Method for automatically identifying rock minerals based on spectral information

Abstract

The invention discloses a method for automatically identifying rock minerals based on spectral information, and particularly relates to the field of rock mineral identification, which comprises the following steps: acquiring spectral information of mixed minerals, screening mineral types matched with a static spectral feature library, and establishing feature-mineral type association. Similar mineral types are retrieved, a set number of samples are obtained, and the samples are put according to the collection coordinates; carrying out time sequence reflection spectrum data acquisition on the similar mineral samples, extracting spectral features and carrying out timestamp alignment; performing three-dimensional structuring on the spectral features, extracting dynamic spectral feature elements by adopting a sliding window, and constructing a three-dimensional matrix; distortion identification is carried out on the time window spectrum characteristic elements of the sunshine cycle, and distortion fragments are eliminated; establishing a feature library according to the effective dynamic spectrum feature element proportion, and generating a ready signal to trigger model training; and training a rock mineral identification model based on the convolutional neural network, and carrying out whole-region accurate scanning and automatic identification on the trained model.

Description

Technical Field

[0001] The present invention relates to the technical field of rock and mineral identification, and more specifically, to a method for automatically identifying rock and mineral based on spectral information. Background Art

[0002] In the task of rock and mineral identification, the spectral characteristics of minerals often have the problem of static spectral feature similarity, especially when multiple minerals have similar spectral features, which makes the traditional classification method based on static spectra face great challenges. The spectral signals of minerals may be highly overlapping in certain bands, and even different minerals may have similar spectral features. In addition, the slight differences between different minerals may be masked by data noise or instrument errors, causing errors in the identification model during classification. Considering that minerals are usually distributed in one area, due to the similarity of their environment, the periodic characteristics of their dynamic spectral characteristics with a daily cycle have not been fully utilized in rock and mineral identification.

[0003] Therefore, how to propose an automatic rock and mineral identification method based on spatiotemporal dynamic spectral characteristics, introduce the time dimension to dynamically capture the subtle differences in mineral spectra over time, and realize the dynamic spectral characteristics of minerals in different time periods within the identification cycle, thereby improving the identification and classification accuracy of similar minerals with similar static spectral characteristics, is an urgent problem to be solved.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method for automatically identifying rock minerals based on spectral information to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions: S1: Collect mixed minerals for spectral information detection, screen the mineral types that hit the spectral features in the static spectral feature library, and establish the corresponding correlation relationship between features and mineral types; S2: Retrieve similar mineral species whose corresponding relationships are not unique in the feature-mineral species corresponding relationship, obtain a set number of similar mineral samples, and place the similar mineral samples according to the distribution of the mixed mineral collection coordinates; S3: collect time series reflectance spectrum data of similar mineral samples and align the time series driven by timestamp to extract the spectral features of the time series reflectance spectrum data of similar mineral samples; S4: Perform three-dimensional structural processing on the time series spectral features of similar mineral samples, extract dynamic spectral feature primitives using the sliding window method, and establish a three-dimensional matrix of dynamic spectral feature primitives; S5: Distortion identification is performed on the spectral characteristic primitives corresponding to the time window of the same previous sunshine cycle, and the corresponding spectral distortion fragments are eliminated; S6: Establish a dynamic spectral feature library, and generate a ready signal to trigger the rock and mineral identification model training based on the ratio of effective dynamic spectral feature primitive data; S7: Using the three-dimensional matrix of dynamic spectral feature primitives in the dynamic spectral feature library as the training data set, a rock and mineral recognition model based on convolutional neural network is established and trained; S8: Deploy the trained rock and mineral identification model to the mobile spectral detection device to accurately scan and automatically identify the entire rock and mineral detection area.

[0007] In a preferred embodiment, in S1, mixed minerals are collected for spectral information detection, mineral types that hit spectral features in the static spectral feature library are screened, and the corresponding association relationship between features and mineral types is established, which specifically includes: Collect mixed minerals with uniform coordinate distribution in the rock and mineral detection area; Retrieve the open-source mineral static spectral feature library, and based on the detection condition constraints required by the static spectral feature library, perform spectral information detection on the individual minerals of the mixed minerals under set experimental conditions to obtain a spectral identification data set of the mixed minerals; Extract static spectral features of mixed mineral spectral identification data set, and compare the static spectral features of mixed minerals with the features in the static spectral feature library for feature similarity, wherein the spectral features include wavelength, absorptivity and absorption peak; A feature similarity threshold is set, and static spectral features whose feature similarity exceeds the set feature similarity threshold are marked as hit features. The possible mineral types corresponding to the hit features in the mineral static spectral feature library are obtained, and a corresponding association relationship between features and mineral types is established.

[0008] In a preferred embodiment, in S2, similar mineral species whose corresponding relationships are not unique in the feature-mineral species corresponding association relationship are retrieved, a set number of similar mineral samples are obtained, and the similar mineral samples are sampled according to the mixed mineral collection coordinate distribution.

[0009] In a preferred embodiment, in S3, the time series reflectance spectrum data of similar mineral samples are collected and the time series is aligned driven by timestamps, and the spectral features of the time series reflectance spectrum data of similar mineral samples are extracted, which specifically includes: Configure a full-sunshine-cycle synchronous spectrum acquisition device in the rock and mineral detection area, set the spectrum acquisition time resolution and the band range of the incident spectrum, and collect reflectance spectrum data in time series for samples with similar mineral sample placement coordinates; Perform time series alignment of all time series reflectance spectral data of similar mineral samples based on spectral acquisition timestamp-driven time series alignment; Extract spectral features of time series reflectance spectral data of similar mineral samples.

[0010] In a preferred embodiment, in S4, the spectral features of the time series of similar mineral samples are subjected to three-dimensional structural processing, and the dynamic spectral feature primitives are extracted by a sliding window method. The establishment of a three-dimensional matrix of dynamic spectral feature primitives specifically includes: The time series spectral features of similar mineral samples are constructed into a three-dimensional tensor according to the three dimensions of acquisition timestamp, spatial coordinates and spectral feature items. Each spatial coordinate corresponds to an independent time series spectral feature sequence, forming a structured storage with dimensions of time, space and spectral feature items. The unit time interval of the spectral acquisition time resolution is set as the window sliding step, the window size is 4 steps, and the sliding window method is used to calculate the first-order derivative of the spectral feature term in the time dimension, extract the dynamic spectral feature primitives, and establish a three-dimensional matrix of dynamic spectral feature primitives.

[0011] In a preferred embodiment, in S5, the distortion identification is performed on the spectral characteristic primitives corresponding to the time window of the same sunshine cycle as the previous sunshine cycle, and the elimination of the corresponding spectral distortion fragments specifically includes: The dynamic spectrum characteristic primitive three-dimensional matrix of the historical full sunshine cycle is obtained, and the dynamic spectrum characteristic primitive is processed with characteristic mean and decomposed in two dimensions based on spatial coordinates. According to the characteristic cosine similarity of the spectral characteristic primitives corresponding to the same time window of the historical sunshine cycle and the current sunshine cycle under the same spatial coordinates, a time-varying anomaly detection model based on characteristic primitive distortion evaluation is constructed. The expression is: ; In the formula, , are the spectral characteristic primitives corresponding to the i-th time window of the historical sunshine cycle and the current sunshine cycle, is the spectral characteristic primitive distortion evaluation value of the i-th time window; The distortion threshold of the spectral feature primitive is set, and the time-varying anomaly detection model detects and selects the time window in which the distortion evaluation value of the spectral feature primitive is greater than the distortion threshold, and removes the spectral distortion fragment corresponding to the time window.

[0012] In a preferred embodiment, in S6, establishing a dynamic spectral feature library, and generating a ready signal based on the ratio of effective dynamic spectral feature primitive data to trigger rock and mineral identification model training specifically includes: Based on the remaining effective dynamic spectral feature primitives after anomaly elimination, a dynamic spectral feature library is established according to mineral categories, and the LRU strategy is used to maintain the library capacity. The number of continuous sunshine cycles covered by each mineral category in the spectral library is counted. When the target mineral category covers more than a set number of complete cycles and the proportion of valid dynamic spectral feature primitive data in each cycle exceeds a set ratio, a ready signal is generated to trigger the rock mineral identification model training.

[0013] In a preferred embodiment, in S7, using the dynamic spectral feature primitive three-dimensional matrix in the dynamic spectral feature library as a training data set, establishing and training a rock and mineral identification model based on a convolutional neural network specifically includes: Using the dynamic spectral feature primitive three-dimensional matrix in the dynamic spectral feature library as the training data set, a rock and mineral recognition model based on convolutional neural network is established, and a three-dimensional convolution kernel corresponding to the dynamic spectral feature primitive three-dimensional matrix structure is designed, covering the three dimensions of time, space and spectral characteristics; The time dimension of the three-dimensional convolution kernel uses continuous time periods for convolution processing, the spatial dimension uses a convolution kernel of size n×m, and the spectral feature dimension selects continuous dynamic spectral feature primitives. The convolution kernel is set to slide to capture the changes of mineral spectra over time. The feature map output by the convolution layer is reduced in dimension through the maximum pooling operation, and the pooled feature map is input into the fully connected layer and connected to the Softmax classifier for mineral type classification; A cross entropy loss function based on time-varying sensitivity weighting is selected to quantify the difference between the predicted results and the actual labels, and the rock and mineral recognition model is iteratively trained; The training data set is divided into multiple subsets through the cross-validation method, and the model performance is evaluated using accuracy, precision, and recall indicators. The training is terminated when all indicators of the rock and mineral identification model reach the preset indicator standards.

[0014] In a preferred embodiment, in S8, the trained rock and mineral identification model is deployed on the mobile spectrum detection device, and the rock and mineral detection area is accurately scanned and automatically identified in the entire area, specifically including: Deploy the trained rock and mineral identification model to the mobile spectral detection equipment to accurately scan the entire rock and mineral detection area; The mobile spectrum detection equipment is time-calibrated with the set full-sunshine cycle synchronous spectrum acquisition device, and the time series reflection spectrum data of all rocks and minerals in the detection area are collected by the mobile spectrum detection equipment, and the corresponding time series spectrum features are extracted; The spectral characteristics of all rock and mineral time series in the detection area are subjected to the same three-dimensional structural processing as similar mineral samples, and a three-dimensional matrix of dynamic spectral characteristic primitives for input identification is established and input into the rock and mineral identification model for automatic rock and mineral identification.

[0015] The technical effects and advantages of the method for automatically identifying rock minerals based on spectral information of the present invention are as follows: By collecting the spectral information of mixed minerals and comparing it with the data in the static spectral feature library, the preliminary screening and classification of minerals are effectively achieved. For the identification of similar mineral types, the spectral changes in different time periods are effectively captured by collecting time series spectral data and performing time alignment. At the same time, the sliding window method extracts the time series spectral features to ensure the efficient construction of dynamic spectral feature primitives, thereby improving the sensitivity and accuracy of the model to mineral changes. Through distortion identification and elimination, the training data is further optimized to avoid errors caused by environmental interference. The establishment of a dynamic spectral feature library provides high-quality training data, which provides a solid foundation for the rock and mineral identification model based on convolutional neural networks and enhances the recognition and generalization capabilities of the model. The model is deployed on a mobile spectral detection device, which can perform full-area precise scanning and automatic identification in practical applications, avoiding the identification error problem of different minerals with similar static spectral features in traditional identification methods, and greatly improving the accuracy of mineral detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of a method for automatically identifying rock minerals based on spectral information according to the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Embodiment 1:

[0018] Figure 1 The present invention provides a method for automatically identifying rock minerals based on spectral information, which comprises the following steps: S1: Collect mixed minerals for spectral information detection, screen the mineral types that hit the spectral features in the static spectral feature library, and establish the corresponding correlation relationship between features and mineral types; S2: Retrieve similar mineral species whose corresponding relationships are not unique in the feature-mineral species corresponding relationship, obtain a set number of similar mineral samples, and place the similar mineral samples according to the distribution of the mixed mineral collection coordinates; S3: collect time series reflectance spectrum data of similar mineral samples and align the time series driven by timestamp to extract the spectral features of the time series reflectance spectrum data of similar mineral samples; S4: Perform three-dimensional structural processing on the time series spectral features of similar mineral samples, extract dynamic spectral feature primitives using the sliding window method, and establish a three-dimensional matrix of dynamic spectral feature primitives; S5: Distortion identification is performed on the spectral characteristic primitives corresponding to the time window of the same previous sunshine cycle, and the corresponding spectral distortion fragments are eliminated; S6: Establish a dynamic spectral feature library, and generate a ready signal to trigger the rock and mineral identification model training based on the ratio of effective dynamic spectral feature primitive data; S7: Using the three-dimensional matrix of dynamic spectral feature primitives in the dynamic spectral feature library as the training data set, a rock and mineral recognition model based on convolutional neural network is established and trained; S8: Deploy the trained rock and mineral identification model to the mobile spectral detection device to accurately scan and automatically identify the entire rock and mineral detection area.

[0019] In S1, mixed minerals are collected for spectral information detection, mineral types that hit the spectral features in the static spectral feature library are screened, and the corresponding correlation relationship between features and mineral types is established.

[0020] Conduct detailed geological analysis in the rock mineral area and divide it into several uniform detection areas. In each area, determine appropriate sampling points based on the mineral distribution density to ensure that the mineral diversity of the entire mining area is covered. Accurately calibrate the coordinates of each sampling point through a total station or GPS positioning system to ensure that the collected samples are evenly distributed in space and avoid excessive concentration or loss of minerals in the area. Properly mix the samples collected from different locations to form a representative mixed mineral sample, while ensuring that the final sample contains as many different mineral components as possible.

[0021] Retrieve the open source mineral static spectral feature library, and based on the detection condition constraints required by the static spectral feature library, including the stability of the light source, temperature, humidity, etc., ensure the consistency of the measurement conditions, and perform spectral information detection on the individual minerals of the mixed minerals under the set experimental conditions (using a high-precision spectrometer, such as a Fourier transform infrared spectrometer, a near-infrared spectrometer, or a UV-visible spectrometer) to obtain the spectral identification data set of the mixed minerals.

[0022] The static spectral features of the mixed mineral spectral identification data set are extracted, and the static spectral features of the mixed minerals are compared with the features in the static spectral feature library for feature similarity (using Euclidean distance or cosine similarity), where the spectral features include wavelength, absorbance and absorption peak.

[0023] Based on the instrument detection accuracy and mineral identification accuracy requirements, the feature similarity threshold is set comprehensively, and the static spectral features whose feature similarity exceeds the set feature similarity threshold are marked as hit features. The possible mineral types corresponding to the hit features in the mineral static spectral feature library are obtained, and the corresponding association relationship between features and mineral types is established. For example, if a spectral feature is similar to the absorption peak of hematite in the static spectral feature library, the feature is associated with hematite.

[0024] In S2, similar mineral species whose corresponding relationships are not unique in the feature-mineral species corresponding association relationship are retrieved, a set number of similar mineral samples are obtained, and the similar mineral samples are sampled according to the mixed mineral collection coordinate distribution.

[0025] In the process of mineral identification, based on the correspondence between features and mineral types, one feature may correspond to multiple mineral types. For example, the absorption characteristics of quartz, feldspar and other silicate minerals in the near-infrared and mid-infrared bands are very similar. In addition, slight differences between different minerals may be masked by data noise or instrument errors, resulting in the inability to accurately identify each similar mineral.

[0026] The number of samples for each similar mineral is set, requiring no less than 50 samples for each mineral to ensure the robustness of the data. At the same time, mineral samples are placed in a uniform random distribution to test the adaptability of the algorithm to the mineral mixing conditions in different regions.

[0027] In S3, time series reflectance spectrum data of similar mineral samples are collected and time series alignment driven by timestamp is performed to extract spectral features of time series reflectance spectrum data of similar mineral samples.

[0028] Select a portable spectrometer with high spectral resolution, such as ASD FieldSpec 4, which can cover a wide band range from 350 nm to 2500 nm to ensure that the spectral bands of most minerals can be covered. Configure the device above the mineral sample in the rock mineral detection area, and set the spectral acquisition time resolution to 15 minutes to ensure that the spectral changes of minerals can be captured within 24 hours a day. This time resolution can balance the needs of frequent acquisition of spectral data and storage capacity. At the same time, set the band range of the incident spectrum (specifically set it in the range of 350nm to 2500nm according to the characteristic intensity of the sample mineral type in different bands), and collect time series reflectance spectral data for samples with similar mineral sample placement coordinates.

[0029] Each time spectral data is collected, the timestamp of the collection is automatically recorded. The accuracy of the timestamp is in seconds, ensuring that each data collection is highly timely. The reflectance spectral data of all samples are time-aligned using the timestamp. If there is a time deviation in the collection of some data (such as network delay or equipment problem), these data are aligned with the surrounding data through interpolation methods (such as linear interpolation or spline interpolation) to ensure the continuity of the entire time series.

[0030] Extract spectral features (including wavelength, absorbance, and absorption peak) of time series reflectance spectrum data of similar mineral samples.

[0031] In S4, the spectral characteristics of the time series of similar mineral samples are processed into three-dimensional structure, and the dynamic spectral characteristic primitives are extracted using the sliding window method to establish a three-dimensional matrix of dynamic spectral characteristic primitives.

[0032] The time series spectral characteristics of similar mineral samples are constructed into a three-dimensional tensor according to the three dimensions of acquisition timestamp, spatial coordinates and spectral feature items. Each spatial coordinate corresponds to an independent time series spectral feature sequence, forming a structured storage with the dimensions of time, space and spectral feature items.

[0033] The unit time interval of the time resolution is set as the window sliding step, that is, the sampling interval at each time point. For example, the time interval is set to 15 minutes (i.e., one unit step). In this way, new spectral data will be collected every 15 minutes, that is, each time the window is slid, the window contains 4 time step data points (i.e., one hour of spectral data). This ensures the smoothing of time series data and can effectively capture the changing trend in the time dimension.

[0034] The sliding window method is used to calculate the first-order derivative of the spectral feature term in the time dimension, and the forward difference method is used to extract the dynamic spectral feature primitives, and a three-dimensional matrix of dynamic spectral feature primitives is established, where the spectral feature primitives are expressed as follows: ; In the formula, represents the jth spectral feature item, is the time step, The rate of change of the spectral characteristic term within a time period, is the rate of change of the jth feature item at the tth time step, is the time step of the sliding window size, which is 4 by default in this embodiment. The dynamic spectrum feature primitive three-dimensional matrix is ​​finally expressed as , where T represents the number of time steps in the time dimension, S represents the spatial coordinates, and F represents the number of each spectral feature item.

[0035] In S5, distortion identification is performed on the spectral feature primitives corresponding to the time window with the same period as the previous sunshine cycle, and the corresponding spectral distortion fragments are eliminated.

[0036] At different timestamps and spatial coordinates, spectral data of the historical full sunshine cycle are collected. These data should cover the reflectance spectra of minerals under different lighting conditions from early morning to dusk, ensuring the timing and spatial uniformity of the data, and then obtaining a three-dimensional matrix of dynamic spectral characteristic primitives of the historical full sunshine cycle.

[0037] The dynamic spectral feature primitives of the historical full sunshine cycle are averaged. By averaging each spectral feature item in the time dimension, the seasonal and temporal fluctuations in the light cycle are eliminated to obtain the average dynamic feature primitives of the spectral features. The dynamic spectral feature primitives of the historical data are decomposed in two dimensions by the principal component analysis (PCA) method to reduce redundant information and extract the most representative feature components. According to the characteristic cosine similarity of the spectral feature primitives corresponding to the same time window of the historical sunshine cycle and the current sunshine cycle under the same spatial coordinates, a time-varying anomaly detection model based on feature primitive distortion evaluation is constructed. The expression is: ; In the formula, , are the spectral characteristic primitives corresponding to the i-th time window of the historical sunshine cycle and the current sunshine cycle, is the distortion evaluation value of the spectral feature primitive in the i-th time window.

[0038] The distortion threshold of spectral characteristic primitives is set based on the overall variation range of the spectral characteristics of the rock minerals in the area and the environmental conditions (such as instantaneous water film interference, the variation range of temperature difference over time, etc.). The time-varying anomaly detection model detects and screens the time windows in which the distortion evaluation value of the spectral characteristic primitives is greater than the distortion threshold, and removes the spectral distortion fragments corresponding to the time windows (if the distortion fragments are too long, data can be filled by interpolation method).

[0039] In S6, a dynamic spectral feature library is established, and a ready signal is generated based on the ratio of effective dynamic spectral feature primitive data to trigger the training of the rock and mineral identification model.

[0040] Based on the remaining valid dynamic spectral feature primitives after anomalies are eliminated, a dynamic spectral feature library is established and organized according to mineral categories. The LRU cache replacement strategy is used to manage the capacity of the dynamic spectral feature library. The LRU strategy evaluates the capacity of the existing library and decides whether to eliminate data of certain mineral categories based on the storage time. For example, in a cross-quarter scenario, the data of the previous quarter is cleaned up to prevent the mixing of atypical feature data caused by excessive environmental differences due to quarterly changes.

[0041] The number of continuous sunshine cycles covered by each mineral category in the spectral library is counted. A complete sunshine cycle contains at least records from early morning to dusk, including the changes in spectral characteristics of mineral reflectance spectra under different lighting conditions. When the target mineral category covers more than a set number of complete cycles (for example, data of at least 5 consecutive sunshine cycles of the target mineral category are counted to ensure data stability) and the proportion of valid dynamic spectral feature primitive data in each cycle exceeds a set ratio (set according to the missing length of the distorted fragment, the default setting is 90%), a ready signal is generated to trigger the rock mineral identification model training.

[0042] In S7, a rock and mineral identification model based on convolutional neural network is established and trained using the three-dimensional matrix of dynamic spectral feature primitives in the dynamic spectral feature library as a training data set.

[0043] Taking the three-dimensional matrix of dynamic spectral feature primitives in the dynamic spectral feature library as the training data set, a rock and mineral recognition model based on convolutional neural network is established, and a three-dimensional convolution kernel corresponding to the three-dimensional matrix structure of the dynamic spectral feature primitives is designed, covering the three dimensions of time, space and spectral characteristics.

[0044] The time dimension of the three-dimensional convolution kernel uses continuous time periods for convolution processing, the spatial dimension uses a convolution kernel of size n×m (3×3, 3×5 or 5×5 can be selected), and the spectral feature dimension selects continuous dynamic spectral feature primitives. The convolution kernel is set to slide to capture the changes in mineral spectra over time.

[0045] The pooling layer will use a 2x2 pooling window, slide with a step size of 2, take the maximum value in each window, and reduce the dimension of the feature map output by the convolution layer through the maximum pooling operation. The pooled feature map is input into the fully connected layer and connected to the Softmax classifier for mineral type classification.

[0046] The cross entropy loss function based on time-varying sensitivity weighting is selected to quantify the difference between the predicted results and the actual labels, and the rock mineral recognition model is iteratively trained. Expressed as: ; In the formula, is the time-varying sensitivity weight of the i-th time window, which is comprehensively set based on the variance within the time window period. and are the actual label and model predicted probability of the zth training sample respectively.

[0047] The training data set is divided into multiple subsets through the cross-validation method, and the model performance is evaluated using the accuracy, precision, and recall rate indicators (the default settings are 70%, 90%, and 80%, respectively). The training is terminated when all indicators of the rock and mineral identification model reach the preset indicator standards.

[0048] In S8, the trained rock and mineral identification model is deployed on the mobile spectral detection equipment to perform precise scanning and automatic identification of the entire rock and mineral detection area.

[0049] The trained rock and mineral identification model is deployed on mobile spectral detection equipment (such as portable spectrometers, spectral sensors mounted on drones, etc.) to accurately scan the entire rock and mineral detection area. The equipment will scan along a preset route within the detection area to ensure that every area is covered.

[0050] The mobile spectrum detection equipment is time-axis corrected with the full-sunshine cycle synchronous spectrum acquisition device. The mobile spectrum detection equipment is used to collect time series reflectance spectrum data of all rocks and minerals in the detection area, and the corresponding time series spectrum features are extracted.

[0051] The spectral characteristics of all rock and mineral time series in the detection area are subjected to the same three-dimensional structured processing as similar mineral samples. A three-dimensional matrix of dynamic spectral feature primitives for input identification is established and input into the rock and mineral identification model for automatic rock and mineral identification. The identification results will include information such as the mineral category label and probability value. For example, the identification results are "quartz" (probability: 95%) and "feldspar" (probability: 3%). Secondary resolution is performed on rock and minerals that cannot be accurately distinguished in the static spectral feature library.

[0052] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0053] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or may be transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium may be a solid-state hard disk.

[0054] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0055] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0056] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0057] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0058] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0059] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0060] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0061] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for automatically identifying rock minerals based on spectral information, characterized in that: The steps include: S1: Collect mixed minerals for spectral information detection, screen the mineral types that hit the spectral features in the static spectral feature library, and establish the corresponding correlation relationship between features and mineral types; S2: Retrieve similar mineral species whose corresponding relationships are not unique in the feature-mineral species corresponding relationship, obtain a set number of similar mineral samples, and place the similar mineral samples according to the distribution of the mixed mineral collection coordinates; S3: collect time series reflectance spectrum data of similar mineral samples and align the time series driven by timestamp to extract the spectral features of the time series reflectance spectrum data of similar mineral samples; S4: Perform three-dimensional structural processing on the time series spectral features of similar mineral samples, extract dynamic spectral feature primitives using the sliding window method, and establish a three-dimensional matrix of dynamic spectral feature primitives; S5: Distortion identification is performed on the spectral characteristic primitives corresponding to the time window of the same previous sunshine cycle, and the corresponding spectral distortion fragments are eliminated; S6: Establish a dynamic spectral feature library, and generate a ready signal to trigger the rock and mineral identification model training based on the ratio of effective dynamic spectral feature primitive data; S7: Using the three-dimensional matrix of dynamic spectral feature primitives in the dynamic spectral feature library as the training data set, a rock and mineral recognition model based on convolutional neural network is established and trained; S8: Deploy the trained rock and mineral identification model to the mobile spectral detection device to accurately scan and automatically identify the entire rock and mineral detection area.

2. The method for automatically identifying rock minerals based on spectral information according to claim 1, characterized in that: In S1, mixed minerals are collected for spectral information detection, mineral types that match the spectral features in the static spectral feature library are screened, and the corresponding correlation relationship between features and mineral types is established, which specifically includes: Collect mixed minerals with uniform coordinate distribution in the rock and mineral detection area; Retrieve the open-source mineral static spectral feature library, and based on the detection condition constraints required by the static spectral feature library, perform spectral information detection on the individual minerals of the mixed minerals under set experimental conditions to obtain a spectral identification data set of the mixed minerals; Extract static spectral features of mixed mineral spectral identification data set, and compare the static spectral features of mixed minerals with the features in the static spectral feature library for feature similarity, wherein the spectral features include wavelength, absorptivity and absorption peak; A feature similarity threshold is set, and static spectral features whose feature similarity exceeds the set feature similarity threshold are marked as hit features. The possible mineral types corresponding to the hit features in the mineral static spectral feature library are obtained, and a corresponding association relationship between features and mineral types is established.

3. The method for automatically identifying rock minerals based on spectral information according to claim 2, characterized in that: In S2, similar mineral species whose corresponding relationships are not unique in the feature-mineral species corresponding association relationship are retrieved, a set number of similar mineral samples are obtained, and the similar mineral samples are sampled according to the mixed mineral collection coordinate distribution.

4. The method for automatically identifying rock minerals based on spectral information according to claim 3, characterized in that: In S3, the time series reflectance spectrum data of similar mineral samples are collected and the time series alignment driven by timestamp is performed, and the spectral features of the time series reflectance spectrum data of similar mineral samples are extracted, which specifically includes: Configure a full-sunshine-cycle synchronous spectrum acquisition device in the rock and mineral detection area, set the spectrum acquisition time resolution and the band range of the incident spectrum, and collect reflectance spectrum data of time series for samples with similar mineral sample placement coordinates; Perform time series alignment of all time series reflectance spectral data of similar mineral samples based on spectral acquisition timestamp-driven time series alignment; Extract spectral features of time series reflectance spectral data of similar mineral samples.

5. The method for automatically identifying rock minerals based on spectral information according to claim 4, characterized in that: In S4, the spectral characteristics of the time series of similar mineral samples are processed in three dimensions, and the dynamic spectral characteristic primitives are extracted by the sliding window method to establish a three-dimensional matrix of dynamic spectral characteristic primitives, which specifically includes: The time series spectral features of similar mineral samples are constructed into a three-dimensional tensor according to the three dimensions of acquisition timestamp, spatial coordinates and spectral feature items. Each spatial coordinate corresponds to an independent time series spectral feature sequence, forming a structured storage with dimensions of time, space and spectral feature items. The unit time interval of the spectral acquisition time resolution is set as the window sliding step, the window size is 4 steps, and the sliding window method is used to calculate the first-order derivative of the spectral feature term in the time dimension, extract the dynamic spectral feature primitives, and establish a three-dimensional matrix of dynamic spectral feature primitives.

6. The method for automatically identifying rock minerals based on spectral information according to claim 5, characterized in that: In S5, the distortion identification is performed on the spectral characteristic primitives corresponding to the time window of the same sunshine cycle as the previous sunshine cycle, and the corresponding spectral distortion fragments are eliminated, specifically including: The dynamic spectrum characteristic primitive three-dimensional matrix of the historical full sunshine cycle is obtained, and the dynamic spectrum characteristic primitive is processed with characteristic mean and decomposed in two dimensions based on spatial coordinates. According to the characteristic cosine similarity of the spectral characteristic primitives corresponding to the same time window of the historical sunshine cycle and the current sunshine cycle under the same spatial coordinates, a time-varying anomaly detection model based on characteristic primitive distortion evaluation is constructed. The expression is: ； In the formula, , are the spectral feature primitives corresponding to the i-th time window of the historical sunshine cycle and the current sunshine cycle, is the distortion evaluation value of the spectral characteristic primitive of the i-th time window; The distortion threshold of the spectral feature primitive is set, and the time-varying anomaly detection model detects and selects the time window in which the distortion evaluation value of the spectral feature primitive is greater than the distortion threshold, and removes the spectral distortion fragment corresponding to the time window.

7. The method for automatically identifying rock minerals based on spectral information according to claim 6, characterized in that: In S6, a dynamic spectral feature library is established, and a ready signal is generated based on the ratio of effective dynamic spectral feature primitive data to trigger the rock and mineral identification model training, which specifically includes: Based on the remaining effective dynamic spectral feature primitives after anomaly elimination, a dynamic spectral feature library is established according to mineral categories, and the LRU strategy is used to maintain the library capacity. The number of continuous sunshine cycles covered by each mineral category in the spectral library is counted. When the target mineral category covers more than a set number of complete cycles and the proportion of valid dynamic spectral feature primitive data in each cycle exceeds a set ratio, a ready signal is generated to trigger the rock mineral identification model training.

8. The method for automatically identifying rock minerals based on spectral information according to claim 7, characterized in that: In S7, the three-dimensional matrix of dynamic spectral feature primitives in the dynamic spectral feature library is used as a training data set to establish and train a rock and mineral identification model based on a convolutional neural network, specifically including: Using the dynamic spectral feature primitive three-dimensional matrix in the dynamic spectral feature library as the training data set, a rock and mineral recognition model based on convolutional neural network is established, and a three-dimensional convolution kernel corresponding to the dynamic spectral feature primitive three-dimensional matrix structure is designed, covering the three dimensions of time, space and spectral characteristics; The time dimension of the three-dimensional convolution kernel uses continuous time periods for convolution processing, the spatial dimension uses a convolution kernel of size n×m, and the spectral feature dimension selects continuous dynamic spectral feature primitives. The convolution kernel is set to slide to capture the changes of mineral spectra over time. The feature map output by the convolution layer is reduced in dimension through the maximum pooling operation, and the pooled feature map is input into the fully connected layer and connected to the Softmax classifier for mineral type classification; A cross-entropy loss function based on time-varying sensitivity weighting is selected to quantify the difference between the predicted results and the actual labels, and the rock and mineral recognition model is iteratively trained; The training data set is divided into multiple subsets through the cross-validation method, and the model performance is evaluated using accuracy, precision, and recall indicators. The training is terminated when all indicators of the rock and mineral identification model reach the preset indicator standards.

9. The method for automatically identifying rock minerals based on spectral information according to claim 8, characterized in that: In S8, the trained rock and mineral identification model is deployed on the mobile spectral detection device to accurately scan and automatically identify the rock and mineral detection area in the entire area. Specifically, the following steps are performed: Deploy the trained rock and mineral identification model to the mobile spectral detection equipment to accurately scan the entire rock and mineral detection area; The mobile spectrum detection equipment is time-calibrated with the set full-sunshine cycle synchronous spectrum acquisition device, and the time series reflection spectrum data of all rocks and minerals in the detection area are collected by the mobile spectrum detection equipment, and the corresponding time series spectrum features are extracted; The spectral characteristics of all rock and mineral time series in the detection area are subjected to the same three-dimensional structural processing as similar mineral samples, and a three-dimensional matrix of dynamic spectral characteristic primitives for input identification is established and input into the rock and mineral identification model for automatic rock and mineral identification.

Images