi-dimensional fused anti-counterfeiting feature.

[0031] In this optional embodiment, the pre-configured coconut oil phase change standard feature library includes: standard appearance time-series features, standard spectral features, and linkage matching rules; wherein, the standard appearance time-series features are used as the benchmark features for homogeneous binding verification; the standard spectral features are used to ensure the consistency of spectral data during the phase change process; and the linkage matching rules are used to clarify the correspondence between appearance and spectral features at each temperature node.

[0032] In this optional embodiment, based on a pre-configured coconut oil phase change standard feature library, the multi-dimensional fused anti-counterfeiting features and the phase change full-cycle anti-counterfeiting dataset are subjected to homogeneous temporal binding verification. A mapping relationship between the visual features and spectral component features of temperature nodes is constructed, and invalid data is removed to obtain an effective anti-counterfeiting feature set. This includes: temporal alignment of the multi-dimensional fused anti-counterfeiting features and the phase change full-cycle anti-counterfeiting dataset based on the pre-configured coconut oil phase change standard feature library; comparison of the multi-dimensional fused anti-counterfeiting features and near-infrared spectral features based on the coconut oil phase change standard feature library to identify abnormal data that does not conform to the phase change law; construction of a mapping relationship between temperature nodes and visual and spectral features, and screening of effective feature data, removing invalid and abnormal data to obtain feature information that conforms to the phase change law; and integration of the screened effective feature information to obtain an effective anti-counterfeiting feature set.

[0033] Specifically, based on the pre-configured coconut oil phase change standard feature library, which includes standard appearance time-series features, standard spectral features, and linkage matching rules, the standard appearance time-series features are set such that the feature parameter fluctuation does not exceed 5% for every 1°C increase in temperature, and the standard spectral features are set to the feature threshold range of 400-1700nm in the near-infrared band. The multi-dimensional fused anti-counterfeiting features are time-series aligned with the phase change full-cycle anti-counterfeiting dataset. Using the phase change critical temperature (e.g., 18-28°C) as a benchmark, the temperature nodes of the test data are matched one-to-one with the temperature nodes of the standard feature library to ensure time-series consistency.

[0034] Simultaneously, the multi-dimensional fused anti-counterfeiting features and near-infrared spectral features were compared one by one against the standard appearance time-series features and standard spectral features in the standard feature library. A feature deviation threshold of ±3% was set, and abnormal data with deviations exceeding the threshold were investigated and eliminated. Furthermore, a mapping relationship between each temperature node and the corresponding visual and spectral features was constructed to clarify the correspondence between appearance and spectral features at different temperature nodes. Valid feature data conforming to the phase transition law was further screened, and invalid data and abnormal data such as missing data and misaligned time sequences were eliminated. The screened valid feature information was integrated to obtain a valid anti-counterfeiting feature set that meets the requirements.

[0035] In this optional embodiment, the effective anti-counterfeiting feature set is subjected to phase change association chain anchoring and encryption processing to generate encrypted anti-counterfeiting feature certificates. The encrypted anti-counterfeiting feature certificates are then standardized and stored to obtain a standardized encrypted anti-counterfeiting feature library. This includes: extracting appearance time-series features, spectral time-series features, and linkage correspondences at each temperature node based on the effective anti-counterfeiting feature set; anchoring the extracted features based on the phase change time sequence to construct a phase change feature association chain; encrypting the phase change feature association chain using a constraint encryption mechanism and protecting it from tampering based on identity binding, permission constraints, and integrity verification to generate encrypted anti-counterfeiting feature certificates; standardizing the format and unifying the index of the encrypted anti-counterfeiting feature certificates, and storing the standardized encrypted anti-counterfeiting feature certificates in a structured manner to obtain a standardized encrypted anti-counterfeiting feature library.

[0036] Specifically, the appearance time-series features, near-infrared spectral time-series features, and their correlation at each temperature node (e.g., 18℃, 20℃, 22℃, 24℃, 26℃, 28℃) are extracted. A constraint encryption mechanism is used to encrypt the phase transition feature association chain. A dedicated encryption key is set, and combined with identity binding, permission constraints, and integrity verification mechanisms, an immutable encrypted anti-counterfeiting feature certificate is generated.

[0037] In this optional embodiment, an edge computing and cloud platform collaborative architecture is used to determine phase transition features in a standardized encrypted anti-counterfeiting feature library. Based on the determination results, key feature data is extracted and uploaded to the cloud. The cloud is then used for in-depth analysis of the abnormal phase transition patterns of adulterated coconut oil to construct and update a dynamic anti-counterfeiting feature library. This enables machine vision to perform multi-dimensional, full-cycle anti-counterfeiting identification of coconut oil. The process includes: constructing an edge computing and cloud platform collaborative architecture, where the edge is responsible for local data determination and the cloud is responsible for in-depth analysis and feature library updates; and retrieving the standardized encrypted anti-counterfeiting feature library from the edge and determining phase transition features to verify the features. The system verifies the integrity, temporal consistency, and validity of encrypted anti-counterfeiting feature certificates within the database. Based on local judgment results at the edge, it filters and extracts key feature data that meets the judgment criteria, and standardizes and organizes the extracted key feature data. Utilizing the collaborative transmission channel between edge computing and the cloud platform, it securely uploads the standardized key feature data to the cloud. It conducts in-depth analysis of the key feature data uploaded to the cloud, analyzing the abnormal linkage patterns between visual and spectral features of adulterated coconut oil throughout the entire solid-liquid phase transition cycle. Based on the abnormal linkage patterns of phase transition, it constructs a dynamic anti-counterfeiting feature database. Based on the key feature data, iteratively updates the dynamic anti-counterfeiting feature database to obtain the updated dynamic anti-counterfeiting feature database.

[0038] In this optional embodiment, the standardized key feature data is securely uploaded to the cloud using a collaborative transmission channel between edge computing and the cloud platform. This includes: performing a pre-transmission integrity check on the standardized key feature data and activating the secure transmission mode of the collaborative transmission channel between edge computing and the cloud platform; using an encrypted transmission mechanism to perform transmission layer encryption on the key feature data, uploading the encrypted key feature data to the cloud, and monitoring the transmission status in real time to investigate transmission interruptions and data loss anomalies; receiving the key feature data through the cloud, using a data verification mechanism to perform integrity verification on the received data, and temporarily archiving the key feature data in the cloud based on the verification results.

[0039] Specifically, the encrypted transmission mechanism is processed collaboratively by the edge encryption terminal and the cloud decryption and verification terminal. The key feature data (such as oil appearance and spectral information) collected at the edge are preprocessed and encrypted using the AES-128 encryption mechanism. At the same time, an encryption key (such as a key length of 128 bits) is set, and a secure transmission mode is enabled by using the collaborative transmission channel between the edge terminal and the cloud. During the transmission process, the data transmission status is monitored in real time to prevent data loss or tampering.

[0040] like Figure 2 As shown, according to another embodiment of the present invention, a full-process tracking and management system for the cultivation of Epimedium sagittatum is also provided, comprising: Visual spectral acquisition module 1 is used to acquire the visual features of the oil surface and packaging texture and the near-infrared spectral composition features of the entire phase change process from the critical temperature range of the solid-liquid phase change of coconut oil using high-resolution multispectral integrated detection technology. The acquired features are then fused to obtain the anti-counterfeiting dataset of the entire phase change cycle of the sample to be tested. Visual feature extraction module 2 is used to build a multi-branch fusion recognition model based on convolutional neural network and self-attention network, extract phase transition time sequence features and global correlation analysis of the phase transition full-cycle anti-counterfeiting dataset of the sample to be tested, and obtain multi-dimensional fusion anti-counterfeiting features; The anti-counterfeiting feature verification module 3 is used to perform homogeneous time-series binding verification of multi-dimensional fused anti-counterfeiting features and phase change full-cycle anti-counterfeiting dataset based on the pre-configured coconut oil phase change standard feature library, construct the mapping relationship between the visual features of temperature nodes and spectral component features, and remove invalid data to obtain an effective anti-counterfeiting feature set. The anti-counterfeiting certificate encryption module 4 is used to perform phase change association chain anchoring and encryption processing on the effective anti-counterfeiting feature set, generate encrypted anti-counterfeiting feature certificates, and standardize the storage of encrypted anti-counterfeiting feature certificates to obtain a standardized encrypted anti-counterfeiting feature library. The oil sample anti-counterfeiting identification module 5 is used to use the collaborative architecture of edge computing and cloud platform to determine the phase change characteristics of the standardized encrypted anti-counterfeiting feature library, extract key feature data based on the determination results and upload them to the cloud, and use the cloud to deeply analyze the abnormal phase change linkage of adulterated coconut oil, build and update the dynamic anti-counterfeiting feature library, so as to realize machine vision to perform multi-dimensional and full-cycle anti-counterfeiting identification of coconut oil.

[0041] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A machine vision-based method for identifying counterfeit coconut oil, characterized in that, include: Using high-resolution multispectral integrated detection technology, visual features of the oil's appearance and packaging texture, as well as near-infrared spectral composition features, are obtained from the continuous temporal sequence of the phase transition critical temperature range of coconut oil. These features are then fused to obtain a full-cycle anti-counterfeiting dataset for the sample. A multi-branch fusion recognition model is constructed based on convolutional neural networks and self-attention networks to extract phase transition temporal features and perform global correlation analysis on the full-cycle anti-counterfeiting dataset, resulting in multi-dimensional fused anti-counterfeiting features. Based on a pre-configured standard feature library of coconut oil phase transition features, the multi-dimensional fused anti-counterfeiting features and the full-cycle anti-counterfeiting dataset are subjected to temporal binding verification to ensure they are from the same source. The process involves constructing a mapping relationship between the visual features of temperature nodes and the spectral composition features, and removing invalid data to obtain an effective anti-counterfeiting feature set. This effective anti-counterfeiting feature set is then anchored and encrypted using phase transition association chains to generate encrypted anti-counterfeiting feature credentials. These credentials are then stored in a standardized manner to obtain a standardized encrypted anti-counterfeiting feature library. Utilizing a collaborative architecture of edge computing and a cloud platform, the standardized encrypted anti-counterfeiting feature library is used to determine phase transition features. Based on the determination results, key feature data is extracted and uploaded to the cloud. The cloud platform is then used to perform in-depth analysis of the abnormal phase transition linkage patterns of adulterated coconut oil, constructing and updating a dynamic anti-counterfeiting feature library to achieve multi-dimensional, full-cycle anti-counterfeiting identification of coconut oil using machine vision.

2. The machine vision-based anti-counterfeiting identification method for coconut oil according to claim 1, characterized in that, The method utilizes high-resolution multispectral integrated detection technology to acquire continuous temporal visual features of the oil's appearance and packaging texture, as well as near-infrared spectral composition features, throughout the entire phase transition process of coconut oil within the critical temperature range of the solid-liquid phase transition. These features are then fused to obtain a full-cycle anti-counterfeiting dataset for the sample under test. This includes: coaxially calibrating the high-resolution multispectral integrated detection technology based on the solid-liquid phase transition characteristics of coconut oil and a preset critical temperature range; placing the coconut oil sample under test within the calibrated temperature-controlled area, and using a gradient temperature control terminal to control the temperature rise and fall of the sample according to a preset temperature gradient step size to obtain feature data for each stage of the phase transition; simultaneously acquiring visual features of the oil's appearance and packaging texture, as well as near-infrared spectral composition features, at each preset temperature node; filtering out interference noise and performing image alignment and spectral baseline correction on the visual and spectral features; and performing homogeneous registration and fusion of multi-source features at the same temperature node to form a temporally continuous multi-dimensional fused data set, thus obtaining the full-cycle anti-counterfeiting dataset for the sample under test.

3. The machine vision-based anti-counterfeiting identification method for coconut oil according to claim 1, characterized in that, The multi-branch fusion recognition model, constructed based on convolutional neural networks and self-attention networks, extracts phase transition temporal features and performs global correlation analysis on the phase transition full-cycle anti-counterfeiting dataset of the test sample to obtain multi-dimensional fused anti-counterfeiting features. This includes: constructing a multi-branch fusion recognition model, wherein the model includes convolutional neural network branches and self-attention network branches; inputting the phase transition full-cycle anti-counterfeiting dataset of the test sample into the input layer of the multi-branch fusion recognition model and performing temporal dimension standardization on the dataset; and using the convolutional neural network branches to perform appearance temporal analysis on the standardized phase transition full-cycle anti-counterfeiting dataset. Feature extraction is performed to obtain the temporal pattern of changes in the visual features of oil appearance and packaging texture with temperature nodes throughout the phase transition process, resulting in an appearance temporal feature vector. Based on the self-attention mechanism in the self-attention network branch, spectral temporal features are extracted from the standardized phase transition full-cycle anti-counterfeiting dataset to obtain the linkage pattern of spectral features with temperature nodes, resulting in a spectral temporal feature vector. The appearance temporal feature vector and the spectral temporal feature vector are input to the feature fusion layer at the output of the multi-branch fusion recognition model. The two-branch features are fused using a preset fusion strategy to obtain multi-dimensional fused anti-counterfeiting features.

4. The machine vision-based anti-counterfeiting identification method for coconut oil according to claim 3, characterized in that, The preset fusion strategy includes: feature weight allocation and vector concatenation fusion; the feature weight allocation is used to assign preset weights to the appearance time-series feature vector and the spectral time-series feature vector respectively; the vector concatenation fusion is used to concatenate the two types of feature vectors after weighting to form a comprehensive recognition vector.

5. The machine vision-based anti-counterfeiting identification method for coconut oil according to claim 1, characterized in that, The pre-configured coconut oil phase change standard feature library includes: standard appearance time-series features, standard spectral features, and linkage matching rules; wherein, the standard appearance time-series features are used as the benchmark features for homogeneous binding verification; the standard spectral features are used to ensure the consistency of spectral data during the phase change process; and the linkage matching rules are used to clarify the correspondence between appearance and spectral features at each temperature node.

6. The machine vision-based anti-counterfeiting identification method for coconut oil according to claim 5, characterized in that, The process involves: first, performing time-series binding verification between multi-dimensional fused anti-counterfeiting features and the full-cycle anti-counterfeiting dataset of phase transition based on a pre-configured coconut oil phase transition standard feature library. This includes: first, performing time-series alignment between multi-dimensional fused anti-counterfeiting features and the full-cycle anti-counterfeiting dataset of phase transition based on the pre-configured coconut oil phase transition standard feature library; second, comparing multi-dimensional fused anti-counterfeiting features and near-infrared spectral features based on the coconut oil phase transition standard feature library to identify abnormal data that does not conform to the phase transition law; third, constructing a mapping relationship between temperature nodes and visual and spectral features, filtering effective feature data, and eliminating invalid and abnormal data to obtain feature information that conforms to the phase transition law; and finally, integrating the filtered effective feature information to obtain the effective anti-counterfeiting feature set.

7. The machine vision-based anti-counterfeiting identification method for coconut oil according to claim 1, characterized in that, The process of anchoring and encrypting the effective anti-counterfeiting feature set using phase change association chains to generate encrypted anti-counterfeiting feature credentials, and then standardizing and storing these credentials to obtain a standardized encrypted anti-counterfeiting feature library, includes: extracting appearance time-series features, spectral time-series features, and linkage correspondences at each temperature node based on the effective anti-counterfeiting feature set; anchoring the extracted features using association chains based on the phase change time sequence to construct phase change feature association chains; encrypting the phase change feature association chains using a constraint encryption mechanism, and protecting them from tampering based on identity binding, permission constraints, and integrity verification to generate encrypted anti-counterfeiting feature credentials; standardizing the format and indexing of the encrypted anti-counterfeiting feature credentials, and then storing the standardized encrypted anti-counterfeiting feature credentials in a structured manner to obtain a standardized encrypted anti-counterfeiting feature library.

8. The machine vision-based anti-counterfeiting identification method for coconut oil according to claim 1, characterized in that, The method utilizes a collaborative architecture of edge computing and cloud platform to determine phase transition characteristics of a standardized encrypted anti-counterfeiting feature library. Based on the determination results, key feature data is extracted and uploaded to the cloud. The cloud then uses in-depth analysis of the abnormal phase transition patterns of adulterated coconut oil to construct and update a dynamic anti-counterfeiting feature library. This enables machine vision to perform multi-dimensional, full-cycle anti-counterfeiting identification of coconut oil. The method includes: constructing a collaborative architecture of edge computing and cloud platform, where the edge is responsible for local data determination, and the cloud is responsible for in-depth analysis and feature library updates; and, based on the edge, retrieving the standardized encrypted anti-counterfeiting feature library and determining phase transition characteristics to verify the encrypted features within the feature library. The integrity, temporal consistency, and validity of anti-counterfeiting feature certificates are assessed. Based on the local judgment results at the edge, key feature data that meets the judgment criteria is screened and extracted, and the extracted key feature data is standardized and regularized. Using the collaborative transmission channel between edge computing and the cloud platform, the regularized key feature data is securely uploaded to the cloud. In-depth analysis is performed on the key feature data uploaded to the cloud to analyze the abnormal linkage between visual and spectral features of adulterated coconut oil throughout the entire solid-liquid phase transition cycle. Based on the abnormal linkage pattern of phase transition, a dynamic anti-counterfeiting feature library is constructed. Based on the key feature data, the dynamic anti-counterfeiting feature library is dynamically iterated and updated to obtain the updated dynamic anti-counterfeiting feature library.

9. The machine vision-based anti-counterfeiting identification method for coconut oil according to claim 8, characterized in that, The method of securely uploading standardized key feature data to the cloud using a collaborative transmission channel between edge computing and the cloud platform includes: performing integrity checks on the standardized key feature data before transmission, activating the secure transmission mode of the channel using the collaborative transmission channel between edge computing and the cloud platform; using an encrypted transmission mechanism to encrypt the key feature data at the transmission layer, uploading the encrypted key feature data to the cloud, and monitoring the transmission status in real time to investigate transmission interruptions and data loss anomalies; receiving the key feature data through the cloud, using a data verification mechanism to perform integrity verification on the received data, and temporarily archiving the key feature data in the cloud based on the verification results.

10. A machine vision-based anti-counterfeiting identification system for coconut oil, used to implement the machine vision-based anti-counterfeiting identification method for coconut oil as described in any one of claims 1-9, characterized in that: The visual spectral acquisition module utilizes high-resolution multispectral integrated detection technology to acquire continuous temporal visual features of the oil's appearance and packaging texture, as well as near-infrared spectral composition features, throughout the entire phase transition process of coconut oil from the critical temperature range of the solid-liquid phase transition. It then fuses these features to obtain a full-cycle anti-counterfeiting dataset of the sample under test. The visual feature extraction module, based on a multi-branch fusion recognition model constructed using convolutional neural networks and self-attention networks, extracts phase transition temporal features and performs global correlation analysis on the full-cycle anti-counterfeiting dataset of the sample under test to obtain multi-dimensional fused anti-counterfeiting features. The anti-counterfeiting feature verification module is used to perform homogeneous time-series binding verification on multi-dimensional fused anti-counterfeiting features and phase change full-cycle anti-counterfeiting datasets based on a pre-configured coconut oil phase change standard feature library. It constructs the mapping relationship between the visual features of temperature nodes and spectral component features and removes invalid data to obtain a valid anti-counterfeiting feature set. The anti-counterfeiting certificate encryption module is used to anchor and encrypt the phase change association chain of the valid anti-counterfeiting feature set, generate encrypted anti-counterfeiting feature certificates, and standardize the storage of the encrypted anti-counterfeiting feature certificates to obtain a standardized encrypted anti-counterfeiting feature library. The oil sample anti-counterfeiting identification module is used to determine the phase change features of the standardized encrypted anti-counterfeiting feature library using a collaborative architecture of edge computing and cloud platform. Based on the determination results, it extracts key feature data and uploads it to the cloud. It uses the cloud to deeply analyze the abnormal phase change linkage of adulterated coconut oil, constructs and updates the dynamic anti-counterfeiting feature library, so as to realize machine vision for multi-dimensional full-cycle anti-counterfeiting identification of coconut oil.

Images