Reduced false alarm identification for spectral classification

Abstract

The invention relates to reduced misinformation identification for spectral classification. A device may receive information identifying results of a set of spectral measurements of a training set of known samples and a validation set of known samples. The device may generate a classification model based on information identifying results of the set of spectral measurements, where the classification model includes at least one category associated with the material of interest for the spectrometric determination, and where the classification model includes at least one classification associated with the material of interest for the spectrometric determination. The classification model includes a mismatch category associated with at least one of at least one material of uninterest or a baseline spectral measurement. A device may receive information identifying a particular outcome of a particular spectral measurement of an unknown sample. The device may determine, using the classification model, whether an unknown sample is included in a mismatch category. The device may provide an output indicating whether the unknown sample is included in the mismatch category.

Description

[0001] Divisional Application Instructions

[0002] This application is a divisional application of application No. 202111346296.9, filed on January 21, 2019, entitled "Reduced False Alarm Identification for Spectral Classification", which is itself a divisional application of application No. 202111346296.9, filed on January 21, 2019, entitled "Reduced False Alarm Identification for Spectral Classification". Technical Field

[0003] This application relates to, but is not limited to, reduced false alarm identification for spectral classification. background

[0004] Raw material identification can be used for quality control of pharmaceutical products. For example, raw material identification can be performed on medical materials to determine whether the components of the medical material correspond to the packaging label associated with the medical material. Similarly, raw material quantification can be performed to determine the concentration of a specific chemical in a specific sample. Compared with other chemometric techniques, spectroscopic methods can facilitate the non-destructive identification and / or quantification of raw materials with reduced preparation and data acquisition time.

[0005] Overview

[0006] According to some possible implementations, the device may include one or more memories and one or more processors communicatively coupled to one or more memories. The device may receive information about the results of a set of spectral measurements identifying a training set of known samples and a validation set of known samples. The device may generate a classification model based on the information about the results of the set of spectral measurements, wherein the classification model includes at least one category associated with the material of interest used for spectroscopic determination, and wherein the classification model includes at least one no-match class associated with at least one material of no interest or at least one baseline spectral measurement. The device may receive information about the specific results of a specific spectral measurement identifying an unknown sample. The device may use the classification model to determine whether the unknown sample is included in the no-match class. The device may provide output indicating whether the unknown sample is included in the no-match class.

[0007] According to some possible implementations, a non-transitory computer-readable medium may store one or more instructions that, when executed by one or more processors, cause one or more processors to receive information identifying the results of spectral measurements performed on an unknown sample. When executed by one or more processors, the instructions may cause one or more processors to aggregate multiple categories of a classification model to generate an aggregated classification model. When executed by one or more processors, the instructions may cause one or more processors to use the aggregated classification model to determine that the spectral measurements were accurately performed. When executed by one or more processors, the instructions may cause one or more processors to determine, based on the determination that the spectral measurements were accurately performed, use the classification model to determine that the unknown sample is not included in a mismatch category of the classification model, wherein the mismatch category is related to a material of no interest or a baseline spectral measurement. When executed by one or more processors, the instructions may cause one or more processors to perform spectral classification of the unknown sample based on the determination that the unknown sample is not included in a mismatch category. When executed by one or more processors, the instructions may cause one or more processors to provide information identifying the unknown sample based on performing spectral classification of the unknown sample.

[0008] According to some possible implementations, the method may include obtaining the results of a set of spectral measurements via a device. The method may include generating a support vector machine (SVM)-based classification model by the device based on the results of the set of spectral measurements. The classification model includes multiple categories corresponding to a plurality of materials of interest for classification, wherein the set of spectral measurements includes a threshold number of measurements of samples of the plurality of materials of interest, wherein the classification model includes specific categories that do not correspond to the plurality of materials of interest for classification, and wherein the set of spectral measurements includes measurements of samples associated with a specific category that are less than a threshold number. The method may include classifying a specific spectral measurement of a specific sample into a specific category by the device using the classification model. The method may include providing information indicating that a specific sample is assigned to a specific category by the device based on the classification of the specific spectral measurements.

[0009] 1) An apparatus comprising:

[0010] One or more memories; and

[0011] One or more processors, communicatively coupled to one or more memories, the one or more processors being used for:

[0012] Receive information about the results of a set of spectral measurements that identify a training set of known samples and a validation set of known samples;

[0013] A classification model is generated based on the information obtained from the results of the set of spectral measurements. The classification model includes at least one category associated with the material of interest used for the spectral measurement, and the classification model includes at least one mismatch category associated with at least one material of no interest or at least one baseline spectral measurement.

[0014] Receive information about the specific results of specific spectral measurements used to identify unknown samples;

[0015] The classification model is used to determine whether the unknown sample is included in the mismatch category; and

[0016] Provides output indicating whether the unknown sample is included in the mismatch category.

[0017] 2) The device according to 1), wherein the one or more processors, when determining whether the unknown sample is included in the mismatch category:

[0018] Based on the classification model, it is determined that the unknown sample is included in the mismatch category; and

[0019] Wherein, when one or more processors provide output indicating whether the unknown sample is included in the mismatch category:

[0020] Provides output indicating whether the unknown sample is included in the mismatch category.

[0021] 3) The device according to 1), wherein the one or more processors, when determining whether the unknown sample is included in the mismatch category, are configured to:

[0022] Based on the classification model, it is determined that the unknown sample is not included in the mismatch category;

[0023] The classification model is used to determine the classification of the unknown sample based on the determination that the unknown sample is not included in the mismatch category; and

[0024] Wherein, the one or more processors, when providing output indicating whether the unknown sample is included in the mismatch category, are used for:

[0025] Provides output that identifies the classification of the unknown sample.

[0026] 4) The apparatus according to 1), wherein the one or more processors, upon receiving the information identifying the results of the set of spectral measurements, are configured to:

[0027] Receive information identifying a set of baseline spectral measurements; and

[0028] Wherein, the one or more processors, when generating the classification model, are used for:

[0029] The mismatch categories are trained based on the set of baseline spectral measurements.

[0030] 5) The apparatus according to 4), wherein the set of baseline spectral measurements is associated with at least one of the following:

[0031] Measurements performed using incorrect measuring distances.

[0032] Measurements performed using an incorrect measurement context

[0033] Measurements performed using incorrect measuring lighting, or

[0034] Measurements performed in the absence of a sample.

[0035] 6) The apparatus according to 1), wherein the one or more processors, upon receiving the information identifying the results of the set of spectral measurements, are configured to:

[0036] Receive information identifying the at least one material of no interest; and

[0037] Wherein, the one or more processors, when generating the classification model, are used for:

[0038] The mismatch category of the classification model is trained based on the information used to identify the at least one type of material that is not of interest.

[0039] 7) The apparatus according to 1), wherein the one or more processors, when using the classification model to determine whether the unknown sample is included in the mismatch category:

[0040] A confidence metric based on support vector machines is used to determine whether the unknown sample is included in the mismatch category.

[0041] 8) The device according to 7), wherein the confidence metric is at least one of the following:

[0042] Probability estimation, or

[0043] Decision value.

[0044] 9) The device according to 1), wherein the classification model is a first classification model; and

[0045] Wherein, when one or more processors determine whether the unknown sample is included in the mismatch category:

[0046] Perform a first classification using the first classification model to identify a set of local categories of the first classification model for the specific spectral measurement;

[0047] A second classification model is generated based on the set of local categories, the second classification model including the mismatch categories; and

[0048] A second classification is performed to determine whether the unknown sample is included in the mismatch category.

[0049] 10) A non-transitory computer-readable medium storing instructions, the instructions comprising:

[0050] One or more instructions, which, when executed by one or more processors, cause the one or more processors to perform the following actions:

[0051] Receive information identifying the results of spectral measurements performed on unknown samples;

[0052] Multiple categories in a clustered classification model are used to generate a clustered classification model;

[0053] The classification model of the cluster is used to determine that the spectral measurement was performed accurately;

[0054] Based on the determination that the spectral measurements were performed accurately and the classification model was used, it was determined that the unknown sample was not included in the mismatch category of the classification model, the mismatch category being related to materials of no interest or baseline spectral measurements;

[0055] Based on the determination that the unknown sample is not included in the mismatch category, spectral classification of the unknown sample is performed; and

[0056] Based on the spectral classification performed on the unknown sample, information for identifying the unknown sample is provided.

[0057] 11) The non-transitory computer-readable medium according to 10), wherein the one or more instructions that cause the one or more processors to determine that the unknown sample is not included in the mismatch category cause the one or more processors to:

[0058] Based on the confidence metric associated with the classification model meeting a threshold, it is determined that the unknown sample is not included in the mismatch category.

[0059] 12) The non-transitory computer-readable medium according to 11), wherein the one or more instructions, when executed by the one or more processors, also cause the one or more processors to:

[0060] The confidence metric is determined by dividing the classification model into multiple sub-models using either a one-to-all technique or a full-to-all technique.

[0061] 13) The non-transitory computer-readable medium according to 11), wherein the classification model includes a greater than threshold number of categories; and

[0062] Wherein, the one or more instructions that cause the one or more processors to execute the spectral classification cause the one or more processors to:

[0063] Perform a first spectral classification of the unknown sample based on the classification model;

[0064] Based on the execution of the first spectral classification, another classification model is generated using a subset of the categories of the classification model;

[0065] Based on the other classification model, it is determined that the unknown sample is not included in the mismatch category; and

[0066] Perform a second classification to identify the unknown samples.

[0067] 14) A method comprising:

[0068] The equipment obtains a set of spectral measurement results;

[0069] The device generates a support vector machine (SVM)-based classification model based on the results of the set of spectral measurements. The classification model includes multiple categories corresponding to multiple materials of interest for classification. The set of spectral measurements includes measurements of a threshold number of samples of the multiple materials of interest. The classification model includes specific categories that do not correspond to the multiple materials of interest for classification. The set of spectral measurements includes measurements of a number of samples associated with the specific category that is less than the threshold number.

[0070] The device uses the classification model to classify a specific spectral measurement of a specific sample into the specific category; and

[0071] The device provides information indicating that a particular sample is assigned to a particular category based on the classification of the particular spectral measurements.

[0072] 15) According to the method of 14), wherein classifying the specific spectral measurements includes:

[0073] The classification model is divided into multiple sub-models, each of which corresponds to a comparison between the corresponding category of the classification model and each other category of the classification model.

[0074] Determine multiple decision values ​​corresponding to the multiple sub-models; and

[0075] The specific category is selected for the specific sample based on the multiple decision values.

[0076] 16) The method according to 14), wherein classifying the specific spectral measurements includes:

[0077] The classification model is divided into multiple sub-models, and the multiple sub-models correspond to comparisons between each category of the classification model;

[0078] Determine multiple decision values ​​corresponding to the multiple sub-models; and

[0079] The specific category is selected for the specific sample based on the multiple decision values.

[0080] 17) The method according to 14), wherein classifying the specific spectral measurements includes:

[0081] The specific spectral measurements are classified using either a radial basis function type kernel function or a linear kernel type kernel function.

[0082] 18) The method according to 14), wherein the set of spectroscopic measurements includes baseline spectroscopic measurements and spectroscopic measurements of materials of no interest; and

[0083] The baseline spectral measurements and the spectral measurements of the materials of no interest are categorized into the specific category.

[0084] 19) The method according to 14), wherein classifying the specific spectral measurements includes:

[0085] The specific spectral measurements are classified using an in-situ local classification model generated based on the classification model.

[0086] 20) According to the method of 14), wherein classifying the specific spectral measurements includes:

[0087] The categories of the classification model are clustered into a single category; and

[0088] The specific spectral measurement is classified based on the single category. Attached Figure Description

[0089] Figure 1A and Figure 1B This is a diagram summarizing the exemplary embodiments described in this article;

[0090] Figure 2 This is a diagram of an example environment in which the systems and / or methods described in this article may be implemented;

[0091] Figure 3yes Figure 2 A diagram of example components of one or more devices;

[0092] Figure 4 This is a flowchart illustrating an example process for generating a classification model for spectral classification.

[0093] Figure 5 Is with Figure 4 The diagrams shown are related to example implementations of the example processes.

[0094] Figure 6 This is a flowchart of an example process for avoiding false positive identification during spectral classification; and

[0095] Figure 7A and Figure 7B Is with Figure 6 The diagram shows an example implementation related to the example process. Detailed description

[0096] The following detailed description of the exemplary embodiments is taken with reference to the accompanying drawings. The same reference numerals in different drawings may identify the same or similar elements.

[0097] Raw material identification (RMID) is a technique used to identify the components (e.g., ingredients) of a particular sample for identification, verification, etc. For example, RMID can be used to verify whether an ingredient in a pharmaceutical material corresponds to a set of ingredients identified on a label. Similarly, raw material quantification is a technique used to perform quantitative analysis on a particular sample, such as determining the concentration of a specific material in a particular sample. A spectrometer can be used to perform spectroscopic methods on a sample (e.g., pharmaceutical material) to determine the components of the sample, the concentration of the components, etc. A spectrometer can determine a set of measurements of a sample and can provide a set of measurements for spectroscopic determination. Spectral classification techniques (e.g., classifiers) can help determine the components of a sample based on a set of measurements.

[0098] However, some unknown samples to be classified spectrally are not actually included in the categories that the classification model is configured to classify. For example, for a classification model trained to distinguish types of fish, a user might inadvertently provide beef for classification. In this case, the control device could perform spectral classification of a specific material and provide a false positive identification of that material as a specific type of fish, which would be inaccurate.

[0099] As another example, classification models can be trained to classify sugar types (e.g., glucose, fructose, galactose, etc.) and quantify the concentration of each type of sugar in an unknown sample. However, users of the spectrometer and control equipment may inadvertently attempt to classify unknown sugar samples based on incorrect measurements performed using the spectrometer. For example, a user might operate the spectrometer at an incorrect distance from the unknown sample, under environmental conditions different from calibration conditions, and / or similar conditions under which spectroscopy was performed to train the classification model. In this case, when the unknown sample is actually a sugar of type two at a second concentration, the control equipment may receive an inaccurate spectrum for the unknown sample, leading to a false positive identification of the unknown sample as a sugar of type one at a first concentration.

[0100] Some implementations described herein can reduce false alarms in spectroscopic methods by utilizing mismatched categories used in classification models. For example, a control device receiving spectral measurements of an unknown sample can determine whether to assign the unknown sample to a mismatched category. In some implementations, the control device can determine that the unknown sample will be assigned to a mismatched category and can provide information indicating that the unknown sample is assigned to a mismatched category, thereby avoiding false alarms for the unknown sample. Alternatively, based on the determination that the unknown sample will not be assigned to a mismatched category, the control device can analyze the spectrum of the unknown sample to provide, for example, spectral determinations of classification, concentration, etc. Furthermore, the control device can utilize confidence metrics (such as probability estimates, decision values, etc.) to filter out false alarms.

[0101] In this way, the accuracy of the spectral method is improved compared to a spectral method performed without using mismatch categories and / or confidence metrics. Furthermore, mismatch categories can be used when generating a classification model based on a training set of known spectral samples. For example, the control device can determine that a sample in the training set does not correspond to a type corresponding to the rest of the training set (e.g., based on human error that leads to the introduction of incorrect samples into the training set), and can determine that data about the samples should not be included when generating the classification model. In this way, the control device improves the accuracy of the classification model used for the spectral method.

[0102] Figure 1A and Figure 1B This is a diagram summarizing the exemplary embodiment 100 described herein. (See diagram for example.) Figure 1A As shown, Example Implementation 100 may include a control device and a spectrometer.

[0103] like Figure 1AAs further illustrated, the control device can cause the spectrometer to perform a set of spectroscopic measurements on the training and validation sets (e.g., known sample sets used for training and validating the classification model). The training and validation sets can be selected to include a threshold number of samples for each category used in the classification model. The categories of the classification model can refer to groups of similar materials that share one or more common characteristics, such as (in a pharmaceutical context) lactose, fructose, acetaminophen, ibuprofen, aspirin, etc. The materials used to train the classification model, and those for which the classification model is to be used to perform ingredient identification, can be referred to as the materials of interest.

[0104] like Figure 1A As further illustrated, the spectrometer can perform a set of spectral measurements on the training and validation sets based on instructions received from the control device. For example, the spectrometer can determine the spectrum for each sample in both the training and validation sets, enabling the control device to generate a set of categories for classifying unknown samples as one of the materials of interest for a classification model.

[0105] A spectrometer can provide a set of spectral measurements to a control device. The control device can then use specific measurement techniques and generate a classification model based on this set of spectral measurements. For example, the control device can use a support vector machine (SVM) technique (e.g., a machine learning technique for information measurement) to generate a global classification model. The global classification model can include information associated with assigning a specific spectrum to a specific category of material of interest, and can also include information associated with identifying the type of material of interest associated with that specific category. In this way, the control device can provide information to identify the type of material in an unknown sample based on assigning the spectrum of the unknown sample to a specific category.

[0106] In some implementations, the control device may receive spectra associated with samples belonging to a mismatch category. For example, the control device may receive spectra determined to be similar to those of the material of interest, spectra associated with materials that may be confused with the material of interest (e.g., visually, chemically, etc.), spectra associated with incorrect operation of the spectrometer (e.g., spectra of measurements performed without samples, spectra of measurements performed at incorrect distances between the sample and the optics of the spectrometer, etc.), and / or similar spectra. Materials that are not of interest, and materials that may be included in the mismatch category, may be referred to as nuisance materials or materials of no interest. In this case, the control device may generate mismatch categories for a classification model and may use the mismatch categories to verify false positive identification avoidance based on the spectra of nuisance materials included in the validation set. Alternatively or concurrently, during the use of the classification model, the control device may receive information identifying nuisance materials and may update the classification model to avoid false positive identifications (e.g., identifying nuisance materials as one of the materials of interest).

[0107] like Figure 1B As shown, the control device can receive a classification model (e.g., from storage, from another control device that generates the classification model, etc.). The control device can cause the spectrometer to perform a set of spectral measurements on an unknown sample (e.g., an unknown sample to which classification or quantification is to be performed). The spectrometer can perform a set of spectral measurements based on instructions received from the control device. For example, the spectrometer can determine the spectrum of the unknown sample. The spectrometer can provide a set of spectral measurements to the control device. The control device can attempt to classify the unknown sample based on a classification model, such as using a multi-stage classification technique.

[0108] about Figure 1B The control device can attempt to use a classification model to determine whether an unknown sample belongs to a mismatched category. For example, the control device can determine a confidence metric corresponding to the probability that the unknown sample belongs to a mismatched category. In this case, based on the control device's determination that the confidence metric (e.g., probability estimate, decision value output of a support vector machine, etc.) meets a threshold, the control device can assign the unknown sample to the mismatched category. In this case, the control device can report that the unknown sample cannot be accurately classified using the classification model, thereby reducing the likelihood of the unknown sample being falsely identified as belonging to a category of material of interest.

[0109] In some implementations, based on a first determination that an unknown sample does not belong to a mismatched category, the control device may attempt to perform a determination for a specific sample in the unknown set using in-situ local modeling. For example, the control device may determine a set of confidence metrics associated with a specific sample and a global classification model. In this case, the control device may select a subset of categories from the global classification model based on one or more corresponding confidence metrics and may generate a local classification model based on this set of categories. The local classification model may be an in-situ classification model generated using SVM techniques and a subset of categories. Based on the generated in-situ classification model, the control device may attempt to classify the unknown sample based on the local classification model. In this case, if one or more confidence metrics associated with the local classification model satisfy a threshold, the control device may determine that the unknown sample does indeed belong to a mismatched category and may report that the unknown sample cannot be classified using the classification model. Alternatively, the control device may determine that the unknown sample does not belong to a mismatched category and may report the classification associated with the unknown sample.

[0110] In this way, based on reducing the possibility of false positives in reporting unknown samples as materials of interest, the control device achieves improved accuracy for spectral methods for unknown samples compared to other classification models.

[0111] As indicated above, Figure 1A and Figure 1B This is provided as an example only. Other examples are possible and may differ from those provided. Figure 1A and Figure 1B Example of the description.

[0112] Figure 2 This is a diagram of an example environment 200 in which the systems and / or methods described herein can be implemented. (See diagram 200 for example.) Figure 2 As shown, environment 200 may include control device 210, spectrometer 220, and network 230. The devices in environment 200 may be interconnected via wired connection, wireless connection, or a combination of wired and wireless connection.

[0113] Control device 210 may include one or more devices capable of storing, processing, and / or routing information associated with spectral classification. For example, control device 210 may include a server, computer, wearable device, cloud computing device, and / or a similar device that generates a classification model based on a set of measurements on a training set, validates a classification model based on a set of measurements on a validation set, and / or utilizes the classification model to perform spectral classification based on a set of measurements in an unknown set. In some embodiments, as described herein, control device 210 may utilize machine learning techniques to determine whether spectral measurements of unknown samples will be classified into a mismatched category to reduce the likelihood of false positives. In some embodiments, control device 210 may be associated with a specific spectrometer 220. In some embodiments, control device 210 may be associated with multiple spectrometers 220. In some embodiments, control device 210 may receive information from and / or transmit information to another device in environment 200 (e.g., spectrometer 220).

[0114] Spectrometer 220 may include one or more devices capable of performing spectral measurements on a sample. For example, spectrometer 220 may include spectroscopic devices that perform spectroscopic methods (e.g., vibrational spectroscopy, such as near-infrared (NIR) spectrometers, mid-infrared (mid-IR) spectrometers, Raman spectroscopy, etc.). In some embodiments, spectrometer 220 may be integrated into a wearable device, such as a wearable spectrometer and / or similar devices. In some embodiments, spectrometer 220 may receive information from and / or transmit information to another device in environment 200 (e.g., control device 210).

[0115] Network 230 may include one or more wired and / or wireless networks. For example, network 230 may include cellular networks (e.g., Long Term Evolution (LTE) networks, 3G networks, Code Division Multiple Access (CDMA) networks, etc.), Public Land Mobile Networks (PLMN), Local Area Networks (LAN), Wide Area Networks (WAN), Metropolitan Area Networks (MAN), telephone networks (e.g., Public Switched Telephone Network (PSTN)), private networks, self-organizing networks, intranets, the Internet, fiber-optic networks, cloud computing networks, etc., and / or combinations of these or other types of networks.

[0116] Figure 2 The number and layout of devices and networks shown are provided as examples. In reality, with... Figure 2 Compared to the devices and / or networks shown, there may be additional devices and / or networks, fewer devices and / or networks, different devices and / or networks, or devices and / or networks arranged differently. Furthermore, Figure 2The two or more devices shown can be implemented within a single device, or Figure 2 The single device shown herein can be implemented as multiple distributed devices. For example, although control device 210 and spectrometer 220 are described herein as two separate devices, control device 210 and spectrometer 220 can be implemented within a single device. Alternatively or alternatively, a group of devices in environment 200 (e.g., one or more devices) can perform one or more functions described as being performed by another group of devices in environment 200.

[0117] Figure 3 This is a diagram of example components of device 300. Device 300 may correspond to control device 210 and / or spectrometer 220. In some embodiments, control device 210 and / or spectrometer 220 may include one or more devices 300 and / or one or more components of device 300. Figure 3 As shown, device 300 may include bus 310, processor 320, memory 330, storage component 340, input component 350, output component 360, and communication interface 370.

[0118] Bus 310 includes components that allow communication among the components of device 300. Processor 320 is implemented in hardware, firmware, or a combination of hardware and software. Processor 320 is a central processing unit (CPU), graphics processing unit (GPU), accelerated processing unit (APU), microprocessor, microcontroller, digital signal processor (DSP), field-programmable gate array (FPGA), application-specific integrated circuit (ASIC), or other type of processing component. In some embodiments, processor 320 includes one or more processors capable of being programmed to perform functions. Memory 330 includes random access memory (RAM), read-only memory (ROM), and / or another type of dynamic or static storage device (e.g., flash memory, magnetic storage, and / or optical storage) for use by processor 320.

[0119] Storage component 340 stores information and / or software related to the operation and use of device 300. For example, storage component 340 may include hard disks (e.g., magnetic disks, optical disks, magneto-optical disks, and / or solid-state drives), compact optical disks (CDs), digital universal disks (DVDs), floppy disks, cartridges, magnetic tapes, and / or other types of non-transitory computer-readable media along with corresponding drives.

[0120] Input component 350 includes components that allow device 300 to receive information, for example, via user input (e.g., a touchscreen display, keyboard, keypad, mouse, button, switch, and / or microphone). Alternatively, input component 350 may include sensors for sensing information (e.g., a Global Positioning System (GPS) component, accelerometer, gyroscope, and / or actuator). Output component 360 includes components that provide output information from device 300 (e.g., a display, speaker, and / or one or more light-emitting diodes (LEDs)).

[0121] Communication interface 370 includes similar transceiver components (e.g., a transceiver and / or separate receiver and transmitter) that enable device 300 to communicate with other devices, for example, via a wired connection, a wireless connection, or a combination of wired and wireless connections. Communication interface 370 may allow device 300 to receive information from and / or provide information to another device. For example, communication interface 370 may include an Ethernet interface, an optical interface, a coaxial interface, an infrared interface, a radio frequency (RF) interface, a universal serial bus (USB) interface, a wireless local area network (WLAN) interface, a cellular network (CNN) interface, etc.

[0122] Device 300 can perform one or more of the processes described herein. Device 300 can perform these processes based on software instructions stored in non-transitory computer-readable media (e.g., memory 330 and / or storage component 340) executed by processor 320. Computer-readable media are defined herein as non-transitory memory devices. Memory devices include memory space within a single physical storage device or memory space distributed across multiple physical storage devices.

[0123] Software instructions may be read into memory 330 and / or storage component 340 via communication interface 370 from another computer-readable medium or from another device. When executed, the software instructions stored in memory 330 and / or storage component 340 may cause processor 320 to perform one or more processes described herein. Alternatively or additionally, hard-wired circuitry may be used in place of or in combination with software instructions to perform one or more processes described herein. Therefore, the embodiments described herein are not limited to any particular combination of hardware circuitry and software.

[0124] Figure 3 The number and arrangement of components shown are provided as an example. In fact, with... Figure 3 Compared to the components shown, device 300 may include additional components, fewer components, different components, or components arranged differently. Alternatively, a set of components of device 300 (e.g., one or more components) may perform one or more functions described as being performed by another set of components of device 300.

[0125] Figure 4 This is a flowchart of an example process 400 for generating a classification model for spectral classification. In some implementations, Figure 4 One or more process frames can be executed by control device 210. In some embodiments, Figure 4 One or more process frames may be performed by another device or a group of devices (such as spectrometer 220) that are separate from or include control device 210.

[0126] like Figure 4 As shown, process 400 may include having a set of spectral measurements performed against a training set and / or a validation set (box 410). For example, control device 210 may (e.g., using processor 320, communication interface 370, etc.) cause spectrometer 220 to perform a set of spectral measurements against the training set and / or validation set of samples to determine the spectrum for each sample in the training set and / or validation set. The training set may refer to a set of samples of one or more known materials used to generate a classification model. Similarly, the validation set may refer to a set of samples of one or more known materials used to validate the accuracy of the classification model. For example, the training set and / or validation set may include one or more versions of a set of materials (e.g., one or more versions manufactured by different manufacturers to control for manufacturing differences).

[0127] In some implementations, the training and / or validation sets may be selected based on a set of materials of interest for which a classification model will be used to perform spectral classification. For example, when it is anticipated to perform spectral quantization on a pharmaceutical material to determine the presence of a specific component of the pharmaceutical material, the training and / or validation sets may include a set of samples of active pharmaceutical ingredients (APIs), excipients, etc., at a set of different possible concentrations.

[0128] In some implementations, the training and / or validation sets can be selected to include a specific number of samples for each type of material. For example, the training and / or validation sets can be selected to include multiple samples (e.g., 5 samples, 10 samples, 15 samples, 50 samples, etc.) and / or their concentrations for a particular material. In some implementations, the number of samples can be less than a threshold. For example, categories of homogeneous organic compounds can be generated based on 50 spectra from 10 samples (e.g., spectral scans), 15 spectra from 3 samples, etc. Similarly, for heterogeneous organic compounds, categories can be generated based on, for example, 100 spectra from 20 samples, 50 spectra from 10 samples, etc. Similarly, categories of biological or agricultural materials can be generated based on 400 spectra from 40 samples, 200 spectra from 20 samples, etc. In some implementations, the number of samples and / or spectra used to interfere with the mismatched class of the material can be associated with the same or reduced number of samples and / or spectra of the non-mismatched class of the same type of material (e.g., homogeneous organic compounds, heterogeneous organic compounds, biological or agricultural materials, etc.). In this way, the control device 210 can be provided with a threshold number of spectra associated with a specific type of material, thereby facilitating the generation and / or verification of classes for a classification model (e.g., a global classification model, a local classification model, etc.) or a quantization model, allowing unknown samples to be accurately assigned to that class, and the quantization model to be used to quantify the spectra assigned to the class associated with the quantization model.

[0129] In some implementations, one or more samples of materials to be assigned to a mismatched category may be included in a training set and / or a validation set. For example, spectrometer 220 may provide measurements of a first material associated with a spectrum similar to that of a second material to be quantized using a quantization model. In this way, control device 210 may use machine learning to train for the avoidance of false alarm identification. In some implementations, control device 210 may select materials for the mismatched category based on received information. For example, control device 210 may receive information identifying interfering materials having spectra, appearances, etc., similar to those of the material of interest at a specific concentration for which a classification model is to be generated. Alternatively or additionally, control device 210 may perform machine learning techniques to automatically identify interfering materials of a particular material of interest. For example, control device 210 may use machine learning to perform pattern recognition to identify spectra of interfering materials that are similar to those of the material of interest, to identify interfering materials that visually appear similar to those of the material of interest, and / or the like.

[0130] In some implementations, control device 210 may enable baseline spectral measurements to be performed to identify spectra in mismatch categories. For example, control device 210 may enable spectral measurements as baseline measurements under conditions such as the absence of a sample, incorrect background, incorrect illumination, and / or similar circumstances, to ensure that incorrect spectral measurements are classified as mismatched categories rather than as belonging to the material of particular interest. In this case, control device 210 may automatically control spectrometer 220, provide information using a user interface to guide the user of spectrometer 220 in performing incorrect measurements, etc. Alternatively, control device 210 may receive information indicating that a particular spectral measurement was incorrectly performed, thus enabling the generation of mismatch categories.

[0131] In some embodiments, the control device 210 may cause multiple spectrometers 220 to perform a set of spectral measurements in response to one or more physical conditions. For example, the control device 210 may cause a first spectrometer 220 and a second spectrometer 220 to perform a set of vibrational spectral measurements using NIR spectroscopy. Alternatively or concurrently, the control device 210 may cause a set of spectral measurements to be performed at multiple times, at multiple locations, under multiple different laboratory conditions, etc. In this way, the control device 210 reduces the possibility of inaccurate spectral measurements as a result of the physical conditions under which a set of spectral measurements is performed by a single spectrometer 220.

[0132] like Figure 4 As further illustrated, process 400 may include receiving information identifying the results of a set of spectral measurements (block 420). For example, control device 210 may (e.g., using processor 320, communication interface 370, etc.) receive information identifying the results of the set of spectral measurements. In some embodiments, control device 210 may receive information identifying a set of spectra corresponding to samples in the training set and / or validation set. For example, control device 210 may receive information identifying a specific spectrum observed when spectrometer 220 performs spectroscopy on the training set. In some embodiments, control device 210 may simultaneously receive information identifying the spectra of both the training set and the validation set. In some embodiments, control device 210 may receive information identifying the spectra of the training set, which may be used to generate a classification model, and may receive information identifying the spectra of the validation set after generating the classification model, enabling testing of the classification model. In some embodiments, control device 210 may receive additional information as a result of a set of spectral measurements, such as information indicating that measurements were performed inaccurately to generate mismatched categories. Alternatively or additionally, control device 210 may receive information associated with identifying energy absorption, energy radiation, energy scattering, etc.

[0133] In some implementations, the control device 210 may receive information from multiple spectrometers 220 identifying the results of a set of spectral measurements. For example, the control device 210 may control physical conditions (e.g., differences between the multiple spectrometers 220, potential differences in laboratory conditions, etc.) by receiving spectral measurements performed by multiple spectrometers 220, performed at multiple different times, performed at multiple different locations, etc.

[0134] In some implementations, control device 210 may remove one or more spectra from the utilization of the generated classification model. For example, control device 210 may perform spectral classification and classify spectra into mismatched categories, and may determine that the sample corresponding to that spectrum is unintentionally interfering material or material of no interest (e.g., based on human error in correctly performing spectroscopy, errors in information identifying the spectra in the training set, etc.), and may determine to remove that spectrum from the training set. In this way, control device 210 can improve the accuracy of the classification model by reducing the likelihood of generating a classification model using incorrect or inaccurate information about the training or validation set.

[0135] like Figure 4 As further illustrated, process 400 may include generating a classification model based on information identifying the results of a set of spectral measurements (box 430). For example, control device 210 may generate a global classification model (e.g., for in-situ local modeling techniques) associated with principal component analysis (PCA)-SVM classifier techniques based on information identifying the results of a set of spectral measurements (e.g., using processor 320, memory 330, storage component 340, etc.).

[0136] In some embodiments, control device 210 may perform a set of measurements to generate a global classification model. For example, control device 210 may generate a set of categories for the global classification model and may assign a set of spectra identified by a set of spectroscopic measurements to local categories based on the use of SVM technology. In some embodiments, during the use of the global classification model, control device 210 uses a confidence metric associated with the global classification model to identify a threshold amount corresponding to the local category of the unknown spectrum, generates a local classification model based on the local category, and determines the identity of the unknown spectrum based on the local classification model. In this case, mismatched categories may be generated for the local classification model (e.g., the local classification model generated in situ from the global classification model may include mismatched categories). In this way, by using in-situ local modeling with a first category and a second category, control device 210 is able to classify a large number of categories (e.g., greater than a threshold, such as greater than 50 categories, greater than 100 categories, greater than 200 categories, greater than 1000 categories, etc.). In some embodiments, control device 210 may generate another type of classification model for classifying unknown spectra and / or use another type of classifier for the classification model.

[0137] SVM can refer to a supervised learning model that performs pattern recognition and uses confidence metrics for classification. In some implementations, when using SVM techniques to generate a global classification model, control device 210 can utilize specific types of kernel functions to determine the similarity of two or more inputs (e.g., spectra). For example, control device 210 can utilize kernel functions of the radial basis function (RBF) type (e.g., referred to as SVM-rbf), which can be represented as k(x,y) = exp(-||xy||^2) for spectra x and y; or kernel functions of the linear function type (e.g., referred to as SVM-linear and hierarchical SVM-linear when used for multi-stage determination techniques), which can be represented as k(x,y) =<x·y> Kernel functions of type sigmoid function; kernel functions of type polynomial function; kernel functions of type exponential function; and / or similar functions.

[0138] In some implementations, the control device 210 may utilize specific types of confidence metrics for the SVM, such as probability-based SVMs (e.g., a determination based on the probability that a sample is a member of a class in a set of classes), decision-based SVMs (e.g., a determination using a decision function to vote for a class in a set of classes as the class to which the sample is a member), etc. For example, during the use of a classification model utilizing a decision-based SVM, the control device 210 may determine whether an unknown sample lies within the boundaries of its constituent classes based on the mapping of the spectrum of the unknown sample, and may assign the sample to a class based on whether the unknown sample lies within the boundaries of its constituent classes. In this way, the control device 210 may determine whether to assign the unknown spectrum to a specific class, to a mismatched class, etc.

[0139] In some implementations, control device 210 may utilize specific category comparison techniques to determine decision values. For example, control device 210 may utilize a one-versus-all technique (sometimes called a one-all-all technique), where the classification model is divided into a set of sub-models, each sub-model based on a comparison of one category with every other category of the classification model, and the decision value is determined based on each sub-model. Alternatively, control device 210 may utilize an all-pair technique, where the classification model is divided into every possible pair of categories to form sub-models, and the decision value is determined from the sub-models.

[0140] Although some of the implementations described herein are presented in the manner of a specific set of machine learning techniques, other techniques may also be used to determine information about unknown spectra, such as the classification of materials, etc.

[0141] In some implementations, the control device 210 may select a specific classifier from a set of classification techniques to be used to generate a global classification model. For example, the control device 210 may generate multiple classification models corresponding to multiple classifiers and may test the multiple classification models, for example by determining the transferability of each model (e.g., how accurate a classification model generated based on spectral measurements performed on the first spectrometer 220 is when applied to spectral measurements performed on the second spectrometer 220), the accuracy of large-scale measurements (e.g., the accuracy with which the classification model can be used to classify a certain number of samples that simultaneously meet a threshold), and so on. In this case, the control device 210 may select a classifier, such as an SVM classifier (e.g., hierarchical-SVM-linear), based on determining that the classifier is associated with superior transferability and / or accuracy of large-scale measurements relative to other classifiers.

[0142] In some embodiments, control device 210 may generate a classification model based on information identifying samples in a training set. For example, control device 210 may use information identifying the type or concentration of a material represented by samples in the training set to identify the category of a spectrum having that material type or concentration. In some embodiments, control device 210 may train the classification model while generating it. For example, control device 210 may train the model using a portion of a set of spectral measurements (e.g., measurements associated with the training set). Alternatively or additionally, control device 210 may perform an evaluation of the classification model. For example, control device 210 may validate the classification model (e.g., for predicted intensity) using another portion of that set of spectral measurements (e.g., a validation set).

[0143] In some implementations, control device 210 may use multi-stage determination techniques to validate the classification model. For example, for classification based on in-situ local modeling, control device 210 may determine that the global classification model is accurate when used in association with one or more local classification models. In this way, control device 210 ensures that the classification model is generated with threshold accuracy before being provided for use, for example, by control device 210, by other control devices 210 associated with other spectrometers 220.

[0144] In some implementations, control device 210 may provide the classification model to other control devices 210 associated with other spectrometers 220 after generating the classification model. For example, a first control device 210 may generate the classification model and provide it to a second control device 210 for use. In this case, for classification based on in-situ local modeling, the second control device 210 may store the classification model (e.g., a global classification model) and may use the classification model to generate one or more in-situ local classification models for classifying one or more samples in an unknown set. Alternatively, control device 210 may store the classification model for use by control device 210 when performing classification, generating one or more local classification models (e.g., for classification based on in-situ local modeling), and / or similar situations. In this way, control device 210 provides the classification model for use in the spectral classification of unknown samples.

[0145] Although Figure 4 An example block of process 400 is shown, but in some implementations, it is different from... Figure 4 Compared to the boxes depicted, process 400 may include additional boxes, fewer boxes, different boxes, or boxes arranged differently. Alternatively, two or more boxes of process 400 may be executed in parallel.

[0146] Figure 5 Is with Figure 4The example process 400 shown is associated with the example implementation 500. Figure 5 An example of generating a classification model with false positive identification for quantization is shown.

[0147] like Figure 5 As shown, control device 210-1 transmits information to spectrometer 220-1 to instruct spectrometer 220-1 to perform a set of spectral measurements on training set and validation set 510. Assume that training set and validation set 510 include a first set of training samples (e.g., measurements used to train a classification model) and a second set of validation samples (e.g., measurements used to verify the accuracy of the classification model). As shown by reference numeral 515, spectrometer 220-1 performs a set of spectral measurements based on received instructions. As shown by reference numeral 520, control device 210-1 receives the first set of spectra for the training samples and the second set of spectra for the validation samples. In this case, validation samples may include samples of multiple materials of interest for classification and one or more samples of mismatched categories used to train the classification model to avoid false positives or incorrect measurements of one or more interfering materials. Assume that control device 210-1 stores information identifying each sample in training set and validation set 510.

[0148] about Figure 5 Suppose that control device 210-1 has chosen to use a hierarchical SVM-linear classifier to generate a classification model (e.g., based on testing the hierarchical SVM-linear classifier against one or more other classifiers). This classification model can be an in-situ local modeling type. As illustrated by reference numeral 525, control device 210-1 trains the classification model using a hierarchical SVM-linear classifier and a first set of spectra, and validates the classification model using a hierarchical SVM-linear classifier and a second set of spectra. Control device 210-1 can use a subset of the first set of spectra to train the classification model to identify interfering materials, and use a subset of the second set of spectra to validate the accuracy of the classification model in identifying interfering materials, thereby generating mismatch categories for the classification model.

[0149] Assume that control device 210-1 determines that the classification model meets a validation threshold (e.g., has an accuracy exceeding the validation threshold). As shown by reference numeral 530, control device 210-1 provides the classification model to control device 210-2 (e.g., for use when performing classification on spectral measurements performed by spectrometer 220-2) and to control device 210-3 (e.g., for use when performing classification on spectral measurements performed by spectrometer 220-3).

[0150] As indicated above, Figure 5 This is provided as an example only. Other examples are possible and may differ from those provided. Figure 5 Example of the description.

[0151] In this way, control device 210 facilitates the generation and distribution of classification models based on selected classification techniques (e.g., techniques selected based on model transferability, accuracy of large-scale classification, etc.) for use by one or more other control devices 210 associated with one or more spectrometers 220. Additionally, control device 210 improves the accuracy of the classification model by including spectral measurements of interfering materials to avoid false alarms.

[0152] Figure 6 This is a flowchart of an example process 600 for avoiding false alarms during raw material identification. In some embodiments, Figure 6 One or more process frames may be executed by control device 210. In some embodiments, Figure 6 One or more process frames may be performed by another device or a group of devices (such as spectrometer 220) that are separate from or include control device 210.

[0153] like Figure 6 As shown, process 600 may include receiving information identifying the results of a set of spectral measurements performed on an unknown sample (block 610). For example, control device 210 may (e.g., using processor 320, communication interface 370, etc.) receive information identifying the results of a set of spectral measurements performed on an unknown sample. In some embodiments, control device 210 may receive information identifying the results of a set of spectral measurements on an unknown set (e.g., multiple samples). The unknown set may include a set of samples (e.g., unknown samples) to which a determination (e.g., spectral classification) is to be performed. For example, control device 210 may cause spectrometer 220 to perform a set of spectral measurements on the set of unknown samples and may receive information identifying a set of spectra corresponding to the set of unknown samples.

[0154] In some implementations, the control device 210 may receive information about identification results from multiple spectrometers 220. For example, the control device 210 may cause multiple spectrometers 220 to perform a set of spectral measurements on an unknown set (e.g., the same sample set), and may receive information about a set of spectra corresponding to samples in the unknown set. Alternatively, the control device 210 may receive information about the results of a set of spectral measurements performed at multiple times, locations, etc., and may classify and / or quantify a particular sample based on the set of spectral measurements performed at multiple times, locations, etc. (e.g., based on averaging the set of spectral measurements or based on another technique). In this way, the control device 210 can respond to physical conditions that may affect the results of the set of spectral measurements.

[0155] Alternatively, the control device 210 may cause the first spectrometer 220 to perform a first portion of a set of spectral measurements on a first portion of the unknown set, and may cause the second spectrometer 220 to perform a second portion of the set of spectral measurements on a second portion of the unknown set. In this way, the control device 210 can reduce the time required to perform a set of spectral measurements compared to having all spectral measurements performed by a single spectrometer 220.

[0156] like Figure 6 As further illustrated, process 600 may include determining whether a set of spectral measurements were performed accurately (block 620). For example, control device 210 may (e.g., using processor 320, memory 330, storage component 340, etc.) determine whether the set of spectral measurements was performed accurately. In some embodiments, control device 210 may determine whether spectral measurements of the unknown sample were performed at a calibrated distance (e.g., between the optical components of spectrometer 220 and the sample, between the optical components of spectrometer 220 and the background of the sample, etc.). Alternatively or additionally, control device 210 may determine whether spectral measurements of the unknown sample were performed at a calibrated temperature, at a calibrated pressure, at a calibrated humidity, using a calibrated background, using a calibrated spectrometer, and / or under similar conditions.

[0157] The calibrated values ​​for calibration conditions (such as calibration distance, calibration temperature, calibration pressure, calibration humidity, calibration background, etc.) may include values ​​where the model was trained and / or validated. For example, control device 210 may receive measurement data from spectrometer 220 that identifies values ​​for measurement conditions (such as temperature, unknown sample, and distance between optical components of spectrometer 220), and control device 210 may verify that the model was trained using a training set and / or validation set associated with calibrated values ​​for calibration conditions within a threshold value range.

[0158] Alternatively, control device 210 may use a single-class SVM (SC-SVM) classifier technique to perform a sanity check to determine whether the unknown spectrum is associated with a correctly performed measurement. For example, control device 210 may aggregate multiple classes in a classification model to form an aggregated classification model with a single class and use an SVM classifier with decision values ​​to determine whether the unknown sample is an outlier. In this case, when the unknown sample is an outlier, control device 210 can determine that the set of spectral measurements was not performed accurately and can allow the set of spectral measurements to be performed again, and can receive another set of results identifying the other set of spectral measurements (box 620 - No). In this way, control device 210 is able to identify unknown spectra that differ from the classification model by a threshold amount without using samples similar to the unknown sample (e.g., also differing from the training set samples of the material of interest by a threshold amount) to train the classification model. In addition, the control device 210 reduces the number of samples to be collected for generating the classification model, thereby reducing costs, time, and utilization of computing resources (e.g., processing and memory resources) compared to acquiring, storing, and processing other samples of interfering materials that differ from the material of interest by a threshold amount.

[0159] Furthermore, compared to performing spectroscopy under uncertain conditions where the measurement conditions do not match the calibration conditions, control device 210 reduces the possibility of inaccurate results from spectroscopy (e.g., inaccurate quantification, inaccurate measurement, etc.). Additionally, based on ensuring that the measurement of the unknown sample is performed correctly before attempting to classify the unknown sample, control device 210 reduces the utilization of computational resources compared to attempting to perform spectroscopy that fails due to incorrect measurement or another attempt to perform spectroscopy.

[0160] like Figure 6As further illustrated, based on the determination that a set of spectral measurements were accurately performed (box 620 - Yes), process 600 may include determining whether an unknown sample is included in a mismatch category based on the results of that set of spectral measurements (box 630). For example, control device 210 may attempt (e.g., using processor 320, memory 330, storage component 340, etc.) to determine whether an unknown sample will be classified into a mismatch category (e.g., uninteresting material or interfering material). In some embodiments, control device 210 may classify the unknown sample to determine whether it is included in a mismatch category. For example, control device 210 may use an SVM-rbf kernel function or an SVM-linear kernel function on the model to determine a decision value for classifying the unknown sample into a mismatch category. Based on the decision value satisfying a threshold decision value, control device 210 may determine that the unknown sample belongs to a mismatch category (e.g., the unknown sample is determined to be interfering material, the spectrum is determined to be associated with baseline spectral measurements, such as measurements performed with incorrect measurement distances, measurements performed with incorrect measurement backgrounds, measurements performed with incorrect measurement illumination, measurements performed in the absence of a sample, etc.). In this way, control device 210 determines that the classification model used for the spectroscopic method has not been calibrated for the spectrum of a particular unknown sample, and avoids false positives for that particular unknown sample. Alternatively, control device 210 can determine that the unknown sample does not belong to a mismatch category.

[0161] like Figure 6 As further illustrated, based on determining that an unknown sample is included in a mismatch category (box 630 - Yes), process 600 may include providing output indicating that the unknown sample is included in a mismatch category (box 640). For example, control device 210 may provide information indicating that an unknown sample is included in a mismatch category, for example, via a user interface (e.g., using processor 320, memory 330, storage component 340, communication interface 370, etc.). In some embodiments, control device 210 may provide information associated with identifying an unknown sample. For example, based on an attempt to quantify the amount of a specific chemical in a specific plant and determining that the unknown sample is not that specific plant (but another plant, e.g., based on human error), control device 210 may provide information to identify the other plant. In some embodiments, control device 210 may obtain another classification model and may use the other classification model to identify the unknown sample based on assigning the unknown spectrum to a mismatch category in the classification model.

[0162] In this way, the control device 210 reduces the possibility of providing incorrect information based on false alarms about unknown samples, and enables technicians to perform error correction by providing information to help determine that the unknown sample is another plant rather than a specific plant.

[0163] like Figure 6 As further illustrated, based on the determination that the unknown sample is not included in the mismatch category (box 630 - No), process 600 may include performing one or more spectroscopic measurements based on the results of a set of spectroscopic measurements (box 650). For example, control device 210 may perform one or more spectroscopic measurements based on the results of a set of spectroscopic measurements (e.g., using processor 320, memory 330, storage component 340, etc.). In some embodiments, control device 210 may assign the unknown sample to a specific category within a set of categories in a global classification model to perform the first measurement. For example, control device 210 may determine, based on a global classification model, that a specific spectrum associated with a particular sample corresponds to a local category of material (e.g., cellulose material, lactose material, caffeine material, etc.).

[0164] In some implementations, control device 210 may assign specific samples based on a confidence metric. For example, control device 210 may determine the probability that a specific spectrum is associated with each category of the global classification model based on a global classification model. In this case, control device 210 may assign an unknown sample to a specific local category based on the fact that a specific probability for a specific local category exceeds other probabilities associated with other non-local categories. In this way, control device 210 determines the type of material associated with the sample, thereby identifying the sample. In some implementations, control device 210 may determine that an unknown sample does not meet a threshold for association with any category and does not meet a threshold for association with a mismatched category. In this case, control device 210 may provide an output indicating that the unknown sample is not included in any category and cannot be assigned to a mismatched category with a confidence level corresponding to the threshold associated with the mismatched category.

[0165] In some embodiments, to perform in-situ local modeling, such as for a classification model of categories with a number greater than a threshold, control device 210 may generate a local classification model based on a first determination. The local classification model may refer to an in-situ classification model generated based on a confidence metric associated with the first determination using SVM determination techniques (e.g., kernel functions of SVM-rbf, SVM-linear, etc.; probability-based SVM, decision-based SVM, etc.; and / or similar techniques). In some embodiments, control device 210 may generate multiple local classification models.

[0166] In some implementations, control device 210 may generate a local quantization model based on performing a first determination using a global classification model. For example, when control device 210 is used to determine the concentration of a substance in an unknown sample, and multiple unknown samples are associated with different quantization models used to determine the concentration of the substance, control device 210 may utilize the first determination to select a subset of categories as local categories for the unknown samples, and may select a quantization model for the unknown samples based on the results of the first determination. In this way, control device 210 utilizes hierarchical determination and quantization models to improve spectral classification.

[0167] In some implementations, control device 210 may perform a second determination based on the results and a local classification model. For example, control device 210 may classify an unknown sample as one of the materials of interest in a global classification model based on a local classification model and a specific spectrum. In some implementations, control device 210 may determine a set of confidence metrics associated with a specific spectrum and a local classification model. For example, control device 210 may determine the probability that a specific spectrum is associated with each category of the local classification model and may assign a specific spectrum (e.g., an unknown sample associated with a specific spectrum) to a category with a higher probability than other categories of the local classification model. In this way, control device 210 identifies unknown samples. In some implementations, control device 210 may determine mismatched categories of the local classification model and may assign a specific spectrum to a mismatched category of the local classification model. In some implementations, control device 210 may determine a threshold confidence metric for an unknown sample that fails to meet the category of the classification model and may determine that the classification of the unknown sample has failed. In this way, based on the use of threshold confidence metrics, control device 210 reduces the likelihood of false positives for unknown samples.

[0168] In some embodiments, control device 210 may perform quantization after performing a first measurement (and / or after performing a second measurement). For example, control device 210 may select a local quantization model based on performing one or more measurements, and may perform quantization associated with a specific sample based on the selection of a local quantization model. As an example, when performing raw material identification to determine the concentration of a specific chemical in plant material, where the plant material is associated with multiple quantization models (e.g., related to whether the plant is indoors or outdoors, whether it is grown in winter or summer, etc.), control device 210 may perform a set of measurements to identify a specific quantization model. In this case, control device 210 may determine that the plant was grown indoors in winter based on performing a set of measurements, and may select a quantization model associated with plants grown indoors in winter for determining the concentration of a specific chemical.

[0169] like Figure 6As further illustrated, based on classification failures when performing one or more spectral classifications (box 650-A), process 600 may include providing an output indicating classification failure and selectively updating the categories of the classification model (box 660). For example, control device 210 may (e.g., using processor 320, memory 330, storage component 340, communication interface 370, etc.) provide information indicating classification failure. For example, based on determining that the confidence level associated with the classification does not meet a threshold confidence level, control device 210 may provide an output indicating classification failure, thereby reducing the likelihood of false alarms. Alternatively or additionally, based on determining that the confidence level does not meet the threshold, control device 210 may selectively update the categories of the classification model for performing classification. For example, control device 210 may obtain (e.g., from an operator, database, etc.) additional information about the identified sample and may determine that the sample belongs to a labeled category. In this case, control device 210 may update the labeled category to achieve improved subsequent spectral classification. Alternatively or additionally, control device 210 may obtain information indicating that the sample does not belong to a labeled category. In this case, control device 210 can update the mismatch category to achieve improved subsequent mismatch classification. In this way, control device 210 enables iterative model enhancement for spectral classification.

[0170] like Figure 6 As further illustrated, based on the success of classification when performing one or more spectral classifications (box 650-B), process 600 may include providing information identifying the classification associated with the unknown sample (box 670). For example, control device 210 may (e.g., using processor 320, memory 330, storage component 340, communication interface 370, etc.) provide information identifying the classification associated with the unknown sample. In some embodiments, control device 210 may provide information identifying a specific category for the unknown sample. For example, control device 210 may provide information indicating that a specific spectrum associated with the unknown sample is determined to be associated with a specific category, thereby identifying the unknown sample.

[0171] In some implementations, the control device 210 may provide information indicating a confidence metric associated with assigning an unknown sample to a particular category. For example, the control device 210 may provide information identifying the probability of an unknown sample being associated with a particular category, etc. In this way, the control device 210 provides information indicating the likelihood that a particular spectrum will be accurately assigned to a particular category.

[0172] In some implementations, control device 210 may provide quantization based on performing a set of classifications. For example, based on a local quantization model that identifies the category associated with the unknown sample, control device 210 may provide information identifying the concentration of a substance in the unknown sample. In some implementations, control device 210 may update a classification model based on performing a set of classifications. For example, control device 210 may generate a new classification model that includes the unknown sample as a training set of samples, based on determining the unknown sample as a material of interest, a distracting material, and / or a similar material.

[0173] Although Figure 6 An example block of process 600 is shown, but in some implementations, it differs from... Figure 6 Compared to the boxes depicted, process 600 may include additional boxes, fewer boxes, different boxes, or boxes arranged differently. Alternatively, two or more boxes of process 600 may be executed in parallel.

[0174] Figure 7A and Figure 7B Is with Figure 6 The example process 600 shown is associated with an example implementation 700 related to the prediction success rate. Figure 7A and Figure 7B Example results of material identification using a technique based on hierarchical support vector machines (hierarchical-SVM linear) are shown.

[0175] like Figure 7A As shown by reference numeral 705, control device 210 can cause spectrometer 220 to perform a set of spectral measurements. For example, control device 210 can provide instructions to cause spectrometer 220 to acquire the spectrum of an unknown sample, thereby determining the classification of the unknown sample as a particular material of interest in a set of materials of interest trained as a classification model. As shown by reference numerals 710 and 715, spectrometer 220 can receive an unknown sample and can perform a set of spectral measurements on the unknown sample. As shown by reference numeral 720, control device 210 can receive the spectrum of the unknown sample based on the set of spectral measurements performed on the unknown sample by spectrometer 220.

[0176] like Figure 7B As shown, control device 210 can use classification model 725 to perform spectral classification. Classification model 725 includes a set of categories 730 for a set of spectral identifications for a training set. For example, classification model 725 includes categories 730-1 to 730-6 of potentially interesting materials and mismatch categories 730-7 of interfering materials (e.g., similar materials; similar spectra; incorrectly obtained spectra, such as incorrect illumination spectra, incorrect distance spectra, incorrect background spectra, etc.; and / or similar substances).

[0177] like Figure 7B As further illustrated by reference numerals 735 and 740, the spectrum of an unknown sample is assigned to a mismatch category, and the unknown sample is identified as distractor material (e.g., a member of the mismatch category). For example, control device 210 may use in-situ local modeling techniques to generate a local model based on a global model (e.g., classification model 725), and may determine whether an unknown sample is distractor material based on the local model. In some embodiments, control device 210 may perform in-situ thresholding techniques to determine whether an unknown sample is distractor material. For example, client device 750 may self-validate or cross-validate decision values ​​associated with the first most probable category and / or the runner-up category (e.g., the second most probable category) of the unknown sample, and may use these decision values ​​to set upper and lower bounds for the prediction threshold. In some embodiments, client device 750 may utilize multiple local modeling strategies. For example, client device 750 may utilize a first modeling technique to determine the winner category and a second modeling technique to determine a confidence metric. In some embodiments, client device 750 may utilize a one-class SVM (SC-SVM) technique to determine whether an unknown sample is distractor material. As shown by reference numeral 745, the control device 210 provides the client device 750 with an output indicating that the unknown sample is interfering material, rather than providing the unknown sample as a false alarm identification of a specific concentration of material of interest in the material of interest.

[0178] As indicated above, Figure 7A and Figure 7B This is provided as an example only. Other examples are possible and may differ from those provided. Figure 7A and Figure 7B Example of the description.

[0179] In this way, the control device 210 reduces the likelihood of providing inaccurate results from the spectroscopic method by avoiding false positives of unknown samples being trained as classification models to identify specific materials of interest.

[0180] The foregoing disclosure provides illustrations and descriptions, but is not intended to be exhaustive or to limit the embodiments to the precise forms disclosed. Modifications and variations are possible or may be obtained from practice of the embodiments based on the above disclosure.

[0181] This article describes some implementation methods in conjunction with thresholds. As used herein, satisfying a threshold can refer to a value greater than the threshold, more than the threshold, higher than the threshold, greater than or equal to the threshold, less than the threshold, less than the threshold, lower than the threshold, less than or equal to the threshold, equal to the threshold, etc.

[0182] It will be apparent that the systems and / or methods described herein may be implemented in various forms, including hardware, firmware, or a combination of firmware and software. The actual dedicated control hardware or software code used to implement these systems and / or methods is not a limitation of the implementation. Therefore, while the operation and behavior of the systems and / or methods are described herein without reference to specific software code, it should be understood that software and hardware can be designed to implement the systems and / or methods based on the descriptions herein.

[0183] Although specific combinations of features are stated in the claims and / or disclosed in the specification, these combinations are not intended to limit the disclosure of possible embodiments. In fact, many of these features can be combined in ways not specifically stated in the claims and / or not disclosed in the specification. Although each appended dependent claim may be directly subordinated to only one claim, the disclosure of possible embodiments includes each dependent claim in combination with every other claim in the claim set.

[0184] No element, action, or instruction used herein should be construed as critical or necessary unless explicitly stated otherwise. Furthermore, as used herein, the articles “a” and “an” are intended to include one or more items and may be used interchangeably with “one or more.” Additionally, as used herein, the term “set” is intended to include one or more items (e.g., related items, unrelated items, a combination of related and unrelated items, etc.) and may be used interchangeably with “one or more.” The term “one” or similar language is used when referring to only one item. Furthermore, as used herein, the terms “has,” “have,” “having,” and / or similar words are intended to be open-ended terms. Moreover, unless explicitly stated otherwise, the phrase “based on” is intended to mean “at least partially based on.”

Claims

1. A method comprising: A device comprising one or more processors determines that an unknown sample is an outlier by using an aggregation classification model. Based on the determination that the unknown sample is the outlier sample, the device determines that one or more spectral measurements were not performed accurately; as well as The device may trigger one or more actions based on the determination that one or more spectral measurements were not performed accurately.

2. The method of claim 1, wherein causing the one or more actions comprises: Based on the determination that one or more spectral measurements were not performed accurately, another set of results is received that identifies another set of spectral measurements.

3. The method according to claim 1, further comprising: Based on the actions that cause the one or more actions, an unknown spectrum that differs from the clustering classification model by a threshold amount is identified.

4. The method according to claim 1, further comprising: To ensure that one or more different spectral measurements are performed accurately; as well as Based on the determination that the one or more different spectral measurements were performed accurately, the unknown sample is classified into mismatch categories in the clustering classification model to determine whether it is included in the mismatch category.

5. The method according to claim 4, further comprising: Provides output indicating whether the unknown sample is included in the mismatch category.

6. The method according to claim 1, further comprising: When performing in-situ thresholding, it is determined whether the unknown sample is a disturbance or abnormal material.

7. The method according to claim 1, further comprising: Cross-validation of decision values ​​associated with the first category and / or the second category of the unknown sample; as well as Use the decision value to set the upper and lower limits for the prediction threshold.

8. An apparatus comprising: One or more memory units; as well as One or more processors, coupled to the one or more memories, are configured to: By using an aggregation classification model, it was determined that the unknown sample was an outlier. Based on the determination that the unknown sample is the outlier sample, it is determined that one or more spectral measurements were not performed accurately. as well as Based on the determination that one or more spectral measurements were not performed accurately, one or more actions are triggered.

9. The device of claim 8, wherein the one or more processors are configured to cause the one or more actions as follows: Based on the determination that one or more spectral measurements were not performed accurately, another set of results is received that identifies another set of spectral measurements.

10. The device of claim 8, wherein the one or more processors are further configured to: Based on the actions that cause the one or more actions, an unknown spectrum that differs from the clustering classification model by a threshold amount is identified.

11. The device of claim 8, wherein the one or more processors are further configured to: It was determined that one or more different spectral measurements were performed accurately; and Based on the determination that the one or more different spectral measurements were performed accurately, the unknown sample is classified into mismatch categories in the clustering classification model to determine whether it is included in the mismatch category.

12. The device of claim 11, wherein the one or more processors are further configured to: Provides output indicating whether the unknown sample is included in the mismatch category.

13. The device of claim 8, wherein the one or more processors are further configured to: When performing in-situ thresholding, it is determined whether the unknown sample is a disturbance or abnormal material.

14. The device of claim 8, wherein the one or more processors are further configured to: Cross-validation of decision values ​​associated with the first category and / or the second category of the unknown sample; and Use the decision value to set the upper and lower limits for the prediction threshold.

15. A non-transitory computer-readable medium storing an instruction set, the instruction set comprising: One or more instructions, when executed by one or more processors of the device, cause the device to: By using an aggregation classification model, it was determined that the unknown sample was an outlier. Based on the determination that the unknown sample is the outlier sample, it is determined that one or more spectral measurements were not performed accurately. as well as Based on the determination that one or more spectral measurements were not performed accurately, one or more actions are triggered.

16. The non-transitory computer-readable medium of claim 15, wherein the one or more instructions causing the device to cause the one or more actions cause the device to: Based on the determination that one or more spectral measurements were not performed accurately, another set of results is received that identifies another set of spectral measurements.

17. The non-transitory computer-readable medium of claim 15, wherein one or more instructions further cause the device to: Based on the actions that cause the one or more actions, an unknown spectrum that differs from the clustering classification model by a threshold amount is identified.

18. The non-transitory computer-readable medium of claim 15, wherein one or more instructions further cause the device to: It was determined that one or more different spectral measurements were performed accurately; and Based on the determination that the one or more different spectral measurements were performed accurately, the unknown sample is classified into mismatch categories in the clustering classification model to determine whether it is included in the mismatch category.

19. The non-transitory computer-readable medium of claim 18, wherein one or more instructions further cause the device to: Provides output indicating whether the unknown sample is included in the mismatch category.

20. The non-transitory computer-readable medium of claim 15, wherein one or more instructions further cause the device to: When performing in-situ thresholding, it is determined whether the unknown sample is a disturbance or abnormal material.

Images