AI drill core identification and automatic logging system for external investigation operation robot

Abstract

The invention discloses an AI drill core identification and automatic logging system for an external investigation operation robot, relates to the technical field of geological investigation, and solves the technical problems that different feature data of a drill core are difficult to analyze and construct a fusion feature value, and an AI model containing multiple feature branches is difficult to construct to identify the drill core. Comprising the following modules: a multi-source data acquisition module used for installing a camera and a plurality of sensors to acquire corresponding data; the sensor cooperative work module is used for the master control system to establish a clock source, synchronize camera and sensor data acquisition and add marks; the data processing and analyzing module is used for analyzing the three kinds of data and constructing corresponding characteristic values; the AI recognition module is used for constructing an AI model containing texture, spectrum and mechanical branches, deploying after feature fusion and cross entropy loss training, and outputting category probability distribution; and the automatic logging module is used for automatically generating logging information after the AI model identifies the drill core.

Description

Technical Field

[0001] The present invention belongs to the technical field of geological exploration, and particularly relates to an AI core identification and automatic cataloging system for outdoor exploration operation robots. Background Art

[0002] In geological outdoor exploration operations, the collection and analysis of core samples are important means to obtain geological information. With the development of artificial intelligence technology, it has become possible to use AI technology to achieve automatic identification and cataloging of cores. An AI core identification and automatic cataloging system for outdoor exploration operation robots collects features associated with cores through various devices, analyzes and constructs corresponding feature values, inputs them into an AI recognition model for training, and deploys the trained model into the system to achieve automatic identification of cores.

[0003] However, traditional core processing procedures often rely on manual classification, marking, and cataloging, which are not only inefficient but also prone to errors and omissions due to human factors. Although existing core processing and identification can meet core identification to a certain extent, there are still problems such as low identification accuracy and incomplete cataloging information, which cannot meet the needs of actual exploration operations. It is difficult to query the corresponding radius value in the mapping table through the comprehensive texture complexity coefficient, use the automatically adjusted radius to extract texture features from core images, and it is also difficult to analyze the spectral and mechanical features of cores and construct corresponding feature values. There is a lack of an AI model that includes texture, spectral, and mechanical branches to automatically classify core types. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art; for this purpose, the present invention provides an AI core identification and automatic cataloging system for outdoor exploration operation robots to solve the following technical problems: Traditional core processing procedures often rely on manual classification, marking, and cataloging, which are not only inefficient but also prone to errors and omissions due to human factors. Although existing core processing and identification can meet core identification to a certain extent, there are still problems such as low identification accuracy and incomplete cataloging information, which cannot meet the needs of actual exploration operations. It is difficult to query the corresponding radius value in the mapping table through the comprehensive texture complexity coefficient, use the automatically adjusted radius to extract texture features from core images, and it is also difficult to analyze the spectral and mechanical features of cores and construct corresponding feature values. There is a lack of an AI model that includes texture, spectral, and mechanical branches to automatically classify core types.

[0005] To solve the above problems, the present invention provides an AI core identification and automatic cataloging system for outdoor exploration operation robots, including the following modules: Multi-source data acquisition module: A high-resolution camera is installed on the external exploration operation robot to collect images of the core samples in real time, and a hyperspectral sensor and a six-axis force sensor are integrated on the same device to collect corresponding data; Sensor collaborative work module: The main control system establishes a clock source to synchronize the data acquisition of the camera and the sensors and add marks, starts the robot according to the motion plan, and uses the core sample reaching the sensor as a trigger for data acquisition; Data processing and analysis module: Analyze the core sample images to determine the relationship between the comprehensive texture complexity coefficient and the LBP radius, use the automatically adjusted radius to extract the texture features of the core sample images, analyze the spectral curves, combine multiple features to obtain the fused spectral features, analyze the force and torque signals, and fuse the time-domain features and frequency-domain features of the mechanical characteristics; AI recognition module: Construct a unified timestamp coordinate system for the core sample, align the data and perform Z-score standardization, construct an AI model including texture, spectral, and mechanical branches, and after feature fusion and cross-entropy loss training, deploy it to output the class probability distribution; Automatic cataloging module: After the AI model recognizes the core sample, it automatically generates cataloging information, constructs a database to store relevant information, and designs a user interface to display the core sample recognition information.

[0006] As a further solution of the present invention, the step of installing a high-resolution camera on the external exploration operation robot to collect images of the core samples in real time, and integrating a hyperspectral sensor and a six-axis force sensor on the same device to collect corresponding data includes the following steps: Fix the high-resolution camera with a seismic support, connect the power line and data transmission line of the camera to the main control system of the robot; determine the installation position of the hyperspectral sensor according to the measurement distance and measurement angle between the hyperspectral sensor and the core sample, connect the power line and signal output line of the hyperspectral sensor to the main control system of the robot, and add a corresponding analog-to-digital conversion module in the main control system to convert it into a digital signal; install the six-axis force sensor at the connection between the drill bit and the robotic arm, and connect its signal line to the main control system of the robot; Perform initialization settings on the camera, hyperspectral sensor, and six-axis force sensor in the main control system.

[0007] As a further solution of the present invention, the step of the main control system establishing a clock source to synchronize the data acquisition of the camera and the sensors and add marks, starting the robot according to the motion plan, and using the core sample reaching the sensor as a trigger for data acquisition includes the following steps: Establish a clock source in the main control system, and use the clock synchronization signal to ensure that the camera and various sensors start data collection at the same time, and add synchronization marks to the collected data; start the movement of the external survey robot according to the preset motion plan, and when the drill core reaches the measurement position of the sensor, trigger the high-resolution camera, hyperspectral sensor and six-dimensional force sensor to start collecting data; The high-resolution camera continuously collects images of the drill core at a set frame rate, and the image data is transmitted to the main control system in the form of digital signals through the data transmission line; the hyperspectral sensor collects spectral information on the surface of the drill core after receiving the trigger signal; the six-dimensional force sensor converts the measured force and torque signals into electrical signals; The storage format of image data, spectral data and mechanical data is determined in the main control system, and metadata is added, including acquisition time, location and drill core number information.

[0008] As a further solution of the present invention, the analysis of the drill core image comprises the following steps: The collected drill core image is preprocessed, including denoising, white balance adjustment and image enhancement, the color image is converted into a grayscale image, the grayscale co-occurrence matrix of the drill core image is calculated, and texture features are extracted, including contrast, correlation and energy. After the extracted texture features are normalized, multiple texture feature parameters are combined by weighted summation to obtain a first texture complexity coefficient; At the same time, the gradient vector of each pixel point in the drill core image is calculated to obtain the gradient matrix, and the structure tensor is obtained by smoothing the gradient matrix. The structure tensor is eigenvalue decomposed to obtain two eigenvalues, and the eigenvalues ​​of multiple pixels in different directions are averaged to obtain the mean of the eigenvalues ​​in the corresponding direction. The variance of the eigenvalue in each direction is calculated, the variance in each direction is sorted, and the corresponding weight coefficients are assigned in order from large to small, and weighted addition is performed to obtain the second texture complexity coefficient.

[0009] As a further solution of the present invention, the determination of the relationship between the comprehensive texture complexity coefficient and the LBP radius and the use of the automatically adjusted radius to extract texture features from the drill core image include the following steps: The first texture complexity coefficient and the second texture complexity coefficient are averaged to obtain a comprehensive texture complexity coefficient. According to experimental data, the relationship between the comprehensive texture complexity coefficient and the LBP radius is determined, and a coefficient-radius mapping table is established to record the optimal LBP radius corresponding to each comprehensive texture complexity coefficient. According to the calculated comprehensive texture complexity coefficient, the corresponding radius value is queried in the mapping table, and the automatically adjusted radius is used to extract texture features of the drill core image.

[0010] As a further solution of the present invention, the analysis of the spectral curve includes the following steps: Determine that the spectral coverage range is 350 - 2500 nanometers, obtain the complete spectral curve for each acquisition point, record its wavelength and the corresponding reflectivity, and perform preprocessing operations on the spectral data, including noise removal and baseline correction; Perform convolution operations on the spectral curve using Gaussian kernel filters of different sizes, detect peaks at different scales, synthesize the detection results of each scale, and determine the final absorption peak position through weighted average; after determining the absorption peak position, take the reflectivity at this position as the starting point, and obtain the absorption peak depth as the distance from the vertical direction to the baseline, and perform standardization processing on the absorption peak depth; search for points with reflectivity reaching the half-height position along the wavelength axis on both sides of the absorption peak, and record the wavelength difference between these two points as the full width at half maximum (FWHM), and use the same method to obtain the full width at one-tenth maximum (FWTM) and the full width at one-quarter maximum (FWQM), and determine the weight coefficients corresponding to the three width characteristics according to experimental experience, and add them with weights to obtain the comprehensive width characteristic.

[0011] As a further solution of the present invention, the combination of multiple features to obtain the fused spectral feature includes the following steps: Traverse the entire spectral curve data, find all the positions identified as absorption peaks, judge whether the distribution of the absorption peak positions is uniform, determine the corresponding mapping function according to whether the distribution is uniform, and convert the absorption peak positions into values in the range of 0 to 1; Determine the weight coefficients corresponding to the position, depth, and width characteristics, and the sum of the three weight coefficients is equal to 1. Add the position, depth, and width characteristics with weights according to the determined weight coefficients to obtain the fused spectral feature.

[0012] As a further solution of the present invention, the analysis of the force and torque signals, fusing the time-domain features and the frequency-domain features of mechanical characteristics, includes the following steps: After performing preprocessing operations on the force and torque signals collected by the six-dimensional force sensor, extract the statistical features of the signals, including the mean value, impact factor, and energy ratio. Use the Min-Max normalization method to map the values of the three features between 0 and 1, and determine the weights corresponding to the three features, and add them with weights to obtain the fused feature value; extract the waveform features of the signals, including the skewness coefficient and kurtosis coefficient of the waveform, divide the two coefficients by the corresponding standard deviations respectively to obtain the standardized skewness coefficient and kurtosis coefficient, and add them with weights to obtain the fused waveform feature, and combine it with the fused statistical feature to obtain the fused time-domain vector; Perform FTT transformation on the preprocessed force and torque signals to obtain the spectral representation of the signals, extract the frequency components and amplitudes, determine the effective frequency range, and obtain the overall kurtosis value of the signals based on the spectrogram. Integrate the frequency components, their amplitudes, and the overall kurtosis value to construct a feature vector containing multiple features, perform normalization processing on it, and fuse each eigenvalue in the normalized feature vector through weighted fusion to obtain the fused frequency-domain features. Add the fused time-domain features and frequency-domain features with weights to obtain the fused mechanical features.

[0013] As a further solution of the present invention, construct an AI model including texture, spectral, and mechanical branches, and after feature fusion and cross-entropy loss training, deploy it to output the category probability distribution, including the following steps: Establish a unified timestamp coordinate system for each core sample. Based on the triggering moment of the mechanical sensor, align the texture images and spectral data within a ±50 ms time window, and perform Z-score normalization processing on the extracted texture features, spectral features, and mechanical feature dimensions. Construct the AI model architecture: The input layer includes a texture branch: RGB image input of 224*224*3, a spectral branch: 420-dimensional spectral curve, and a mechanical branch: 6-dimensional force signal * 1000 time steps; The texture network uses an improved ResNet18 architecture, changing the last fully connected layer to a 512-dimensional output. The spectral network uses a 1D convolutional layer and adds a spectral attention module. The mechanical network uses a bidirectional LSTM layer and adds a time attention mechanism; The feature fusion layer uses a cross-modal attention mechanism to splice the high-order features of each modality to form a 1024-dimensional joint feature vector. Select weighted cross-entropy loss as the loss function to train the model: In the first stage, freeze the texture network and train the spectral or mechanical branch. In the second stage, unfreeze all networks and perform end-to-end training, using the Lookhead optimizer, k = 5, α = 0.5; Deploy the trained AI model to the actual core identification task, and the model outputs the probability distribution of each category.

[0014] As a further solution of the present invention, after the AI model identifies the core, it automatically generates catalog information, constructs a database to store relevant information, and designs a user interface to display the core identification information, including the following steps: According to the identification result of the AI model, automatically generate catalog information for the core. The catalog information includes the category, source, feature description, and identification date of the core, and supports customizing catalog fields and templates. Design a user interface to display the image, spectral data, mechanical features, and identification results of the core, provide editing and viewing functions for the catalog information, and allow users to modify and supplement the catalog information.

[0015] Advantages of the present invention: By determining the LBP radius through the comprehensive texture complexity coefficient, the present invention can make the scale of texture feature extraction more accurately adapt to the texture changes in the image. A fixed radius may not provide the optimal feature representation under different texture complexities. By dynamically adjusting the radius according to the comprehensive texture complexity coefficient, the features extracted by LBP can better reflect the actual texture structure of the image, thereby providing a more discriminative feature vector for subsequent classification, recognition, or analysis tasks; In the present invention, three types of core-related feature data are respectively set in the input layer. The feature fusion layer adopts a cross-modal attention mechanism to splice the high-order features of each modality to form a 1024-dimensional joint feature vector, and a phased training strategy is used to optimize the model performance. Setting the weighted cross-entropy loss can assign different weights to different classes, enabling the model to pay more attention to those difficult-to-recognize or minority-class samples during training, avoiding the model from being biased towards the majority-class samples due to class imbalance, achieving the improvement of the generalization ability of the model on different datasets, and when the model is deployed to the actual core recognition task, it can better adapt to various unseen data situations and ensure a high recognition accuracy. Brief Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0017] Figure 1 It is a block diagram of the module flow of the present invention. Detailed Embodiments

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0019] Please refer to Figure 1 As shown, the present invention is an AI core recognition and automatic cataloging system for outdoor exploration operation robots, including the following modules: Multi-source data acquisition module: A high-resolution camera is installed on the outdoor exploration operation robot to collect images of the core in real time, and a hyperspectral sensor and a six-axis force sensor are integrated on the same device to collect corresponding data; Sensor Collaboration Working Module: The main control system establishes a clock source, synchronizes the data acquisition of the camera and sensors, adds markers, starts the robot according to the motion plan, and triggers the acquisition when the core drilling reaches the sensor. Data Processing and Analysis Module: Analyze the core drilling images to determine the relationship between the comprehensive texture complexity coefficient and the LBP radius, extract the texture features of the core drilling images using the automatically adjusted radius, analyze the spectral curves, combine multiple features to obtain the fused spectral features, analyze the force and torque signals, and fuse the time-domain features and frequency-domain features into mechanical features. AI Recognition Module: Construct a unified timestamp coordinate system for the core drilling samples, align the data, perform Z-score standardization, construct an AI model including texture, spectral, and mechanical branches, and after feature fusion and cross-entropy loss training, deploy it to output the class probability distribution. Automatic Cataloging Module: After the AI model recognizes the core drilling, it automatically generates cataloging information, constructs a database to store relevant information, and designs a user interface to display the core drilling recognition information.

[0020] Specifically, install a camera and multiple sensors in the system respectively. Create a synchronous signal generator module in the main control system, configure the module so that it can generate synchronous pulse signals according to the signals of the clock source. Determine the storage formats of image data, spectral data, and mechanical data in the main control system. Create a metadata database or data structure in the main control system to store information such as acquisition time, location, and core drilling number. Analyze the image data collected by the camera, calculate the first and second texture complexity coefficients, add them to obtain the comprehensive texture complexity coefficient, construct a corresponding mapping table with the LBP radius, and determine the best radius to analyze the texture features of the core drilling. Analyze the spectral data collected by the spectral sensor, weight and add the preprocessed position, depth, and width features according to the determined weight coefficients to obtain the fused spectral features. Analyze the force and torque signals collected by the six-axis force sensor, and fuse the analyzed time-domain features and frequency-domain features to obtain the final mechanical features. Use the three features as the input of the model to construct an AI model that can recognize the core drilling type. Design an automatic cataloging system according to the recognition structure of the AI model.

[0021] In one embodiment of the present invention, a high-resolution camera is installed on the external exploration operation robot to collect images of the core drilling in real time, and a hyperspectral sensor and a six-axis force sensor are integrated on the same device to collect corresponding data, including the following steps: Fix a high - resolution camera with a seismic support, and connect the power cord and data transmission line of the camera to the main control system of the robot; determine the installation position of the hyperspectral sensor according to the measured distance and measurement angle between the hyperspectral sensor and the core sample, connect the power cord and signal output line of the hyperspectral sensor to the main control system of the robot, and add a corresponding analog - to - digital conversion module in the main control system to convert it into a digital signal; install a six - axis force sensor at the connection between the drill bit and the robotic arm, and connect its signal line to the main control system of the robot; Perform initialization settings for the camera, hyperspectral sensor, and six - axis force sensor in the main control system.

[0022] Specifically, check whether the seismic support firmly fixes the high - resolution camera to ensure that it will not be displaced due to factors such as vibration during subsequent operations. Re - confirm that the power cord and data transmission line of the camera have been correctly and firmly connected to the main control system of the robot. After installation, according to the configuration file or setting interface provided by the driver, and according to the requirements of the measured distance and measurement angle between the hyperspectral sensor and the core sample, carefully confirm whether its installation position is correct and determine that it is successfully connected to the main control system. In the main control system, find the corresponding analog - to - digital conversion module configuration interface, set various parameters of the analog - to - digital conversion so that it can correctly convert the analog signal output by the hyperspectral sensor into a digital signal. Check whether the six - axis force sensor is correctly installed at the connection between the drill bit and the robotic arm and whether the installation is firm and reliable, and send an instruction through the main control system to read the data of the six - axis force sensor.

[0023] The initialization settings for the camera, hyperspectral sensor, and six - axis force sensor in the main control system are shown in Table 1: Table 1.

[0024] In one embodiment of the present invention, the main control system establishes a clock source, synchronizes the data acquisition of the camera and the sensors and adds markers, and starts the robot according to the motion planning. The core sampling to the sensor is used as a trigger for data acquisition, including the following steps: Establish a clock source in the main control system. Through the clock synchronization signal, determine that the camera and each sensor start data acquisition at the same moment, and add synchronization markers to the acquired data; start the motion of the external exploration operation robot according to the preset motion planning. When the core sample reaches the measurement position of the sensor, trigger the high - resolution camera, hyperspectral sensor, and six - axis force sensor to start data acquisition; The high - resolution camera continuously acquires images of the core sample at the set frame rate, and the image data is transmitted to the main control system in the form of a digital signal through the data transmission line; after receiving the trigger signal, the hyperspectral sensor acquires the spectral information on the surface of the core sample; the six - axis force sensor converts the measured force and torque signals into electrical signals; The storage format of image data, spectral data and mechanical data is determined in the main control system, and metadata is added, including acquisition time, location and drill core number information.

[0025] Specifically, connect the clock source device to the main control system through a suitable interface, and install the corresponding driver so that it can be recognized by the main control system, enter the operating system of the main control system, find the clock source setting option, set the clock source parameters, including frequency and phase, according to the technical manual of the clock source, create a synchronization signal generator module in the main control system, configure the module so that it can generate a synchronization pulse signal according to the signal of the clock source, and use the robot programming language to write a motion planning program according to the requirements of the field survey operation. Define the robot's motion trajectory, speed, acceleration and other parameters in detail in the program to ensure that it can accurately move to the core collection position according to the preset path, integrate the written motion planning program into the robot's control system, and after confirming that the motion planning program and the robot control system are working properly, control the robot through the control interface of the main control system or the remote control terminal Send a start command to start the movement of the outdoor survey robot, enable the position sensor in the main control system to monitor the position of the drill core, and when the drill core approaches the measurement position of the sensor, the position sensor can promptly feedback the position information to the main control system. According to the information fed back by the position sensor, when the drill core reaches the set measurement position, the main control system sends a trigger signal to the high-resolution camera, the hyperspectral sensor and the six-dimensional force sensor through the control line. According to the type of data acquisition and the needs of subsequent processing, the storage format of the image data, the spectral data and the mechanical data is determined in the main control system, and a metadata database or data structure is created in the main control system to store information such as the acquisition time, location and drill core number. When a group of data is acquired, the corresponding metadata is associated with the group of data. For example, this information can be recorded in the header of the data file or through a separate metadata file.

[0026] In one embodiment of the present invention, the analyzing of the drill core image comprises the following steps: The collected drill core image is preprocessed, including denoising, white balance adjustment and image enhancement, the color image is converted into a grayscale image, the grayscale co-occurrence matrix of the drill core image is calculated, and texture features are extracted, including contrast, correlation and energy. After the extracted texture features are normalized, multiple texture feature parameters are combined by weighted summation to obtain a first texture complexity coefficient; Meanwhile, calculate the gradient vector of each pixel point in the core drilling image to obtain a gradient matrix. Smooth the gradient matrix to obtain a structure tensor, perform eigenvalue decomposition on the structure tensor to obtain two eigenvalues, average the eigenvalues of multiple pixel points in different directions respectively to obtain the mean value of the eigenvalues in the corresponding direction, then calculate the variance of the eigenvalues in each direction, sort the variances of each direction, assign corresponding weight coefficients in descending order, and perform weighted summation to obtain the second texture complexity coefficient.

[0027] Specifically, according to the selected conversion algorithm, traverse each pixel point in the image, convert its RGB value to a grayscale value to obtain a grayscale image. Select the directions and distances for calculating the gray-level co-occurrence matrix. The directions include 0°, 45°, 90°, and 135°, and the distance is 1 pixel. For the given directions and distances, count the number of occurrences of pixel pairs that satisfy the corresponding positional relationship in the grayscale image to construct a gray-level co-occurrence matrix. Calculate the contrast, correlation, and energy respectively using the corresponding calculation formulas, and normalize the extracted texture features using methods such as min-max normalization, including normalizing the contrast, correlation, and energy texture features respectively to obtain the normalized texture feature values. Set the weight of the contrast to 0.5, the weight of the correlation to 0.3, and the weight of the energy to 0.2. Multiply the normalized texture feature values by the corresponding weight coefficients, and then add the results of each feature to obtain the first texture complexity coefficient. For each pixel point in the grayscale image, calculate its gradient vector, combine the gradient vectors of each pixel point to construct a gradient matrix, select methods such as Gaussian filtering to smooth the gradient matrix to obtain a smoothed gradient matrix. According to the smoothed gradient matrix, calculate the structure tensor, perform eigenvalue decomposition on the structure tensor to obtain two eigenvalues λ1 and λ2, average the eigenvalues of multiple pixel points in the 0° and 90° directions respectively to obtain the mean value of the eigenvalues in the corresponding direction. According to the mean value of the eigenvalues in each direction, calculate the variance of the eigenvalues in each direction, and sort the variances of the eigenvalues in each direction in descending order. Then, according to the sorting results, assign weight coefficients. The weight of the first direction is 0.6, and the weight of the second direction is 0.4. Multiply the variance in each direction by the corresponding weight coefficient, and then add the results of each direction to obtain the second texture complexity coefficient; Formula for calculating the first texture complexity coefficient:

[0028] Where, is the first texture complexity coefficient, , and are the contrast, correlation, and energy after normalization processing, , and is the corresponding weight coefficient; The calculation formula for the second texture complexity coefficient:

[0029] Wherein, is the second texture complexity coefficient, and are the variances of the eigenvalues in each direction, and are the corresponding weight coefficients.

[0030] In one embodiment of the present invention, the relationship between the determined comprehensive texture complexity coefficient and the LBP radius is used to extract texture features from the core image using the automatically adjusted radius, including the following steps: Average the first texture complexity coefficient and the second texture complexity coefficient to obtain the comprehensive texture complexity coefficient. According to the experimental data, determine the relationship between the comprehensive texture complexity coefficient and the LBP radius, establish a coefficient - radius mapping table, and record the optimal LBP radius corresponding to each comprehensive texture complexity coefficient; According to the calculated comprehensive texture complexity coefficient, query the corresponding radius value in the mapping table, and use the automatically adjusted radius to extract texture features from the core image.

[0031] Specifically, collect a series of representative core image samples covering different texture complexities and already labeled with category labels, including low complexity, medium complexity, and high complexity. For each image sample, use cross - validation to train an SVM classifier to determine the optimal LBP radius: for each candidate radius R ∈ , extract the LBP histogram features of the image, use 5 - fold cross - validation to train the SVM classifier, calculate the classification accuracy of each fold of the validation set, take the average of the classification accuracies as the score of this radius, compare the average accuracies of different radii, select the radius with the highest average accuracy as the optimal LBP radius. At the same time, calculate the comprehensive texture complexity coefficient of each sample image according to the previous steps, record (comprehensive texture complexity coefficient, optimal LBP radius), draw a scatter plot of the two, observe whether there is an interval division or a continuous relationship, find that the two show an obvious interval division, perform equal - width binning according to the comprehensive texture complexity coefficient range, assign the samples to the corresponding intervals according to the binning boundaries, and count the mode of the optimal LBP radii in each interval to construct a coefficient - radius mapping table.

[0032] The coefficient - radius mapping table is shown in Table 2: Table 2 Comprehensive texture complexity coefficient range (Z) Recommended LBP radius Theoretical basis 0.0＜Z＜0.3 1 Low complexity 0.3≤Z＜0.6 2 Medium complexity 0.6≤Z＜1.0 3 High complexity In one embodiment of the present invention, the analysis of the spectral curve includes the following steps: Determine that the spectral coverage range is 350 - 2500 nanometers, obtain a complete spectral curve for each acquisition point, record its wavelength and the corresponding reflectance, and perform preprocessing operations on the spectral data, including noise removal and baseline correction; Perform convolution operations on the spectral curve using Gaussian kernel filters of different sizes, detect peaks at different scales, synthesize the detection results for each scale, and determine the final absorption peak position through weighted averaging; after determining the absorption peak position, take the reflectance at this position as the starting point, and obtain the absorption peak depth as the distance from the vertical direction to the baseline, and perform standardization processing on the absorption peak depth; search for points where the reflectance reaches the half-height position along the wavelength axis on both sides of the absorption peak, record the wavelength difference between these two points as the full width at half maximum (FWHM), and use the same method to obtain the full width at one-tenth maximum and the full width at one-quarter maximum, and determine the weight coefficients corresponding to the three width features according to experimental experience, and add them with weights to obtain the comprehensive width feature.

[0033] Specifically, during the acquisition process, ensure that the spectrometer can cover the spectral range of 350 - 2500 nanometers, obtain a complete spectral curve for each acquisition point, perform median filtering on the acquired spectral data in the wavelength direction to remove impulse noise and spike noise, and subtract the baseline value obtained by fitting from the original spectral data to obtain the corrected spectral data. Select Gaussian kernels of different sizes from 3×3 to 11×11 with a step size of 2, and perform convolution operations on the spectral curve for each size. For each data point on the spectral curve, multiply the value of the corresponding adjacent data point by the weight coefficient of the Gaussian kernel and then sum to obtain the convolution result. Among the convolution results at different scales, detect peaks respectively. According to experimental experience or preset rules, perform weighted averaging on the detection results at different scales to determine the absorption peak position. After determining the absorption peak position, take the reflectance at this position as the starting point, and obtain the absorption peak depth as the distance from the vertical direction to the baseline, and perform standardization processing on all detected absorption peak depths to eliminate the influence caused by differences in overall reflectance or other factors between different samples. Search for points where the reflectance reaches the half-height position along the wavelength axis on both sides of the absorption peak, record the wavelength difference between these two points as the full width at half maximum (FWHM). Similarly, search for points where the reflectance reaches one-tenth and one-quarter of the total absorption peak depth positions, calculate their wavelength differences to obtain the full width at one-tenth maximum and the full width at one-quarter maximum, and determine the weight coefficients corresponding to the three width features according to experimental experience. The weight of the full width at half maximum is 0.5, the weight of the full width at one-tenth maximum is 0.3, and the weight of the full width at one-quarter maximum is 0.2. Multiply each width feature by the corresponding weight coefficient and then add them to obtain the comprehensive width feature; Comprehensive width feature calculation formula:

[0034] Where, is the comprehensive width feature, when is 1, 2, and 3, corresponding to three width features of full width at half maximum, full width at one-tenth maximum, and full width at one-quarter maximum, is the th width feature, is the weight coefficient corresponding to the

[0035] In one embodiment of the present invention, the step of obtaining the fused spectral feature by combining multiple features includes the following steps: Traverse the entire spectral curve data, find all the positions identified as absorption peaks, determine whether the distribution of absorption peak positions is uniform, determine the corresponding mapping function according to whether the distribution is uniform, and convert the absorption peak positions into values in the range of 0 to 1; Determine the weight coefficients corresponding to the position, depth, and width features, and the sum of the three weight coefficients is equal to 1. Weight and add the position, depth, and width features according to the determined weight coefficients to obtain the fused spectral feature.

[0036] Specifically, along the wavelength axis direction of the spectral curve, with a certain step size, set to 1 nanometer, check each data point in turn. If the reflectivity of this point is less than the surrounding neighborhood points and meets a certain depth threshold of the absorption peak, then mark it as an absorption peak position, record the wavelength value at this position, calculate the wavelength difference between adjacent absorption peak positions to obtain a data set of wavelength differences, perform statistical analysis on these wavelength difference data. If the coefficient of variation is small, it is considered that the distribution of absorption peak positions is relatively uniform; otherwise, it is non-uniform. If the distribution of absorption peak positions is uniform, a linear mapping function can be used to convert the absorption peak positions into values in the range of 0 to 1; if the distribution of absorption peak positions is non-uniform, a mapping based on the cumulative distribution function is used. First, calculate the cumulative distribution function of the wavelengths of all absorption peak positions, and then for the wavelength of each absorption peak position, complete the mapping to the range of 0 to 1 by looking up its corresponding value in the CDF. According to experimental experience or theoretical analysis, initially set the weight coefficient ranges for the position, depth, and width features. The weight coefficient corresponding to the position feature is 0.4, and both the depth feature and the width feature are 0.3. Weight and add the pre-processed position, depth, and width features according to the determined weight coefficients to obtain the fused spectral feature.

[0037] In one embodiment of the present invention, the step of analyzing the force and torque signals and fusing the time-domain features and frequency-domain features of mechanical features includes the following steps: After preprocessing the force and torque signals collected by the six-dimensional force sensor, the statistical features of the signals are extracted, including the mean, impact factor and energy ratio. The Min-Max normalization method is used to map the values ​​of the three features between 0 and 1, and the weights corresponding to the three features are determined. The weighted addition is performed to obtain the fused feature values. The waveform features of the signals are extracted, including the skewness coefficient and kurtosis coefficient of the waveform. The two coefficients are divided by the corresponding standard deviation to obtain the standardized skewness coefficient and kurtosis coefficient. The weighted addition is performed to obtain the fused waveform features, which are then combined with the fused statistical features to obtain the fused time domain vector. Perform FTT transformation on the preprocessed force and torque signals to obtain the spectrum representation of the signal, extract the frequency components and amplitudes, determine the effective frequency range, and obtain the overall kurtosis value of the signal based on the spectrum diagram. Fuse the frequency components and their amplitudes as well as the overall kurtosis value to construct a feature vector containing multiple features, normalize it, and fuse each eigenvalue in the normalized feature vector through weighted fusion to obtain the fused frequency domain feature. The fused time domain features and frequency domain features are weighted and added to obtain the fused mechanical features.

[0038] Specifically, add up all the data points and divide by the number of data points to obtain the mean vector. Calculate the corresponding eigenvalues according to the calculation formulas of the impact factor and the energy ratio. Use the min-max normalization formula to map the values of the three features, namely the mean, the impact factor, and the energy ratio, to the range between 0 and 1. Determine the weight coefficients corresponding to the three features of the mean, the impact factor, and the energy ratio according to experimental experience or knowledge in related fields. Add the normalized features and the weights to obtain the fused statistical features. For each force and torque signal, calculate its skewness coefficient and kurtosis coefficient, and divide them by the corresponding standard deviations respectively to obtain the normalized skewness coefficient and the normalized kurtosis coefficient. Determine the weight coefficients corresponding to the skewness coefficient and the kurtosis coefficient according to experimental experience or knowledge in related fields. Multiply the eigenvalues of the normalized skewness coefficient and the normalized kurtosis coefficient features by the corresponding weight coefficients respectively, and then add them up to obtain the fused waveform features. Combine the fused statistical features and the fused waveform features to obtain the fused time-domain feature vector; perform a fast Fourier transform on the preprocessed force and torque signals to obtain the frequency spectrum representation of the signals. According to the frequency spectrum diagram, determine the main frequency components of the signals and their corresponding amplitudes. Select a frequency range that contains the main frequency components and has a relatively high energy proportion as the effective frequency interval. Calculate the overall kurtosis value of the signal according to the frequency spectrum diagram. Construct a feature vector F_frequency = [frequency component 1, amplitude 1, frequency component 2, amplitude 2,..., overall kurtosis value]. Normalize the feature vector. Determine the weight coefficients corresponding to each feature in the feature vector, and then multiply each eigenvalue in the normalized feature vector by the corresponding weight coefficient and add them up to obtain the fused frequency-domain feature. The weight coefficients corresponding to the fused time-domain and frequency-domain features are both 0.5.

[0039] In one embodiment of the present invention, the steps of constructing an AI model including texture, spectral, and mechanical branches, performing feature fusion and cross-entropy loss training, and then deploying it to output the category probability distribution include the following: Establish a unified timestamp coordinate system for each core sample. Based on the triggering moment of the mechanical sensor, align the texture images and spectral data within a time window of ±50 ms. Perform Z-score normalization processing on the dimensions of the extracted texture features, spectral features, and mechanical features. Construct the AI model architecture: The input layer includes a texture branch: RGB image input of 224*224*3, a spectral branch: 420-dimensional spectral curve, and a mechanical branch: 6-dimensional force signal * 1000 time steps; The texture network uses an improved ResNet18 architecture and changes the last fully connected layer to a 512-dimensional output. The spectral network uses a 1D convolutional layer and adds a spectral attention module. The mechanical network uses a bidirectional LSTM layer and adds a time attention mechanism; The feature fusion layer uses a cross-modal attention mechanism to splice the high-order features of each modality to form a 1024-dimensional joint feature vector; The weighted cross-entropy loss is selected as the loss function to train the model: in the first stage, the texture network is frozen, and the spectral or mechanical branch is trained. In the second stage, all networks are unfrozen and trained end-to-end. The Lookhead optimizer is used with k = 5 and α = 0.5. The trained AI model is deployed to the actual core identification task, and the model outputs the probability distribution of each category.

[0040] Specifically, in the construction of the AI model, the input layer settings include a texture branch, a spectral branch, and a mechanical branch. The texture branch is set to input an RGB image with a size of 224×224×3. The spectral branch is set to input a spectral curve with a size of 420 dimensions. The mechanical branch is set to input a 6-dimensional force signal × 1000 time steps. The network architecture construction includes a texture network, a spectral network, and a mechanical network. The texture network uses an improved ResNet18 architecture, and the last fully connected layer is changed to a 512-dimensional output. The spectral network uses a 1D convolutional layer to process the spectral curve and adds a spectral attention module to automatically assign different weights according to the importance of different wavelength points of the spectral data. The mechanical network uses a bidirectional LSTM layer to process the mechanical signal and adds a time attention mechanism. The feature fusion layer uses a cross-modal attention mechanism for feature fusion to automatically learn the relationships and weights between different modal features. In this way, the high-order features output by the three branches are concatenated into a 1024-dimensional joint feature vector. The weighted cross-entropy loss is selected as the loss function, and the Lookahead optimizer is used to train the model. The trained model is deployed to the actual core identification task to obtain the probability distribution results of each category for users to make decisions and analyze.

[0041] In one embodiment of the present invention, after the AI model identifies the core, it automatically generates catalog information, constructs a database to store relevant information, and designs a user interface to display the core identification information, including the following steps: According to the identification results of the AI model, catalog information is automatically generated for the core. The catalog information includes the category, source, feature description, and identification date of the core, and supports customizing catalog fields and templates. Design a user interface to display the image, spectral data, mechanical features, and identification results of the core, provide editing and viewing functions for the catalog information, and allow users to modify and supplement the catalog information.

[0042] Specifically, obtain the recognition results of the core samples from the trained AI model, and select the category with the highest probability as the category of the core samples. At the same time, extract the source information of the core samples from the original data, such as sampling location, sampling depth, etc. Generate a feature description of the core samples based on the previously extracted and standardized feature data. Design a catalog information template, including fixed fields such as "category", "source", "feature description", "recognition date", and customizable catalog fields. Set appropriate data types and formats for each field. Automatically generate the catalog information of the core samples according to the catalog information template and the extracted information. Design a user interface, which is divided into four main areas: an image display area, a spectral data display area, a mechanical feature display area, and a recognition result display area. The image display area supports the loading and display of core sample images in multiple image formats, such as JPEG, PNG, etc. The spectral data display area graphically displays the spectral data of the core samples, with the abscissa representing the wavelength and the ordinate representing the spectral intensity. The mechanical feature display area displays the mechanical feature data of the core samples in the form of charts. The recognition result display area intuitively displays the recognition results of the AI model, such as category name and recognition probability. In the catalog information table, click on a catalog field to pop up an edit box, allowing the user to modify the field content. Set a "save" button. After the user finishes modifying the catalog information and clicks save, the updated catalog information will be stored in the database.

[0043] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A system for AI core identification and automatic cataloging of an outdoor exploration operation robot, characterized in that, Includes the following modules: Multi-source data acquisition module: A high-resolution camera is installed on the outdoor survey robot to collect images of the drill core in real time, and a hyperspectral sensor and a six-dimensional force sensor are integrated into the same device to collect corresponding data; Sensor collaborative work module: The main control system establishes a clock source, synchronizes the camera and sensor data collection and adds markers, starts the robot according to the motion plan, and drills the core to the sensor to trigger the collection; Data processing and analysis module: Analyze the drill core image, determine the relationship between the comprehensive texture complexity coefficient and the LBP radius, use the automatically adjusted radius to extract the texture features of the drill core image, analyze the spectral curve, combine multiple features to obtain the fused spectral features, analyze the force and torque signals, and fuse the time domain features and frequency domain features to obtain mechanical features; AI recognition module: builds a unified timestamp coordinate system for drill core samples, aligns data and performs Z-score standardization, builds an AI model including texture, spectrum and mechanical branches, and deploys it after feature fusion and cross entropy loss training to output category probability distribution; Automatic cataloging module: After the AI ​​model identifies the drill core, it automatically generates cataloging information, builds a database to store relevant information, and designs a user interface to display the drill core identification information.

2. The system for AI core identification and automatic cataloging of an outdoor exploration operation robot according to claim 1, characterized in that, The method of installing a high-resolution camera on an outdoor survey robot to collect images of the drill core in real time and integrating a hyperspectral sensor and a six-dimensional force sensor on the same device to collect corresponding data includes the following steps: Use an anti-vibration bracket to fix the high-resolution camera, and connect the power cord and data transmission line of the camera to the main control system of the robot; determine the installation position of the hyperspectral sensor according to the measured distance and angle between the hyperspectral sensor and the drill core, connect the power cord and signal output line of the hyperspectral sensor to the main control system of the robot, and add a corresponding analog-to-digital conversion module to the main control system to convert it into a digital signal; install the six-dimensional force sensor at the connection between the drill bit and the robotic arm, and connect its signal line to the main control system of the robot; Initialize the camera, hyperspectral sensor and six-dimensional force sensor in the main control system.

3. The system for AI core identification and automatic cataloging of an outdoor exploration operation robot according to claim 1, characterized in that, The main control system establishes a clock source, synchronizes the camera and sensor data collection and adds markers, starts the robot according to the motion plan, and drills the core to the sensor as a trigger for collection, including the following steps: Establish a clock source in the main control system, and use the clock synchronization signal to ensure that the camera and various sensors start data collection at the same time, and add synchronization marks to the collected data; start the movement of the external survey robot according to the preset motion plan, and when the drill core reaches the measurement position of the sensor, trigger the high-resolution camera, hyperspectral sensor and six-dimensional force sensor to start collecting data; The high-resolution camera continuously collects images of the drill core at a set frame rate, and the image data is transmitted to the main control system in the form of digital signals through the data transmission line; the hyperspectral sensor collects spectral information on the surface of the drill core after receiving the trigger signal; the six-dimensional force sensor converts the measured force and torque signals into electrical signals; The storage format of image data, spectral data and mechanical data is determined in the main control system, and metadata is added, including acquisition time, location and drill core number information.

4. The system for AI core identification and automatic cataloging of an outdoor exploration operation robot according to claim 1, characterized in that, The analysis of the drill core image comprises the following steps: The collected drill core image is preprocessed, including denoising, white balance adjustment and image enhancement, the color image is converted into a grayscale image, the grayscale co-occurrence matrix of the drill core image is calculated, and texture features are extracted, including contrast, correlation and energy. After the extracted texture features are normalized, multiple texture feature parameters are combined by weighted summation to obtain a first texture complexity coefficient; At the same time, the gradient vector of each pixel point in the drill core image is calculated to obtain the gradient matrix, and the structure tensor is obtained by smoothing the gradient matrix. The structure tensor is eigenvalue decomposed to obtain two eigenvalues, and the eigenvalues ​​of multiple pixels in different directions are averaged to obtain the mean of the eigenvalues ​​in the corresponding direction. The variance of the eigenvalue in each direction is calculated, the variance in each direction is sorted, and the corresponding weight coefficients are assigned in order from large to small, and weighted addition is performed to obtain the second texture complexity coefficient.

5. The system for AI core identification and automatic cataloging of an outdoor exploration operation robot according to claim 1, characterized in that, Determining the relationship between the comprehensive texture complexity coefficient and the LBP radius, and using the automatically adjusted radius to extract texture features from the drill core image, comprises the following steps: The first texture complexity coefficient and the second texture complexity coefficient are averaged to obtain a comprehensive texture complexity coefficient. According to experimental data, the relationship between the comprehensive texture complexity coefficient and the LBP radius is determined, and a coefficient-radius mapping table is established to record the optimal LBP radius corresponding to each comprehensive texture complexity coefficient. According to the calculated comprehensive texture complexity coefficient, the corresponding radius value is queried in the mapping table, and the automatically adjusted radius is used to extract texture features of the drill core image.

6. The system for AI core identification and automatic cataloging of an outdoor exploration operation robot according to claim 1, characterized in that, The analysis of the spectral curve comprises the following steps: Determine that the spectrum covers a range of 350-2500 nanometers, obtain a complete spectrum curve for each acquisition point, record its wavelength and corresponding reflectance, and perform preprocessing operations on the spectrum data, including noise removal and baseline correction; Use Gaussian kernel filters of different sizes to perform convolution operations on the spectral curve, detect peaks at different scales, integrate the detection results of each scale, and determine the final absorption peak position by weighted average; after determining the absorption peak position, take the reflectivity at that position as the starting point, and obtain the absorption peak depth along the vertical distance to the baseline, and then standardize the absorption peak depth; search for points where the reflectivity reaches the half-height position along the wavelength axis on both sides of the absorption peak, and record the wavelength difference between these two points as the half-height full width. Use the same method to obtain one-tenth the full width and one-quarter the full width, and determine the weight coefficients corresponding to the three width features based on experimental experience, and perform weighted addition to obtain the comprehensive width feature.

7. The system for AI core identification and automatic cataloging of an outdoor exploration operation robot according to claim 1, characterized in that, The method of combining multiple features to obtain a fused spectral feature comprises the following steps: Traverse the entire spectral curve data, find all positions identified as absorption peaks, determine whether the distribution of absorption peak positions is uniform, determine the corresponding mapping function according to whether the distribution is uniform, and convert the absorption peak positions into values in the range of 0 to 1; Determine the weight coefficients corresponding to the position, depth, and width features, and the sum of the three weight coefficients is equal to 1. Weight and add the position, depth, and width features according to the determined weight coefficients to obtain the fused spectral features.

8. The system for AI core identification and automatic cataloging of an outdoor exploration operation robot according to claim 1, characterized in that,The analysis of the force and torque signals, fusing the time-domain features and frequency-domain features of the mechanical features, includes the following steps: After preprocessing the force and torque signals collected by the six-axis force sensor, extract the statistical features of the signals, including the mean value, impact factor, and energy ratio. Use the Min-Max normalization method to map the values of the three features between 0 and 1, and determine the weights corresponding to the three features, and perform weighted addition to obtain the fused feature value; extract the waveform features of the signal, including the skewness coefficient and kurtosis coefficient of the waveform, divide the two coefficients by the corresponding standard deviation respectively to obtain the normalized skewness coefficient and kurtosis coefficient, and perform weighted addition to obtain the fused waveform feature, and combine it with the fused statistical feature to obtain the fused time-domain vector; Perform FTT transformation on the preprocessed force and torque signals to obtain the spectral representation of the signals, extract the frequency components and amplitudes, determine the effective frequency interval, and obtain the overall kurtosis value of the signals according to the spectrogram. Fuse the frequency components, their amplitudes, and the overall kurtosis value to construct a feature vector containing multiple features, perform normalization processing on it, and fuse each feature value in the normalized feature vector through weighted fusion to obtain the fused frequency-domain feature; Perform weighted addition on the fused time-domain features and frequency-domain features to obtain the fused mechanical features.

9. The AI core drilling identification and automatic cataloging system for an outdoor exploration operation robot according to claim 1, wherein, The construction of an AI model including texture, spectral, and mechanical branches, and after feature fusion and cross-entropy loss training, it is deployed to output the class probability distribution, including the following steps: Establish a unified timestamp coordinate system for each core sample, with the trigger moment of the mechanical sensor as the reference, align the texture images and spectral data within a ±50ms time window, and perform Z-score normalization processing on the extracted texture features, spectral features, and mechanical feature dimensions; Construct the AI model architecture: The input layer includes a texture branch: input of 224*224*3 RGB images, a spectral branch: 420-dimensional spectral curve, and a mechanical branch: 6-dimensional force signal * 1000 time steps; The texture network uses an improved ResNet18 architecture, and changes the last fully connected layer to a 512-dimensional output. The spectral network uses a 1D convolutional layer and adds a spectral attention module. The mechanical network uses a bidirectional LSTM layer and adds a time attention mechanism; The feature fusion layer uses a cross-modal attention mechanism to splice the high-order features of each modality to form a 1024-dimensional joint feature vector; The weighted cross-entropy loss is selected as the loss function to train the model: in the first stage, the texture network is frozen and the spectral or mechanical branch is trained; in the second stage, all networks are unfrozen and trained end-to-end, using the Lookahead optimizer with k = 5 and α = 0.

5. The trained AI model is deployed to the actual core identification task, and the model outputs the probability distribution of each category.

10. The AI core drilling identification and automatic cataloging system for an outdoor exploration operation robot according to claim 1, wherein, After the AI model identifies the core, it automatically generates catalog information, constructs a database to store relevant information, and designs a user interface to display the core identification information, including the following steps: According to the identification results of the AI model, catalog information is automatically generated for the core. The catalog information includes the category, source, feature description, and identification date of the core, and supports customizing catalog fields and templates. Design a user interface to display the core's image, spectral data, mechanical features, and identification results, provide functions for editing and viewing catalog information, and allow users to modify and supplement the catalog information.

Images