Local auto-scaling classification of spectral data sets

Abstract

The invention relates to local auto-scaling classification of spectral data sets. In some implementations, a device may receive a spectral data set associated with an unknown sample. The apparatus may obtain a multi-category classification model used to classify an unknown sample into at least one category of a plurality of categories; wherein the multi-category classification model includes a plurality of local auto-scaling one-to-one (OVO) binary classifiers, each of the plurality of local auto-scaling OVO binary classifiers being associated with a different category pair of a plurality of categories. The apparatus may apply local auto-scaling to a spectral data set associated with an unknown sample to create a local auto-scaled spectral data set. The apparatus may perform classification of unknown samples based on the locally auto-scaled spectral data set and using a multi-class classification model that includes a plurality of locally auto-scaled OVO binary classifiers.

Description

Background Art

[0001] Classification is a predictive modeling technique that can be used to assign samples to categories. A binary classification model can be configured to assign samples to one of two categories, while a multi-class classification model can be configured to assign samples to one of multiple categories. The type of classification model used depends on the application. Some algorithms, such as support vector machine (SVM) algorithms, are designed for binary classification. Algorithms designed for binary classification cannot be directly used for multi-class classification models. However, heuristic techniques can be used to split the multi-class classification problem into multiple binary classification data sets and train a binary classification model on each binary classification data set. Examples of these heuristic techniques include one-to-many (OVR) techniques and one-to-one (OVO) techniques. Summary of the invention

[0002] Some implementations described herein relate to a method. The method may include receiving, by a device, a spectral data set associated with an unknown sample. The method may include obtaining, by the device, a multi-class classification model, the multi-class classification model being used to classify the unknown sample into at least one of a plurality of classes, wherein the multi-class classification model includes a plurality of locally automatically scaled one-to-one (OVO) binary classifiers, each of the plurality of locally automatically scaled OVO binary classifiers being associated with a different class pair from a plurality of classes. The method may include applying, by the device, local automatic scaling to the spectral data set associated with the unknown sample to create a locally automatically scaled spectral data set. The method may include performing, by the device, classification of the unknown sample based on the locally automatically scaled spectral data set and using a multi-class classification model including a plurality of locally automatically scaled OVO binary classifiers.

[0003] Some implementations described herein relate to a device. The device may include one or more memories and one or more processors coupled to the one or more memories. One or more processors may be configured to receive a spectral data set associated with an unknown sample. One or more processors may be configured to obtain a multi-class classification model, which is used to classify the unknown sample into at least one of a plurality of categories, wherein the multi-class classification model includes a plurality of local automatic scaling OVO binary classifiers, each of the plurality of local automatic scaling OVO binary classifiers being associated with different class pairs from a plurality of categories. One or more processors may be configured to apply local automatic scaling to a spectral data set associated with an unknown sample to create a locally automatically scaled spectral data set. One or more processors may be configured to perform classification of an unknown sample based on a locally automatically scaled spectral data set and using a multi-class classification model including a plurality of locally automatically scaled OVO binary classifiers.

[0004] Some implementations described herein relate to a method for generating a multi-class classification model. The method may include obtaining, by a device, a training data set of spectral data. The method may include identifying, by the device and based on the training data set, multiple categories of the multi-class classification model. The method may include generating, by the device, multiple OVO binary classifiers of the multi-class classification model, wherein each OVO binary classifier in the multiple OVO binary classifiers is associated with a different pair of categories from multiple categories. The method may include applying, by the device, local automatic scaling to each OVO binary classifier in the multiple OVO binary classifiers to create multiple local automatic scaling OVO binary classifiers of the multi-class classification model. The method may include storing, by the device, a multi-class classification model comprising multiple local automatic scaling OVO binary classifiers. BRIEF DESCRIPTION OF THE DRAWINGS

[0005] Figure 1A-1I is a diagram illustrating an example associated with local auto-scaling classification of a spectral dataset.

[0006] Figure 2 is a diagram of an example environment in which the systems and / or methods described herein may be implemented.

[0007] Figure 3 is a diagram of example components of an apparatus associated with local automatic scaling classification of a spectral dataset.

[0008] Figure 4 is a flow chart of an example process associated with classifying an unknown sample based on a locally autoscaled spectral dataset and using a multi-class classification model including multiple locally autoscaled one-vs-one (OVO) binary classifiers.

[0009] Figure 5 is a flow chart of an example process associated with generating a multi-class classification model including multiple local auto-scaled OVO binary classifiers. DETAILED DESCRIPTION

[0010] The following detailed description of example implementations refers to the accompanying drawings. The same reference numerals in different drawings may identify the same or similar elements. A spectrometer is used as an example below. However, the techniques, principles, processes, and methods described herein can be used with any sensor, including but not limited to other optical sensors and spectral sensors.

[0011] A multispectral sensor device may be capable of measuring, acquiring, collecting, or otherwise determining a spectral data set associated with a sample (e.g., a material, a compound, a chemical, a fluid, etc.). The spectral data set may include time-series spectral data associated with multiple (e.g., 16, 32, 64, etc.) wavelength channels.

[0012] The spectral data set can be provided to the classification model for the classification of the sample. For example, the multi-class classification model can be configured to assign a given sample to one of n (n>1) categories. In general, the spectral data set is provided as input to the multi-class classification model, and an indication of a specific category in the n categories to which the sample is assigned is provided as output. However, due to the heterogeneity and complexity of the spectral data set structure (for example, particularly when samples of multiple categories tend to provide similar spectral data sets), the performance of the multi-class classification model generated based on such spectral data sets and classified using such spectral data sets is affected. As a result, the reliability of the multi-class classification model using the spectral data set is reduced, and the possibility of misclassification of a given sample is increased.

[0013] Some implementations described herein enable the use of local automatic scaling techniques to classify spectral data sets. In some implementations, a classification device may receive a spectral data set associated with an unknown sample. The classification device applies local automatic scaling to the spectral data set associated with the unknown sample to create a locally automatically scaled spectral data set. The classification device may obtain a multi-class classification model that is used to classify the unknown sample into at least one of a plurality of classes. Here, the multi-class classification model includes a plurality of locally automatically scaled one-to-one (OVO) binary classifiers, wherein each locally automatically scaled OVO binary classifier is associated with a different class pair. The classification device then performs classification of the unknown sample based on the locally automatically scaled spectral data set and using a multi-class classification model including a plurality of locally automatically scaled OVO binary classifiers. In some implementations, the techniques and devices described herein improve the reliability of the multi-class classification model using the spectral data set, and similarly reduce the possibility of misclassification of a given sample. More details are provided below.

[0014] Figure 1A-1I is a diagram illustrating an example associated with local auto-scaling classification of a spectral dataset. Figure 1A is a diagram illustrating an example 100 associated with generating a multi-class classification model including multiple local auto-scaling OVO binary classifiers.

[0015] like Figure 1AAs shown at reference numeral 102, the classification device 210 can obtain a training data set of spectral data. The training data set may include spectral data, based on which a multi-class classification model can be generated, the multi-class classification model including multiple OVO binary classifiers that can be used to classify unknown samples. For example, the training data set may include a spectrum of each sample in a sample group (e.g., multivariate time series data, such as near infrared (NIR) spectra) (e.g., measured by a multi-spectral sensor device). Here, the category of each sample in the sample group is known (e.g., an indication of the category of each sample can be indicated in the training data set). In some implementations, the training data set includes spectra for multiple samples of the same category (e.g., to improve the robustness of the multi-class classification model). In addition, the training data set may include spectra for one or more samples in each category of a plurality of n (n>1) categories.

[0016] As shown at reference numeral 104, in a first operation associated with generating a multi-class classification model including a plurality of OVO binary classifiers, the classification device 210 may identify a plurality of n classes of the multi-class classification model. In some implementations, the classification device 210 may identify the plurality of n classes based on a training data set. For example, the classification device 210 may obtain a training data set including spectra for one or more samples associated with each class, wherein the class of each sample is indicated in the training data set. Thus, the classification device 210 may identify a plurality of n classes based on the training data set (e.g., based on an indication of the class of each sample in the training data set).

[0017] Next, as indicated at reference numeral 106 , the classification device 210 may generate a plurality of OVO binary classifiers of the multi-class classification model, wherein each of the plurality of OVO binary classifiers is associated with a different class pair from the plurality of classes.

[0018] OVO technology is a technology that enables binary classification algorithms to be used for multi-class classification. According to OVO technology, a multi-class classification data set is split into a binary classification problem. That is, the data set is split into a data set for each category and each other category. For example, a multi-class classification problem with n categories is split into m (m = [(n×(n-1)) / 2]) binary classification data sets. As an example, a multi-class classification problem may include four categories (n=4)-Cls1, Cls2, Cls3, and Cls4. Here, the multi-class classification problem is split into six (for example, m = [(4×(4-1)) / 2] = 6) binary classification problems-Cls1 and Cls2, Cls1 and Cls3, Cls1 and Cls4, Cls2 and Cls3, Cls2 and Cls4, and Cls3 and Cls4. In this example, six data sets are used-there is one data set for each category pair. Then create a binary classification model from each binary classification data set, wherein a given binary classification model is designed to assign samples to one of the two categories associated with the binary classification data set, and a given binary classification model is generated based on the binary classification data set. This binary classification model is referred to as an OVO binary classifier in this article. When performing classification, samples are assigned to the category that obtains the most votes from multiple OVO binary classifiers (i.e., the category with the highest number of votes in the output of multiple OVO binary classifiers).

[0019] In some implementations, the classification device 210 generates multiple OVO binary classifiers based on a training data set. For example, the classification device 210 can create m binary classification data sets based on a training data set associated with n categories. Next, the classification device 210 can generate an OVO binary classifier based on each of the m binary classification data sets. That is, the classification device 210 can generate m binary classification models, each of which can assign samples to one of the two categories associated with the binary classification data set based on which the given binary classification model is generated. In some implementations, each OVO binary classifier in a plurality of OVO binary classifiers is a binary classification model. For example, each OVO binary classifier in a plurality of OVO binary classifiers can be a support vector machine (SVM) model. Therefore, in an implementation, the multi-class classification model can be a model based on an SVM.

[0020] In some implementations, each column of data associated with a given OVO binary classifier corresponds to a different wavelength of light. That is, in some applications, each column of data in a given one of the m binary classification datasets may correspond to a different wavelength of light (e.g., a channel of the multispectral sensor device 205), meaning that a given column may include spectral data associated with a single wavelength and multiple scans. In addition, in some implementations, each row of data associated with a given OVO binary classifier corresponds to a spectral scan spanning a range of wavelengths of light. That is, in some applications, each row of data in a given one of the m binary classification datasets may correspond to a particular scan (e.g., as collected by the multispectral sensor device 205), meaning that a given row may include spectral data associated with multiple wavelengths for a single scan.

[0021] Next, as indicated at reference numeral 108 , the classification device 210 may apply local auto-scaling to each of the plurality of OVO binary classifiers to create a plurality of local auto-scaling OVO binary classifiers of the multi-class classification model.

[0022] Automatic scaling is a data preprocessing technique that can be used, for example, to improve model performance (e.g., improve the accuracy of predictions), reduce the impact of data outliers, or ensure that the data is at the same scale. Traditionally, automatic scaling is applied to the global level. For example, according to traditional practice, before generating multiple binary classification models, automatic scaling (if applied) is applied to the training data (e.g., so that automatic scaling will be applied to the entire training data set). As used herein, local automatic scaling refers to automatic scaling performed at a local level-this means that automatic scaling is performed separately for each OVO binary classifier. That is, local automatic scaling can include automatic scaling of a binary classification data set associated with a specific OVO binary classifier (e.g., instead of automatically scaling the entire training data set). In some implementations, by applying local automatic scaling, the perspective of the high-dimensional spectral pattern is centered, and the spectral pattern is scaled to equal variance (e.g., so that all channels have a standard deviation of 1), which significantly improves the resolution of all channels. In some implementations, local automatic scaling is a column-by-column operation.

[0023] In some implementations, in order to apply local automatic scaling to a given OVO binary classifier, the classification device 210 first performs mean centering on each column of data associated with the given OVO binary classifier. For example, the classification device 210 may calculate the mean of the data in each column of the binary classification data set associated with the given OVO binary classifier. The classification device 210 may then subtract the corresponding mean from the data in each column (e.g., so that the mean of each column is zero). Next, the classification device 210 may scale each column of data associated with the given OVO binary classifier to unit variance (e.g., so that the variance in all directions is scaled to 1). For example, the classification device 210 may calculate the standard deviation (STD) of each column of the binary classification data set associated with the OVO binary classifier centered on the mean, and then the data from the given column may be divided by the corresponding STD of the given column. In some implementations, after applying local automatic scaling to the binary classification data set, the mean of each column is zero and the STD is equal to 1. In some implementations, the classification device 210 may apply local automatic scaling to each OVO binary classifier. That is, the classification device 210 can apply local autoscaling to each binary classification dataset associated with multiple OVO binary classifiers separately (for example, so that the classification device 210 applies local autoscaling m times, once for each of the m binary classification datasets).

[0024] In this way, the classification device 210 can generate a multi-class classification model including multiple local auto-scaling OVO binary classifiers. In some implementations, the multi-class classification model can be used to perform classification of unknown samples, as described below. Figure 1B An example thereof is described.

[0025] As shown at reference numeral 110, the classification device 210 can store a multi-class classification model including a plurality of local auto-scaling OVO binary classifiers. For example, the classification device 210 can store the multi-class classification model in a memory of the classification device 210 (e.g., so that the classification device 210 can access the multi-class classification model later). Additionally or alternatively, the classification device 210 can provide the multi-class classification model for storage in another device (e.g., the multispectral sensor device 205, the user device 215, etc.).

[0026] Figure 1B is a diagram illustrating an example 150 associated with classifying an unknown sample based on a locally auto-scaled spectral dataset and using a multi-class classification model including a plurality of locally auto-scaled OVO binary classifiers.

[0027] like Figure 1BAs shown at reference numeral 152 in the figure, the classification device 210 can receive a spectral dataset associated with an unknown sample. The spectral dataset can include spectral data corresponding to the sample to be classified using a multi-class classification model including multiple local auto-scaling OVO binary classifiers. For example, the spectral dataset can include a spectrum (e.g., multivariate time series data, such as NIR spectrum) of the sample (e.g., as measured by the multi-spectral sensor device 205). Here, the class of the sample is unknown. In some aspects, such as Figure 1B As indicated in , the classification device 210 may receive a spectral dataset from the multi-spectral sensor device 205 .

[0028] In some implementations, each column of data in the spectral data set corresponds to a different wavelength of light. That is, in some applications, each column of data in the spectral data set can correspond to a different wavelength of light (e.g., a channel of the multi-spectral sensor device 205), which means that a given column can include spectral data associated with a single wavelength. In addition, in some implementations, each row of data in the spectral data set can correspond to a scan across a range of wavelengths of light. That is, in some applications, each row of data in the spectral data set can correspond to a specific scan (e.g., collected by the multi-spectral sensor device 205), which means that a given row can include spectral data associated with multiple wavelengths for a single scan.

[0029] As shown at reference numeral 154, the classification device 210 can obtain a multi-class classification model, which is used to classify an unknown sample into at least one of a plurality of classes, wherein the multi-class classification model includes a plurality of local automatic scaling OVO binary classifiers, each of the plurality of local automatic scaling OVO binary classifiers being associated with a different class pair from a plurality of classes. For example, the classification device 210 can generate, store, or otherwise access a multi-class classification model including a plurality of local automatic scaling OVO binary classifiers (e.g., as described above with respect to Figure 1A The generated multi-class classification model).

[0030] As shown at reference numeral 156, the classification device 210 can apply local automatic scaling to the spectral data set associated with the unknown sample to create a locally automatically scaled spectral data set. In some implementations, the classification device 210 applies local automatic scaling to the spectral data set for each OVO binary classifier. Therefore, in some implementations, the classification device 210 applies local automatic scaling to the spectral data set m times-once for each classifier in the m locally automatically scaled OVO binary classifiers. In some implementations, in order to apply local automatic scaling to the spectral data set with respect to a given locally automatically scaled OVO binary classifier, the classification device 210 first performs mean centering on each column of data in the spectral data set. For example, the classification device 210 can calculate (or determine from a memory) the average value of the data in each column of the binary classification data set associated with a given locally automatically scaled OVO binary classifier. The classification device 210 can then subtract the corresponding average value from the data in each column of the spectral data set. Next, the classification device 210 can scale each column of data in the spectral data set. For example, the classification device 210 may calculate (or determine from memory) the STD of each column of the binary classification data set centered on the mean value associated with a given local auto-scaling OVO binary classifier, and then the data from the given column of the spectral data set may be divided by the corresponding STD. In some implementations, the classification device 210 applies local auto-scaling with respect to each local auto-scaling OVO binary classifier, which means that for each OVO binary classifier, the spectral data set associated with the unknown sample is locally auto-scaled once. Then, the iteration of the locally auto-scaled spectral data set associated with the given local auto-scaling OVO binary classifier can be used by the locally auto-scaling OVO binary classifier in association with classifying the unknown sample. In this way, the classification device 210 can create a locally auto-scaled spectral data set, which includes m iterations of the locally auto-scaled spectral data set, wherein each of the m iterations of the locally auto-scaled spectral data set is associated with a different one of the multiple m locally auto-scaling OVO binary classifiers.

[0031] As shown at reference numeral 158, the classification device 210 can perform classification of unknown samples based on the locally automatically scaled spectral data set and using a multi-class classification model including multiple locally automatically scaled OVO binary classifiers. In some implementations, when performing classification of unknown samples, the classification device 210 can determine the votes provided by each of the multiple locally automatically scaled OVO binary classifiers based on the locally automatically scaled spectral data set. Then, the classification device 210 can identify the category with the highest number of votes in multiple categories, and the unknown sample can be classified as being included in the category with the highest number of votes in multiple categories. For example, the multi-class classification model can include m locally automatically scaled OVO binary classifiers. Here, the classification device 210 can provide the locally automatically scaled spectral data set to each of the m locally automatically scaled OVO binary classifiers. Then, a given locally automatically scaled OVO binary classifier provides a vote indicating one of the two categories associated with the given locally automatically scaled OVO binary classifier. The classifier 210 receives votes from all local auto-scaling OVO binary classifiers and then identifies the class that receives the most votes among the n classes. The classifier 210 then assigns the unknown sample to the class that receives the most votes.

[0032] As shown at reference numeral 160, the classification device 210 can provide information associated with the classification results of the unknown sample. Figure 1B As indicated in , the classification device 210 can provide information associated with the results of the classification to the user device 215. In some implementations, the information associated with the results of the classification may include an indication of the category to which the unknown sample is assigned. As an example, each category can be a different type of material, a different type of compound, a different type of chemical, or a different type of fluid, etc. Additionally or alternatively, the information associated with the classification results may include other information, such as an indication of the number of votes received by the category to which the unknown sample is assigned, an indication of the number of votes received by one or more other categories (e.g., the number of votes received by the category with the second most votes), or other types of information.

[0033] In some implementations, local automatic scaling of multiple OVO binary classifiers and local automatic scaling of spectral data sets associated with unknown samples improves the performance of the multi-class classification model (e.g., compared to a multi-class classification model using OVO binary classifiers without local automatic scaling). For example, the techniques and devices described herein improve the discrimination ability of the multi-class classification model. That is, the techniques and devices described herein increase the likelihood that the multi-class classification model will assign unrelated samples to different categories. Similarly, the techniques and devices described herein improve the versatility of the multi-class classification model. That is, the techniques and devices described herein improve the ability of the multi-class classification model to adapt to new or unseen spectral data sets, which in turn increases the classification success rate of the multi-class classification model.

[0034] In addition, local automatic scaling of multiple OVO binary classifiers improves the reliability and robustness of multi-class classification models that perform classification based on spectral data sets, and similarly reduces the possibility of misclassification of a given sample. As a result, resource consumption associated with performing classification (e.g., consumption of processing resources, energy consumption, consumption of network resources) can be reduced. For example, for traditional (less reliable) multi-class classification models, it may be necessary (e.g., by the same classification device or by multiple classification devices) to perform multiple iterations of the same classification, and then use some other data processing techniques to average the results of the classification to ensure that accurate classification results are obtained. However, using a multi-class classification model that includes multiple local automatic scaling OVO binary classifiers improves the classification accuracy of the multi-class classification model. This means that no additional classification iterations need to be performed (by the classification device 210 or any other device). As a result, the consumption of processing resources, energy consumption, and network resources associated with performing classification are reduced.

[0035] As indicated above, Figure 1A and Figure 1B are provided as examples. Other examples may be related to Figure 1A and Figure 1B Different than described. Figure 1A and Figure 1B The number and arrangement of the devices in are provided as an example. Figure 1A and Figure 1B There may be additional devices, fewer devices, different devices, or differently arranged devices than those shown in FIG. Figure 1A and Figure 1B Two or more of the devices shown in may be implemented in a single device, or Figure 1A and Figure 1B The single device shown in can be implemented as multiple distributed devices. Additionally or alternatively, Figure 1A and Figure 1BA set of devices (eg, one or more devices) shown in FIG. 1 may perform the operations described as being performed by Figure 1A and Figure 1B One or more functions performed by another set of devices shown in .

[0036] Figure 1C is a diagram illustrating an example algorithm associated with performing classification of unknown samples using a multi-class classification model including multiple local auto-scaling OVO binary classifiers.

[0037] like Figure 1C As shown in , the multi-class classification model includes a plurality of n classes (e.g., Cls1 to Clsn) and generates a plurality of m OVO binary classifiers (e.g., OVO 1_2 to OVO n-1_n). As further shown, local auto-scaling is applied to each OVO binary classifier to create a plurality of m local auto-scaling OVO binary classifiers (e.g., OVO 1_2 to OVO n-1_n). Here, when performing classification of an unknown sample, each local auto-scaling OVO binary classifier indicates a vote for one of the two associated classes (e.g., LA OVO 1_2 indicates a vote for Cls1 or Cls2). The class Clsk that receives the most votes is identified, and the unknown sample is classified as being included in the class Cksk.

[0038] As indicated above, Figure 1C is provided as an example. Other examples may be related to Figure 1C Described differently.

[0039] Figure 1D-1F is a graph illustrating the improvement in the performance of a multi-class classification model by using local autoscaling. Figure 1D-1F In the associated example, the multi-class classification model includes 10 classes of polymer materials.

[0040] Figure 1D and Figure 1E is a diagram illustrating a principal component analysis (PCA) plot associated with a multi-class classification model including a traditional OVO binary classifier associated with 10 classes.

[0041] exist Figure 1D In the example, the horizontal axis is the score of principal component 1 (PC1), and the vertical axis is the score of principal component 2 (PC2). Figure 1D As indicated in , a multi-class classification model including a conventional OVO binary classifier provides class boundaries (e.g., computed by SVM) for each of the 10 classes. Figure 1DAs shown in the figure, for a multi-class classification model including a traditional OVO binary classifier, several class boundaries overlap or almost overlap. Each point corresponds to a spectrum projected into the PCA space. It is worth noting that Figure 1D is provided to illustrate overlap or near overlap of category boundaries, and Figure 1D The distinction between the points in is not important.

[0042] Figure 1E It's a picture. Figure 1D , which is near the overlap of the boundary associated with class Cls1 and the boundary associated with class Cls2. Figure 1E As shown in , the class Cls1 boundary and the class Cls2 boundary overlap, which means that the probability of misclassification associated with classes Cls1 and Cls2 increases. The points corresponding to the training data based at least in part on which the multi-class classification model is generated are in Figure 1E In the example, the training data is labeled as "Cls1 training" and "Cls2 training". Figure 1E As shown in , when a spectral dataset including test data for Cls2 samples (e.g., a spectral dataset of samples known to be in category Cls2) is provided to a multi-class classification model including a traditional OVO binary classifier, shown as "Cls2 test", the data is projected into the PCA space within the boundary of category Cls2, which means that the multi-class classification model accurately classifies the Cls2 samples as being included in category Cls2. However, when a spectral dataset including test data for Cls1 samples (e.g., a spectral dataset of samples known to be in category Cls1) is provided to a multi-class classification model including a traditional OVO binary classifier, shown as "Cls1 test", the data is projected into the PCA space within the boundary of category Cls2, which means that the multi-class classification model incorrectly classifies the Cls1 samples as being included in category Cls2.

[0043] Figure 1F is a diagram illustrating a region of a PCA plot near the overlap of a boundary associated with class Cls1 and a boundary associated with class Cls2 in a scenario where a multi-class classification model includes a plurality of local auto-scaling OVO binary classifiers. That is, Figure 1F The figure shows the boundary between class Cls1 and class Cls2 when local autoscaling has been applied to multiple OVO binary classifiers of a multi-class classification model. Note that Figure 1F The viewing angle and resolution of mid-spectrum patterns are improved (e.g., Figure 1E compared to).

[0044] like Figure 1FAs shown in , the use of local auto-scaling OVO binary classifiers increases the interval between points associated with classes Cls1 and Cls2. As a result, when a spectral dataset (e.g., Figure 1E When the multi-class classification model is used in the same spectral dataset as that used in ), the data is projected into the PCA space within the boundary of class Cls1, which means that the multi-class classification model accurately classifies the Cls1 samples as being included in class Cls1. As shown in the figure, the multi-class classification model including multiple local auto-scaling OVO binary classifiers also accurately classifies the Cls2 samples as being included in class Cls2.

[0045] As indicated above, Figure 1D-1F are provided as examples. Other examples may be related to Figure 1D-1F Different than described.

[0046] Figure 1G-1I : is a diagram illustrating an example of the classification success rate of an SVM-based multi-class classification model including a traditional OVO binary classifier and an SVM-based multi-class classification model including a local auto-scaling OVO binary classifier. In this example, the SVM-based multi-class classification model is designed to classify a sample into one of 46 categories. Figure 1G-1I , the points along the horizontal axis correspond to different preprocessing schemes associated with preprocessing spectral data, and the vertical axis corresponds to the number of misclassifications. The line labeled "SVM" corresponds to misclassifications by a SVM-based multi-class classification model including a conventional OVO binary classifier, while the line labeled "SVM with LA" corresponds to misclassifications by a SVM-based multi-class classification model including a local autoscaling OVO binary classifier.

[0047] like Figure 1G It can be seen that, on different preprocessing schemes, the SVM-based multi-class classification model including the local auto-scaling OVO binary classifier generally performs significantly better (e.g., has fewer misclassifications) than the SVM-based multi-class classification model including the traditional OVO binary classifier.

[0048] Figure 1H It's a picture. Figure 1G FIG. 1 is a diagram of a close-up view of the first indicated area of ​​the figure shown in FIG. Figure 1HThe region shown in corresponds to a specific set of preprocessing schemes, each of which includes applying a standard normal variable (SNV) operation to the spectral data and then calculating the first derivative (first) of a specific moving window size Wx (x=5, 7, 9, 11, 13, 15 or 17) with a third-order polynomial (P3). As shown in the figure, for this set of preprocessing schemes, the multi-class classification model including multiple local automatic scaling OVO binary classifiers has zero misclassifications (e.g., the multi-class classification model including multiple local automatic scaling OVO binary classifiers has a 100% success rate), while the multi-class classification model including the traditional OVO binary classifier has a large number of misclassifications (e.g., 48 or more).

[0049] Fig. 1I It's a picture. Figure 1G FIG. 2 is a diagram of a close-up view of the second indicated area of ​​the figure shown in FIG. Fig. 1I The region shown in corresponds to a specific set of preprocessing schemes, each of which includes calculating the first derivative (first) associated with a spectral data set for a specific moving window size Wx (x=5, 7, 9, 11, 13, 15, or 17) with a third-order polynomial (P3), and then applying a standard normal variable (SNV) operation. As shown in the figure, for this set of preprocessing schemes, the multi-class classification model including multiple local automatic scaling OVO binary classifiers has zero misclassifications (e.g., the multi-class classification model including multiple local automatic scaling OVO binary classifiers has a 100% success rate), while the multi-class classification model including a traditional OVO binary classifier has a large number of misclassifications (e.g., 30 or more).

[0050] As indicated above, Figure 1G-1I are provided as examples. Other examples may be related to Figure 1G-1I Different than described.

[0051] Figure 2 2 is a diagram of an example environment 200 in which the systems and / or methods described herein may be implemented. Figure 2 As shown in FIG. 2 , environment 200 may include multispectral sensor device 205, classification device 210, user device 215, and network 220. The devices of environment 200 may be interconnected via wired connections, wireless connections, or a combination of wired and wireless connections.

[0052] The multi-spectral sensor device 205 includes a device capable of measuring, gathering, collecting, or otherwise determining absorption spectral data associated with multiple wavelength channels. For example, the multi-spectral sensor device 205 may include a multi-spectral sensing device capable of determining absorption data on each of the 64 wavelength channels. In some implementations, the multi-spectral sensor device 205 may operate in the visible spectrum, near infrared spectrum, and / or infrared spectrum. In some implementations, the multi-spectral sensor device 205 may be, for example, a handheld device or a wearable device (e.g., a device that can be worn on a wrist, finger, arm, leg, head, or ear). In some implementations, the multi-spectral sensor device 205 may receive information from and / or transmit information to another device in the environment 200 (such as a classification device 210).

[0053] The classification device 210 includes a device capable of performing classification of unknown samples according to a local automatic scaling technique, as described herein. For example, the classification device 210 may include an application specific integrated circuit (ASIC), an integrated circuit, a server, a server group, and / or another type of communication and / or computing device. In some implementations, the classification device 210 may be integrated with the multi-spectral sensor device 205 (e.g., such that the multi-spectral sensor device 205 and the classification device 210 are located on the same chip, in the same package, in the same housing, and / or in the same handheld or wearable device). Alternatively, in some implementations, the classification device 210 may be separated from the multi-spectral sensor device 205. In some implementations, the classification device 210 may receive information from and / or transmit information to another device in the environment 200 (such as the multi-spectral sensor device 205).

[0054] The user device 215 includes one or more devices capable of receiving, generating, storing, processing, and / or providing information associated with the classification of an unknown sample according to a local automatic scaling technique, as described herein. For example, the user device 215 may include a communication and / or computing device, such as a mobile phone (e.g., a smart phone or a wireless phone), a laptop computer, a tablet computer, a handheld computer, a desktop computer, a gaming device, a wearable communication device (e.g., a smart watch or smart glasses), or a similar type of device. In some implementations, the multispectral sensor device 205 and / or the classification device 210 may be integrated with the user device 215 (e.g., such that the multispectral sensor device 205 and / or the classification device 210 is included in a housing of the user device 215).

[0055] The network 220 includes one or more wired and / or wireless networks. For example, the network 220 may include a wired network (e.g., when the multispectral sensor device 205 and the classification device 210 are included in the same package and / or the same chip). As another example, the network 220 may include a cellular network (e.g., a long-term evolution (LTE) network, a code division multiple access (CDMA) network, a 3G network, a 4G network, a 5G network, or another type of next-generation network), a public land mobile network (PLMN), a local area network (LAN), a wide area network (WAN), a metropolitan area network (MAN), a telephone network (e.g., a public switched telephone network (PSTN)), a private network, an ad hoc network, an intranet, the Internet, a fiber-optic-based network, a cloud computing network, etc. and / or a combination of these or other types of networks.

[0056] Figure 2 The number and arrangement of devices and networks shown in FIG. 1 are provided as an example. In practice, Figure 2 There may be additional devices and / or networks, fewer devices and / or networks, different devices and / or networks, or differently arranged devices and / or networks than those shown. Figure 2 Two or more devices shown in FIG. 1 may be implemented in a single device, or in Figure 2 A single device shown in FIG. 200 may be implemented as multiple distributed devices. Additionally or alternatively, a set of devices (eg, one or more devices) of environment 200 may perform one or more functions described as being performed by another set of devices of environment 200.

[0057] Figure 3 is a diagram of example components of a device 300 associated with local automatic scaling classification of a spectral data set. The device 300 may correspond to the multispectral sensor device 205, the classification device 210, and / or the user device 215. In some implementations, the multispectral sensor device 205, the classification device 210, and / or the user device 215 may include one or more devices 300 and / or one or more components of the device 300. Figure 3 As shown in , device 300 may include a bus 310 , a processor 320 , a memory 330 , an input component 340 , an output component 350 , and / or a communication component 360 .

[0058] The bus 310 may include one or more components that enable wired and / or wireless communications between components of the device 300. The bus 310 may include: Figure 3Two or more components of a computer program product may be coupled together, such as via operational coupling, communication coupling, electronic coupling, and / or electrical coupling. For example, bus 310 may include electrical connections (e.g., wires, traces, and / or leads) and / or wireless buses. Processor 320 may include a central processing unit, a graphics processing unit, a microprocessor, a controller, a microcontroller, a digital signal processor, a field programmable gate array, an application specific integrated circuit, and / or another type of processing component. Processor 320 may be implemented in hardware, firmware, or a combination of hardware and software. In some implementations, processor 320 may include one or more processors that can be programmed to perform one or more operations or processes described elsewhere herein.

[0059] The memory 330 may include volatile and / or non-volatile memory. For example, the memory 330 may include random access memory (RAM), read-only memory (ROM), a hard drive, and / or other types of memory (e.g., flash memory, magnetic memory, and / or optical memory). The memory 330 may include internal memory (e.g., RAM, ROM, or hard drive) and / or removable memory (e.g., removable via a universal serial bus connection). The memory 330 may be a non-transitory computer-readable medium. The memory 330 may store information, one or more instructions, and / or software (e.g., one or more software applications) related to the operation of the device 300. In some implementations, the memory 330 may include one or more memories, which are coupled to (e.g., communicatively coupled to) one or more processors (e.g., processor 320), such as via bus 310. The communicative coupling between the processor 320 and the memory 330 may enable the processor 320 to read and / or process information stored in the memory 330 and / or store information in the memory 330.

[0060] Input component 340 can enable device 300 to receive input, such as user input and / or sensed input. For example, input component 340 can include touch screen, keyboard, keypad, mouse, button, microphone, switch, sensor, GPS sensor, GNSS sensor, accelerometer, gyroscope and / or actuator. Output component 350 can enable device 300 to provide output such as via display, speaker and / or light emitting diode. Communication component 360 can enable device 300 to communicate with other devices via wired connection and / or wireless connection. For example, communication component 360 can include receiver, transmitter, transceiver, modem, network interface card and / or antenna.

[0061] Device 300 can perform one or more operations or processes described herein. For example, a non-transient computer-readable medium (e.g., memory 330) can store a set of instructions (e.g., one or more instructions or codes) for execution by processor 320. Processor 320 can execute the set of instructions to perform one or more operations or processes described herein. In some implementations, the execution of the set of instructions by one or more processors 320 causes one or more processors 320 and / or device 300 to perform one or more operations or processes described herein. In some implementations, hard-wired circuits can be used instead of instructions or in combination with instructions to perform one or more operations or processes described herein. Additionally or alternatively, processor 320 can be configured to perform one or more operations or processes described herein. Therefore, the implementation described herein is not limited to any specific combination of hardware circuits and software.

[0062] Figure 3 The number and arrangement of components shown in FIG. 3 are provided as an example. Device 300 may include Figure 3 More components, fewer components, different components, or differently arranged components than those shown in . Additionally or alternatively, one set of components (e.g., one or more components) of device 300 may perform one or more functions described as being performed by another set of components of device 300.

[0063] Figure 4 is a flow chart of an example process 400 associated with classifying an unknown sample based on a locally auto-scaled spectral dataset and using a multi-class classification model including multiple locally auto-scaled OVO binary classifiers. In some implementations, Figure 4 One or more processing blocks of are performed by a classification device (e.g., classification device 210). In some implementations, Figure 4 One or more processing blocks of are performed by another device or group of devices (such as multispectral sensor device 205, classification device 210 and / or user device 215) separate from or including the classification device. Additionally or alternatively, Figure 4 One or more process blocks of may be performed by one or more components of device 300 , such as processor 320 , memory 330 , input component 340 , output component 350 , and / or communication component 360 .

[0064] like Figure 4 As shown in , process 400 may include receiving a spectral data set associated with an unknown sample (block 410). For example, a classification device may receive a spectral data set associated with an unknown sample, as described above.

[0065] like Figure 4As further shown in , process 400 may include obtaining a multi-class classification model that is used to classify an unknown sample into at least one of a plurality of classes, wherein the multi-class classification model includes a plurality of local auto-scaling OVO binary classifiers, each of the plurality of local auto-scaling OVO binary classifiers being associated with a different class pair from a plurality of classes (block 420). For example, a classification device may obtain a multi-class classification model that is used to classify an unknown sample into at least one of a plurality of classes, wherein the multi-class classification model includes a plurality of local auto-scaling OVO binary classifiers, each of the plurality of local auto-scaling OVO binary classifiers being associated with a different class pair from a plurality of classes, as described above.

[0066] like Figure 4 As further shown in , process 400 may include applying local autoscaling to the spectral dataset associated with the unknown sample to create a locally autoscaled spectral dataset (block 430). For example, the classification device may apply local autoscaling to the spectral dataset associated with the unknown sample to create a locally autoscaled spectral dataset, as described above.

[0067] like Figure 4 As further shown in , process 400 may include performing classification of unknown samples based on the locally auto-scaled spectral dataset and using a multi-class classification model including multiple locally auto-scaled OVO binary classifiers (block 440). For example, a classification device may perform classification of unknown samples based on the locally auto-scaled spectral dataset and using a multi-class classification model including multiple locally auto-scaled OVO binary classifiers, as described above.

[0068] Process 400 may include additional implementations, such as any single implementation or any combination of implementations described below and / or in combination with one or more other processes described elsewhere herein.

[0069] In a first implementation, applying local autoscaling to the spectral dataset includes performing mean centering on each column of data in the spectral dataset, and scaling each column of data in the spectral dataset after performing mean centering.

[0070] In a second implementation, either alone or in combination with the first implementation, each column of data in the spectral data set corresponds to a different wavelength of light.

[0071] In a third implementation, either alone or in combination with one or more of the first and second implementations, each row of data in the spectral data set corresponds to a spectral sample spanning a wavelength range of light.

[0072] In a fourth implementation, alone or in combination with one or more of the first to third implementations, performing classification of an unknown sample includes: determining, based on a locally auto-scaled spectral data set, votes provided by each of a plurality of locally auto-scaled OVO binary classifiers, identifying a class among a plurality of classes having a highest number of votes, and classifying the unknown sample as being included in the class among the plurality of classes having a highest number of votes.

[0073] In a fifth implementation, alone or in combination with one or more of the first to fourth implementations, process 400 includes obtaining a training data set of spectral data, generating a plurality of OVO binary classifiers based on the training data set, wherein each OVO binary classifier in the plurality of OVO binary classifiers is associated with a different class pair from a plurality of classes, and applying local auto-scaling to each OVO binary classifier in the plurality of OVO binary classifiers to create a plurality of local auto-scaled OVO binary classifiers.

[0074] In a sixth implementation, alone or in combination with one or more of the first to fifth implementations, the multi-class classification model is a SVM-based model.

[0075] In a seventh implementation, alone or in combination with one or more of the first to sixth implementations, process 400 includes providing information associated with a classification result for an unknown sample.

[0076] Although Figure 4 Example blocks of process 400 are shown, but in some implementations, Figure 4 Process 400 may include more blocks, fewer blocks, different blocks, or differently arranged blocks than those depicted in . Additionally or alternatively, two or more blocks of process 400 may be performed in parallel.

[0077] Figure 5 is a flow chart of an example process 500 associated with generating a multi-class classification model including multiple local auto-scaling OVO binary classifiers. In some implementations, Figure 5 One or more processing blocks of are performed by a classification device (e.g., classification device 210). In some implementations, Figure 5 One or more processing blocks of are performed by another device or group of devices separate from or including the classification device, such as a multispectral sensor device (e.g., multispectral sensor device 205) and / or a user device (e.g., user device 215). Additionally or alternatively, Figure 5 One or more processing blocks of may be performed by one or more components of device 300 , such as processor 320 , memory 330 , input component 340 , output component 350 , and / or communication component 360 .

[0078] like Figure 5 As shown in , process 500 may include obtaining a training data set of spectral data (block 510). For example, a classification device may obtain a training data set of spectral data, as described above.

[0079] like Figure 5 As further shown in , process 500 may include identifying multiple categories of the multi-class classification model based on the training data set (block 520). For example, the classification device may identify multiple categories of the multi-class classification model based on the training data set, as described above.

[0080] like Figure 5 As further shown in , process 500 may include generating a plurality of OVO binary classifiers of a multi-class classification model, wherein each OVO binary classifier in the plurality of OVO binary classifiers is associated with a different class pair from the plurality of classes (block 530). For example, a classification device may generate a plurality of OVO binary classifiers of a multi-class classification model, wherein each OVO binary classifier in the plurality of OVO binary classifiers is associated with a different class pair from the plurality of classes, as described above.

[0081] like Figure 5 As further shown in , process 500 may include applying local autoscaling to each OVO binary classifier in the plurality of OVO binary classifiers to create a plurality of local autoscaling OVO binary classifiers of the multi-class classification model (block 540). For example, the classification device may apply local autoscaling to each OVO binary classifier in the plurality of OVO binary classifiers to create a plurality of local autoscaling OVO binary classifiers of the multi-class classification model, as described above.

[0082] like Figure 5 As further shown in , process 500 may include storing a multi-class classification model including multiple local auto-scaling OVO binary classifiers (block 550). For example, the classification device may store a multi-class classification model including multiple local auto-scaling OVO binary classifiers, as described above.

[0083] Process 500 may include additional implementations, such as any single implementation or any combination of implementations described below and / or in combination with one or more other processes described elsewhere herein.

[0084] In a first implementation, process 500 includes receiving a spectral dataset associated with an unknown sample, applying local autoscaling to the spectral dataset associated with the unknown sample to create a locally autoscaled spectral dataset, and performing classification of the unknown sample based on the locally autoscaled spectral dataset and using a multi-class classification model including multiple locally autoscaled OVO binary classifiers.

[0085] In a second implementation, either alone or in combination with the first implementation, applying local autoscaling to each OVO binary classifier includes performing mean centering on each column of data associated with the given OVO binary classifier, and after performing mean centering, scaling the data associated with the given OVO binary classifier to unit variance.

[0086] In a third implementation, either alone or in combination with one or more of the first and second implementations, each column of data associated with a given OVO binary classifier corresponds to a different wavelength of light.

[0087] In a fourth implementation, either alone or in combination with one or more of the first to third implementations, each row of data with a given OVO binary classifier corresponds to a spectral scan spanning a wavelength range of light.

[0088] Although Figure 5 Example blocks of process 500 are shown, but in some implementations, Figure 5 Process 500 may include more blocks, fewer blocks, different blocks, or differently arranged blocks than those depicted in FIG. Additionally or alternatively, two or more blocks of process 500 may be performed in parallel.

[0089] The foregoing disclosure provides illustration and description, but is not intended to be exhaustive or to limit the implementations to the precise form disclosed. Modifications and variations may be made in light of the above disclosure or may be acquired from practice of the implementations.

[0090] As used herein, the term "component" is intended to be broadly interpreted as hardware, firmware, or a combination of hardware and software. Obviously, the systems and / or methods described herein can be implemented in different forms of hardware, firmware, and / or a combination of hardware and software. The actual dedicated control hardware or software code used to implement these systems and / or methods is not a limitation to the implementation. Therefore, the operation and behavior of the systems and / or methods are described herein without reference to specific software codes-it should be understood that software and hardware can be used to implement the systems and / or methods based on the description herein.

[0091] As used herein, satisfying a threshold may refer to being greater than a threshold, greater than or equal to a threshold, less than a threshold, less than or equal to a threshold, equal to a threshold, not equal to a threshold, etc., depending on the context.

[0092] Although specific combinations of features are described in the claims and / or disclosed in the specification, these combinations are not intended to limit the disclosure of various implementations. In fact, many of these features can be combined in a manner that is not specifically described in the claims and / or not disclosed in the specification. Although each dependent claim listed below may be directly subordinate to only one claim, the disclosure of various implementations includes the combination of each dependent claim with each other claim in the claim set. As used herein, the phrase "at least one" referring to a list of items refers to any combination of those items, including single members. As an example, "at least one of a, b, or c" is intended to cover a, b, c, ab, ac, bc, and abc, as well as any combination of multiple identical items.

[0093] When (within a single claim or across multiple claims) a “processor” or “one or more processors” (or another device or component, such as a “controller” or “one or more controllers”) is claimed or described as performing or being configured to perform multiple operations, the language is intended to broadly cover a variety of processor architectures and environments. For example, unless otherwise expressly stated (e.g., by use of “a first processor” and “a second processor” or other language that distinguishes the processors in the claims), the language is intended to cover a single processor that performs or is configured to perform all of the operations, a group of processors that collectively perform or are configured to perform all of the operations, a first processor that performs or is configured to perform a first operation and a second processor that performs or is configured to perform a second operation, or any combination of processors that perform or are configured to perform operations. For example, when a claim has the form “one or more processors configured to: perform X; perform Y; and perform Z,” the claim should be interpreted as “one or more processors configured to perform X; one or more (possibly different) processors configured to perform Y; and one or more (possibly different) processors configured to perform Z.”

[0094] Any element, action or instruction used herein should not be interpreted as key or necessary unless explicitly described as such. In addition, as used herein, the articles "one" and "an" are intended to include one or more items, and can be used interchangeably with "one or more". In addition, as used herein, the article "the" is intended to include one or more items related to the article "the", and can be used interchangeably with "the one or more". In addition, as used herein, the term "set" is intended to include one or more items (e.g., a combination of related items, unrelated items, or related and unrelated items), and can be used interchangeably with "one or more". In the case of only meaning one item, the phrase "only one" or similar language is used. In addition, as used herein, the terms "have", "have", "have", etc. are intended to be open terms. In addition, unless otherwise clearly stated, the phrase "based on" is intended to mean "based at least in part". In addition, as used herein, unless otherwise clearly stated (e.g., if used in combination with "any" or "only one"), the term "or" is intended to be inclusive when used in series, and can be used interchangeably with "and / or".

Claims

1. A method comprising: receiving, by a device, a spectral data set associated with an unknown sample; Obtaining a multi-class classification model by the device, the multi-class classification model is to be used to classify the unknown sample into at least one of a plurality of classes; wherein the multi-class classification model comprises a plurality of local auto-scaling one-to-one OVO binary classifiers, each of the plurality of local auto-scaling OVO binary classifiers being associated with a different class pair from the plurality of classes; applying, by the device, local autoscaling to the spectral dataset associated with the unknown sample to create a locally autoscaled spectral dataset; and Classification of the unknown sample is performed by the device based on the locally auto-scaled spectral dataset and using the multi-class classification model including the plurality of locally auto-scaled OVO binary classifiers.

2. The method of claim 1 , wherein applying the local automatic scaling to the spectral data set comprises: Performing mean centering on each column of data in the spectral dataset; as well as Each column of data of the spectral dataset is scaled after performing the mean centering. The method according to claim 1 , wherein each column of data in the spectral data set corresponds to light of a different wavelength. The method of claim 1 , wherein each row of data in the spectral data set corresponds to a spectral sample spanning a wavelength range of light.

5. The method of claim 1, wherein performing classification of the unknown sample comprises: determining a vote provided by each of the plurality of locally auto-scaled OVO binary classifiers based on the locally auto-scaled spectral dataset, identifying the category of the plurality of categories having the highest number of votes, and The unknown sample is classified as being included in the category having the highest number of votes among the plurality of categories.

6. The method according to claim 1, further comprising: Obtain a training dataset of spectral data, generating a plurality of OVO binary classifiers based on the training data set, wherein each OVO binary classifier of the plurality of OVO binary classifiers is associated with a different class pair from the plurality of classes, and Local autoscaling is applied to each OVO binary classifier in the plurality of OVO binary classifiers to create the plurality of local autoscaling OVO binary classifiers.

7. The method according to claim 1, wherein the multi-class classification model is a support vector machine (SVM) based model.

8. The method according to claim 1, further comprising: Information associated with the results of the classification of the unknown sample is provided.

9. A device comprising: one or more memories; as well as one or more processors coupled to the one or more memories, the one or more processors configured to: receiving a spectral data set associated with an unknown sample; Obtaining a multi-class classification model, the multi-class classification model is to be used to classify the unknown sample into at least one of a plurality of classes; wherein the multi-class classification model comprises a plurality of local auto-scaling one-to-one OVO binary classifiers, each of the plurality of local auto-scaling OVO binary classifiers being associated with a different class pair from the plurality of classes; applying local autoscaling to the spectral dataset associated with the unknown sample to create a locally autoscaled spectral dataset; and Classification of the unknown sample is performed based on the locally autoscaled spectral dataset and using the multi-class classification model including the plurality of locally autoscaled OVO binary classifiers.

10. The apparatus of claim 9, wherein to apply the local autoscaling to the spectral data set, the one or more processors are configured to: performing mean centering on each column of data in the spectral dataset; and Each column of data of the spectral dataset is scaled after performing the mean centering.

11. The apparatus of claim 9, wherein each column of data in the spectral data set corresponds to light of a different wavelength.

12. The apparatus of claim 9, wherein each row of data in the spectral data set corresponds to a spectral sample spanning a wavelength range of light.

13. The apparatus of claim 9, wherein to perform the classification of the unknown sample, the one or more processors are configured to: determining a vote provided by each of the plurality of locally auto-scaled OVO binary classifiers based on the locally auto-scaled spectral dataset, identifying the category of the plurality of categories having the highest number of votes, and The unknown sample is classified as being included in the category having the highest number of votes among the plurality of categories.

14. The device of claim 9, wherein the one or more processors are further configured to: Obtain a training dataset of spectral data, Generate multiple one-to-one OVO binary classifiers based on the training data set, wherein each OVO binary classifier of the plurality of OVO binary classifiers is associated with a different class pair from the plurality of classes, and Local autoscaling is applied to each OVO binary classifier in the plurality of OVO binary classifiers to create the plurality of local autoscaling OVO binary classifiers.

15. The apparatus of claim 9, wherein the multi-class classification model is a support vector machine (SVM) based model.

16. The apparatus of claim 9, wherein the one or more processors are further configured to provide information associated with a result of the classification of the unknown sample.

17. A method for generating a multi-class classification model, comprising: A training data set of spectral data obtained by the device; identifying, by the device and based on the training data set, a plurality of classes of the multi-class classification model; generating, by the device, a plurality of one-to-one OVO binary classifiers of the multi-class classification model, wherein each OVO binary classifier in the plurality of OVO binary classifiers is associated with a different class pair from the plurality of classes; applying, by the device, local autoscaling to each of the plurality of OVO binary classifiers to create a plurality of local autoscaling OVO binary classifiers of the multi-class classification model; and The multi-class classification model including the plurality of local auto-scaling OVO binary classifiers is stored by the device.

18. The method according to claim 17, further comprising: receiving a spectral data set associated with an unknown sample; applying local autoscaling to the spectral dataset associated with the unknown sample to create a locally autoscaled spectral dataset; and Classification of the unknown sample is performed based on the locally autoscaled spectral dataset and using the multi-class classification model including the plurality of locally autoscaled OVO binary classifiers.

19. The method of claim 17, wherein applying the local autoscaling to each OVO binary classifier comprises: Perform mean centering on each column of data associated with a given OVO binary classifier; as well as Each column of data associated with the given OVO binary classifier is scaled to unit variance after performing the mean centering.

20. The method of claim 19, wherein each column of data associated with the given OVO binary classifier corresponds to a different wavelength of light.

Images