Soil classification system and method based on hyperspectral imaging

Abstract

The present disclosure relates to a method for predicting soil characteristic information, the method comprising: receiving data representative of a hyperspectral image of a soil sample; determining a hyperspectral reflectance curve of the soil sample based on the received data, wherein the hyperspectral reflectance curve comprises reflectance intensity information; determining one or more soil characteristics based on the determined hyperspectral reflectance curve of the soil sample, wherein the one or more soil characteristics comprise one or more of the following characteristics: soil type classification information of the soil sample; chemical composition information of the soil sample; and particle size information of the soil sample. A machine learning method for classifying soil samples is also disclosed, as well as a training model and an associated system for performing the method according to the present disclosure. The invention releases insight from geographic data and also relates to improvements in sustainability and environmental development: we together create a safe and livable world.

Description

Technical Field

[0001] This disclosure relates to a method and system for predicting soil characteristic information based on hyperspectral image data representing soil samples. This disclosure also provides a method for training a machine learning model to identify soil sample characteristics, a method for using the machine learning model to obtain soil classification estimates, and a system and computer-readable medium configured to perform such operations. This disclosure unlocks insights from geographic data and also relates to improvements in sustainability and environmental development: together, we create a safe and livable world. Background Art

[0002] There is a widespread need for systems and methods for determining or estimating the characteristics of collected or observed soil samples. One goal of soil classification is to quantify or estimate the proportions of soil types that make up a soil sample. This information is useful because it can serve as a predictor or indicator of the soil's physical properties, which are relevant to the planning and design of structures that interact with the soil environment. For example, the physical characteristics of the soil at a given construction site may impose different requirements on the structure's foundation.

[0003] In situ soil classification can be performed during geological surveys. During such surveys, experts in geology and soil mechanics observe the soil's appearance, including color and particle size. Because this preliminary analysis relies on human perception of the soil, consensus among peers is often required to provide an unbiased assessment of the soil classification. In addition to the increasing logistical challenges of assembling sufficient expertise for in situ soil assessments, analyses based on human perception are prone to unquantifiable errors and biases and lack reproducibility.

[0004] Furthermore, soil classification techniques may be affected by the dryness of the sample (or other factors), as the visual appearance of the sample may vary significantly depending on the saturation and moisture content of the sample, and in situ soil classification is susceptible to fluctuations due to changing environmental conditions at the sample site.

[0005] In addition to in situ soil classification, soil samples can be collected and returned to a laboratory testing environment for analysis. For example, soil samples can be evaluated in a controlled environment and compared to reference samples available in the laboratory. Samples can also be prepared under controlled conditions prior to analysis (for example, they can be dried in an oven). Soil properties, such as chemical composition and pH, can be analyzed.

[0006] There remains a need for improved soil classification techniques that can accurately and reliably classify soil types, have improved repeatability, and do not present the logistical challenges associated with current standard techniques. Summary of the Invention

[0007] The present disclosure provides a system and method for achieving improved soil classification.

[0008] According to a first aspect of the present disclosure, a method for predicting soil characteristic information is provided, the method comprising: receiving data representing a hyperspectral image of a soil sample; determining a hyperspectral reflectance curve of the soil sample based on the received data, wherein the hyperspectral reflectance curve includes reflectance intensity information; and determining one or more soil characteristics based on the determined hyperspectral reflectance curve of the soil sample, wherein the one or more soil characteristics include one or more of the following: soil type classification information of the soil sample; chemical composition information of the soil sample; and particle size information of the soil sample. The method may be a computer-implemented method. The reflectance curve may include a reflectance intensity spectrum within a hyperspectral range of interest, such as within the VNIR range. While the examples described herein include curves measured within the VNIR range, the present disclosure is not limited thereto.

[0009] Optionally, determining one or more soil characteristics based on the determined hyperspectral reflectance curve of the soil sample comprises comparing the determined hyperspectral reflectance curve to one or more known reflectance spectra and / or curves.

[0010] The hyperspectral reflectance curve may include reflectance intensity information for wavelengths within the following ranges: 400 nm to 2000 nm; 400 nm to 1000 nm, or between 400 nm and 550 nm and between 850 nm and 1000 nm. In at least one embodiment, the spectral curve includes reflectance intensity information for wavelengths below approximately 600 nm and above approximately 800 nm, optionally below 550 nm and above 850 nm.

[0011] The hyperspectral image of the soil sample may be a snapshot hyperspectral image.

[0012] In some embodiments, the hyperspectral image can include a spectral sampling interval of approximately 20 nm or less, approximately 15 nm or less, approximately 10 nm or less, or approximately 5 nm or less. The sampling interval can be constant and substantially continuous across the entire captured wavelength range. Alternatively, the spectral interval can vary across the entire wavelength range. The sampling interval can be less than 20 nm, optionally less than 10 nm, and optionally less than 5 nm.

[0013] The method may further include: identifying one or more regions of interest (ROIs) in the hyperspectral image; and determining hyperspectral reflectance curves of the one or more ROIs.

[0014] A plurality of ROIs may be identified, and determining the hyperspectral reflectance profiles of the ROIs may include determining an average hyperspectral reflectance profile of the plurality of ROIs.

[0015] The hyperspectral image is captured at a distance D from the soil sample, and wherein D is about 100 m or less, optionally about 25 m or less, optionally about 10 m or less, optionally about 5 m or less, optionally about 1 m or less.

[0016] The method may further comprise the step of capturing a hyperspectral image.

[0017] The image may be captured in situ from the soil sample, or wherein the image is captured from the soil sample in a test environment.

[0018] The soil type classification information may include soil classification labels, such as clay, silt, and / or sand.

[0019] The step of determining one or more soil characteristics based on the determined hyperspectral reflectance curve of the soil sample may include predicting the soil characteristics using a machine learning model.

[0020] In another aspect of the present disclosure, a system for capturing hyperspectral reflectance information of a soil sample is provided, the system comprising: a remotely operated vehicle (ROV) comprising one or more hyperspectral cameras configured to capture hyperspectral images of the soil sample; wherein the one or more hyperspectral cameras are configured to detect a hyperspectral reflectance curve comprising reflectance intensity information of wavelengths within the following ranges: 400 nm to 2000 nm; 400 nm to 1000 nm, or between 400 nm and 550 nm and between 850 nm and 1000 nm.

[0021] The ROV may also include a light source and, optionally, a calibration surface.

[0022] The system may further include a communication module configured to transmit the hyperspectral image data to a remote storage module.

[0023] The system may also include one or more processors configured to perform the steps of any of the methods described herein.

[0024] According to another aspect, a method for training a machine learning model to identify characteristics of a soil sample is provided, the method comprising: obtaining a hyperspectral reflectance curve associated with the soil sample; obtaining a target characteristic associated with the soil sample; providing the hyperspectral reflectance curve to the machine learning model to obtain an output of the machine learning model, the output of the machine learning model comprising the determined characteristics of the soil sample; and adjusting parameters of the machine learning model to reduce an error between the determined characteristics of the soil sample and the target characteristics of the soil sample.

[0025] In this way, the machine learning model is trained to classify the hyperspectral reflectance spectra associated with images of soil samples, thereby outputting (predicting) specific soil characteristics of interest. For example, the machine learning model can be trained to output labels indicating the main components of the soil sample, such as clay, sand, or silt. The accuracy of the output is evaluated based on the baseline true value target labels, and the model parameters are adjusted to improve the accuracy of the predictions. By iteratively training and adjusting the model parameters in this manner, the machine learning model becomes better at the classification task and can then be used to implement the soil classification method described above and described in more detail herein. Therefore, this training method is conducive to establishing an automated, reliable soil sample classification mechanism that overcomes the shortcomings of the above-mentioned existing classification and analysis methods.

[0026] The hyperspectral reflectance curve can advantageously include reflectance data for wavelengths between 400 and 1000 nm. Spectra of this wavelength are readily acquired using relatively inexpensive cameras. Alternatively, the analyzed wavelength range can be expanded to include wavelengths in the infrared and ultraviolet ranges, which can aid in classification, although this typically requires a more complex camera.

[0027] As described above, the output of the machine learning model is a determined (predicted) soil sample feature, which can be compared to a (ground truth) target feature to assist in training. As described above, an example of a soil feature includes a label indicating the major (i.e., most prevalent) component of a particular soil sample, which is typically sand, silt, or clay. That is, the training method can focus on any appropriate soil feature of interest. For example, the determined soil sample feature and the target soil sample feature can include one or more of the following features: soil type classification information for the soil sample; chemical composition information for the soil sample; and / or particle size information for the soil sample. Analysis of each of these features can provide useful insights into the structure, composition, and characteristics of the soil sample.

[0028] The inventors have determined that the effectiveness of the above-mentioned classifier training process can be improved by using a pre-training process performed before the main classifier training. Specifically, before the machine learning model is used for classification training, performing reconstruction pre-training on the machine learning model (or one or more components thereof) can facilitate subsequent classification training. Therefore, the method may include performing a pre-training process on one or more components of the machine learning model before providing the hyperspectral reflectance curve to the machine learning model. The pre-training process may advantageously include: obtaining a target high-resolution hyperspectral reflectance curve; and obtaining a low-resolution hyperspectral reflectance curve. The low-resolution hyperspectral reflectance curve may include a downsampled or binned (or binned) version of the target high-resolution hyperspectral reflectance curve. The method also includes: providing the low-resolution hyperspectral reflectance curve to one or more components of the machine learning model to obtain an output of one or more components of the machine learning model, the output including a determined high-resolution hyperspectral reflectance curve; and adjusting the parameters of one or more components of the machine learning model to reduce the error between the determined high-resolution hyperspectral reflectance curve and the target high-resolution hyperspectral reflectance curve.

[0029] In this way, the machine learning model (or one or more components thereof) is pre-trained to reconstruct high-resolution hyperspectral reflectance spectra from low-resolution hyperspectral reflectance spectra. Through this pre-training, the pre-trained machine learning model components effectively learn to "understand" the characteristics of the hyperspectral reflectance spectra. Thereafter, when the machine learning model is subsequently subjected to classifier training associated with soil sample spectra, this acquired understanding assists the machine learning model. A key advantage of reconstruction-based pre-training is that any hyperspectral reflectance spectrum can be used during this pre-training, including unlabeled hyperspectral reflectance spectra that are not associated with images of soil samples. Such data is readily available in large quantities compared to the more specific labeled hyperspectral soil spectra required for comprehensive classifier training. Therefore, pre-training solves the problem of scarcity of hyperspectral images and associated spectra of soil samples, which may limit the possibility of effective training of classifiers using only labeled soil sample input data.

[0030] It should be understood that when pre-training is used, it is not necessary for all components of the machine learning model to undergo pre-training. Instead, in some embodiments, only one or some components of the model are pre-trained. More generally, in other words, different model architectures can be used during pre-training and full classifier training.

[0031] In one embodiment, the machine learning model includes an encoder. If pre-training is used, the encoder can be pre-trained, for example, in an encoder-decoder architecture. After pre-training, the pre-trained encoder is returned to the original machine learning model architecture and used in classifier training for the model. It has been found that using the encoder in this manner produces a particularly effective classifier for hyperspectral soil sample data.

[0032] In one embodiment, the machine learning model may include an implicit neural representation network, optionally including a sinusoidal implicit neural representation network (SIREN). The inventors have determined that implicit neural representation networks, and in particular SIREN, are particularly well suited for hyperspectral reconstruction and classification tasks. Advantageously, the machine learning model may include a residual network (ResNet), which the inventors have found to be particularly well suited for hyperspectral reconstruction and classification tasks.

[0033] The machine learning model can also be trained with hyperspectral image data containing one or more labels indicating sample moisture. This can help the machine learning model understand the impact of soil sample moisture on the hyperspectral reflectance curve.

[0034] According to another aspect, a method for obtaining an estimation result of soil classification using a machine learning model trained according to any of the above-mentioned machine learning methods is provided, wherein the method includes: obtaining input data, the input data including data representing a hyperspectral image of a soil sample; applying the input data to the machine learning model to obtain an output of the machine learning as an estimation result.

[0035] According to another aspect of the present disclosure, a system is provided, comprising: one or more processors; and one or more memories having computer-readable instructions stored thereon, the computer-readable instructions being configured to cause the one or more processors to perform operations including the steps of any method described herein.

[0036] The system may include one or more sensor systems, wherein the one or more sensor systems optionally include a hyperspectral camera.

[0037] According to another aspect of the present disclosure, one or more computer-readable media containing instructions that, when executed by one or more data processing devices, cause the one or more data processing devices to perform operations including the steps of any of the methods described herein are provided.

[0038] According to another aspect of the present disclosure, there is provided a machine learning model stored on one or more computer-readable media, wherein the model has been trained according to one or more training methods described herein. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The present disclosure will be further described below with reference to exemplary embodiments and in conjunction with the following drawings, in which:

[0040] Figures 1a to 1c Three different soil samples are shown;

[0041] Figures 2a to 2c Hyperspectral reflectance curves of three different soil samples are shown;

[0042] Figure 3 An exemplary system for capturing hyperspectral images and determining soil classification information based on the captured images according to the present disclosure is shown;

[0043] Figure 4 shows a schematic diagram of an exemplary remote-controlled vehicle including a hyperspectral image capture device according to the present disclosure;

[0044] Figure 5 A flow chart showing a soil classification method according to the present invention is shown;

[0045] Figure 6 A method for training a machine learning model to identify soil sample characteristics according to the present disclosure is shown in schematic form;

[0046] Figure 7 Shown in schematic form Figure 6 a specific embodiment of the method;

[0047] Figure 8 The pre-training method according to the present disclosure is shown in the form of a schematic diagram;

[0048] Figures 9a to 9c shows exemplary inputs and outputs of a machine learning model according to the present disclosure;

[0049] Figure 10a and 10b An exemplary network architecture that may be used during pre-training according to the present disclosure is shown;

[0050] Figure 11 A computer system is shown for executing the various methods of the present disclosure. DETAILED DESCRIPTION

[0051] The following detailed description is merely exemplary in nature and is not intended to limit the present application and its uses. Furthermore, no one should be bound by any theory, expressed or implied, presented in the preceding technical field, background technology, summary of the invention, or the following detailed description. As used herein, the term "module" refers to any hardware, software, firmware, electronic control component, processing logic and / or processor device in any independent form or combination, including but not limited to: application specific integrated circuits (ASICs), electronic circuits, processors (shared, dedicated or grouped) and memories that execute one or more software or firmware programs, combinational logic circuits and / or other appropriate components that provide the described functionality.

[0052] The embodiments of the present disclosure may be described herein in terms of functional and / or logical block components and various processing steps. It should be understood that such block components can be implemented by any number of hardware, software and / or firmware components configured to perform the specified functions. For example, exemplary embodiments of the present disclosure may employ various integrated circuit components, such as memory elements, digital signal processing elements, logic elements, lookup tables, etc., which can perform various functions under the control of one or more microprocessors or other control devices. In addition, it should be understood by those skilled in the art that the embodiments of the present disclosure may be practiced in conjunction with any number of systems, and the systems described herein are merely exemplary embodiments of the present disclosure.

[0053] For the sake of brevity, conventional techniques compared to signal processing, data transmission, signaling, control, and other functional aspects of the system (and the various operating components of the system) may not be described in detail herein. In addition, the connecting lines shown in the various figures included herein are intended to represent exemplary functional relationships and / or physical couplings between the various elements. It should be noted that many alternative or additional functional relationships or physical connections may exist in embodiments of the present disclosure.

[0054] The systems and methods described herein generally relate to a system and method for predicting soil characteristic information, such as soil type, based on received hyperspectral data. The received data may be a hyperspectral image of a soil sample. Determining the soil type (e.g., clay, silt, sand) can classify the soil sample into a set of markers that can provide important insights into the soil's physical properties, which can be used to predict how the soil will behave during and after the construction of a structure.

[0055] The present disclosure generally provides a method and system for determining soil type based on an observed sample. The system and method described herein implement automatic (or semi-automatic) determination of soil type. The automated system is configured to receive data representing a hyperspectral image of a soil sample and determine the soil type of the sample by comparing a hyperspectral reflectance curve with a reference spectrum and / or value based on the comparison.

[0056] As will become apparent from the following description of detailed embodiments, the systems and methods described herein may also utilize machine learning models to determine soil type based on received hyperspectral information.

[0057] Figures 1a to 1c Soil samples 100a, 100b, and 100c are shown, with enlarged portion B schematically illustrating the visually distinct appearance of each sample. Visually distinct characteristics of the soil samples may include color, particle shape, particle size, and combinations thereof. In addition to the different visual attributes of the soil samples, each different sample may also have a different hyperspectral profile.

[0058] Figures 2a to 2c The hyperspectral reflectance curves of soil samples are shown respectively. Figures 2a to 2c The reflectance intensity of the incident light of different wavelengths on the soil sample is shown. Figures 2a to 2c In the example shown, the reflectivity curve spans a wavelength range of about 400 nm to about 1000 nm. In other words, the hyperspectral reflectivity curve is within the visible and near infrared range (VNIR). However, the present disclosure is not limited to Figures 2a to 2c For example, the range in which reflectivity can be measured can extend to the mid-infrared or beyond. In at least one embodiment, reflectivity curves can be collected over a wavelength range of 400 nm to 2500 nm, optionally over a wavelength range of 400 nm to 200 nm, optionally over a wavelength range of 400 nm to 1500 nm, and optionally over a wavelength range of 400 nm to 100 nm.

[0059] Figure 2a FIG. 2 shows a hyperspectral reflectance curve 200a of a soil sample classified as clay. Figure 2a As shown in FIG, the sample exhibits an identifiable main reflectivity peak at a wavelength of about 960 nm, which is about 0.36. Figure 2a As shown, the reflectance curve 200a of the clay sample exhibits a (relatively) small linear reflectance increase from about 400 nm to about 900 nm, when a sharp increase in reflectance intensity is observed. The reflectance peaks at a wavelength of about 960 nm and decreases sharply until about 1000 nm. This curve, including its shape, the reflectance gradient of all or part of the spectrum, and the number and / or location of identified peaks (within the wavelength range), can be characteristic of a particular soil type, which in this case is clay.

[0060] Figure 2b FIG. 2 shows a hyperspectral reflectance curve 200b of a soil sample classified as sand. Figure 2bAs shown in Figure 2, the sample exhibits a recognizable reflectivity peak at a wavelength of 950 nm, which is approximately 0.56. Figure 2b As shown, curve 200b does not show a linear growth (like Figure 2a In contrast, there is a steeper increase in reflectivity from 400 nm to 500 nm. There is a substantially linear increase from 600 nm to 900 nm, before a sharp increase in reflectivity from about 900 nm. A peak is observed at 950 nm, after which the reflectivity drops sharply again towards 1000 nm. This curve, including its shape, the reflectivity gradient of all or part of the spectrum, and the number and / or location of identified peaks (within the wavelength range), can be characteristic of a particular soil type, in this case sand.

[0061] Figure 2c A hyperspectral reflectance curve 200c of a soil sample classified as silt is shown. Figure 2b As shown, the sample shows an identifiable reflectivity peak at about 950nm, which is about 0.52. A secondary peak can be observed at about 880nm, and the reflectivity is about 0.44. Reflectivity is roughly stable at 400nm to 450nm, and increases linearly basically at 450nm to 850nm. Reflectivity declines between the first peak and the second peak (at about 850nm and 950nm), and sharply declines again towards 1000nm from the main peak at about 950nm. This curve, including the shape of the curve, the reflectivity gradient of all or part of the spectrum, the quantity and / or the position (in wavelength range) of the peaks identified, can be the characteristics of a specific soil type, which is a silt in this case.

[0062] exist Figures 2a to 2c In the spectra shown, the reflectivity intensity is measured at a spectral sampling interval of 4 nm. However, it should be understood that other spectral sampling ranges can be selected. The spectral interval is preferably less than 20 nm, preferably less than 15 nm, preferably less than 10 nm, and preferably less than 5 nm.

[0063] Figures 2a to 2c The spectral interval in is also constant over the entire measurement range (400 nm to 100 nm). However, it should be understood that the spectral interval can vary within the measurement range. For example, in a spectral region with fewer characteristic peaks associated with soil types, the spectral sampling interval may be larger than in other regions. For example, from Figures 2a to 2cAs can be seen in the spectra, for the three exemplary soil types shown, identifiable spectral curve features can be found in the wavelength range less than about 600 nm and at wavelengths greater than about 800 nm. Therefore, the sampling frequency can be higher (resulting in a smaller sampling interval) in spectral regions of interest, such as those below about 600 nm or about 550 nm and above 800 nm or 850 nm, than in less interesting regions, such as those between 600 nm and 800 nm or between 550 nm and 850 nm.

[0064] It can be understood from the above discussion that the hyperspectral reflectance curve of a soil sample can indicate the soil type. Figures 2a to 2c In the examples provided, the maximum reflectance is shown in each curve on the y-axis. However, it will be understood that the maximum reflectance value of a sample is less relevant to the characterization of soil type than the location of the peaks and the relative intensity between the peaks. This method of identifying soil samples may be particularly advantageous because the moisture content of the sample does not affect the curve to a large extent. The inventors of the present application have observed that although the visual appearance of the sample varies with the water saturation of the sample (e.g., the wetter the sample, the darker it appears), and although the intensity of the reflected radiation varies with humidity, the shape of the curve and the location of the peaks remain unchanged. Therefore, the systems and methods according to the present disclosure can provide additional improvements over traditional methods of determining soil type, which may require additional steps.

[0065] Please turn to Figure 3 , an exemplary system for determining a hyperspectral curve will now be described. Figure 3 As shown, system 300 includes a hyperspectral image capture device 304. The device can be configured to capture snapshot hyperspectral images. The hyperspectral image capture device can include a hyperspectral camera, such as the VNIR 4250 available from Hinalea Imaging. The image capture device is configured to capture an image in which the intensity of (reflected) electromagnetic radiation of multiple wavelengths within and outside the visible light range is recorded for each image pixel. Therefore, in addition to the reflectivity intensity information of multiple wavelengths, the hyperspectral image also provides spatial information (as in a conventional RBG image). As mentioned above, the range of interest can include the VNIR range. The hyperspectral image capture device can also be configured to measure the reflectivity intensity of the shortwave infrared range (SWIR).

[0066] Although the above reference Figure 3 The illustrated examples show a single hyperspectral image capture device, but the present disclosure also includes embodiments using multiple hyperspectral image capture devices configured to determine hyperspectral images of different (and optionally overlapping) wavelength ranges.

[0067] like Figure 3 As shown, an image capture device 304 (or multiple image capture devices) is in operative communication with a processor 306 and a data storage device 308. The processor 306 can be configured to control the image capture device. The data storage device is configured to store captured image data. The captured image data may include a data cube representing a two-dimensional array of pixels, the data cube having intensity information at multiple wavelengths provided by multiple channels for each pixel. The data cube representing the hyperspectral image may contain information required to determine one or more hyperspectral curves of the sample being imaged, which will be explained in further detail below. The processor of the image capture device can be configured to perform some or all of the data processing steps described below. Alternatively, as Figure 3 As shown, the image capture device can be configured to output image data to an external system for processing.

[0068] In embodiments where multiple image capture devices are used to capture multiple overlapping or non-overlapping wavelength ranges, the data may be stored in multiple data cubes, or may be merged into a single data cube.

[0069] It will be appreciated that in the context of the present disclosure, a hyperspectral profile can be determined from a single pixel of a captured hyperspectral image by determining the reflectance intensity for each wavelength interval within the measured wavelength range. Alternatively, a hyperspectral profile can be determined based on a plurality of pixels, for example by averaging the reflectance intensities of a plurality of selected pixels. The selected pixels may include all pixels of a region of interest (ROI) having a selected hyperspectral image. A plurality of ROIs may be identified, and a spectral profile determined for each ROI based on, for example, the average intensity of the pixels of the respective ROIs. The hyperspectral profiles of each ROI may then be compared to one another to identify inconsistencies between the spectral profiles. Such a comparison step may be used to exclude pixels of non-representative ROIs from a hyperspectral image determined for a sample. Alternatively, hyperspectral profiles for a plurality of ROIs may be determined from a single hyperspectral image to identify a mixed-type sample, wherein a first ROI may be classified as a first soil type and a second ROI may be classified as a second soil type.

[0070] In yet another example, a hyperspectral curve can be determined based on all pixels of a captured hyperspectral image or based on all pixels of a hyperspectral image associated with a sample. In one embodiment, the image data can be processed to exclude data representing any pixel regions having a reflectance exceeding a threshold value (as representative of a background or calibration surface). The hyperspectral curve information can then be determined as the average of the reflectances of all pixels that are not considered to form part of the background.

[0071] In other embodiments, each pixel exceeding a reflectivity threshold may be excluded as a non-representative pixel of the sample.

[0072] In the preceding paragraphs, exemplary image processing steps were described in which one or more hyperspectral curves representing a soil sample can be determined. It should be understood that system 300 can be configured to perform some or all of these image processing steps. Alternatively, system 300 can be configured to store hyperspectral image data and output the raw data to an external system for processing.

[0073] like Figure 3 As shown, system 300 may be a standalone capture system configured to capture hyperspectral images and store the image data for subsequent processing in an external system. In such an embodiment, system 300 may include a wired or wireless connection to the external data processing system.

[0074] exist Figure 3 In the illustrated embodiment, the system 300 is in operative communication with a network device 310. The network device 310 is in operative communication with a first terminal 314, which may include a personal computer, and a second terminal 312, which may include a mobile device. The end-user terminal may also include additional processing means for processing the hyperspectral image data to determine soil type based on the captured image.

[0075] While the system 300 described above is configured for wireless communication via the network device 310, it should be understood that one or more of the connections may be wired.

[0076] Although not shown, the system 300 may also include a calibration surface or be configured for use with a calibration surface. The calibration surface may include a surface having known high spectral reflectance properties, such as a bright white matte plate. The calibration surface may be configured to be within the field of view of the camera 304 when capturing an image of the sample.

[0077] The processor 306 and / or a processor in communication with the system 300 (e.g., at one of the terminal devices) can be configured to determine one or more regions of interest (ROIs) of the soil sample. The ROIs can be, for example, regions toward the edge of the soil sample, adjacent to the calibration surface in the field of view of the camera 304. Identifying the ROIs by the image capture device or a processor in communication with the system 300 can be advantageous because it can reduce the amount of data required to be uploaded to another system to perform additional processing steps.

[0078] While the ROI can be automatically determined by the processor, it should be understood that a human operator can identify one or more ROIs from the hyperspectral image. It should also be understood that the ROI does not necessarily have to be a subregion of the image, but can also be the entire image of the soil sample. However, selecting one or more subregions of the hyperspectral image as the ROI may be advantageous in reducing the amount of data required for processing and may also be advantageous in selecting a ROI that is representative of the sample and avoiding anomalous material that is not representative of the broader sample.

[0079] Now go to Figure 4 In at least one exemplary embodiment, an image capture device is provided on a remotely operated vehicle (ROV). The ROV may include a land-based ROV, an aerial ROV (e.g., an unmanned aerial vehicle or UAV), or a surface or underwater ROV. The remotely operated vehicle 400 includes a body 402 that includes a hyperspectral image capture device 404, a processor 406, and a storage device 408. The ROV may also include a light source 410 for illuminating the sample and / or a calibration surface 412 having known hyperspectral reflectance characteristics. One or more of the light source 410 and the calibration surface 412 may be mounted on a retractable arm that allows the light source and the calibration surface to be repositioned relative to the sample, such as on a robotic arm.

[0080] ROV 400 may also include a propulsion device 414 configured to move the ROV. For example, the propulsion device 414 may take the form of one or more propellers for an underwater ROV. The propulsion device 414 may take the form of one or more rotating blades for a drone.

[0081] The ROV may be configured to capture hyperspectral images of soil samples in situ from a distance of less than 100 m, optionally less than 50 m, optionally less than 25 m, optionally less than 10 m, optionally less than 5 m, and optionally approximately 1 m or less.

[0082] In the preceding paragraphs, referring to Figures 1 to Figure 4 An exemplary system for capturing one or more hyperspectral images of a soil sample to be classified is described.

[0083] Now please turn to Figure 5 , the method of classifying soil samples will now be described. Figure 5As shown, the method includes an optional preparatory step 500, in which a hyperspectral image of a soil sample is captured. In step 502, the method includes receiving data representing the hyperspectral image of the soil sample. In step 504, a hyperspectral reflectance curve of the soil sample is determined based on the received data, wherein the hyperspectral reflectance curve includes reflectance intensity data at multiple wavelengths. In steps 506 and 508, one or more soil characteristics are determined based on the hyperspectral reflectance curve of the soil sample by comparing the determined hyperspectral reflectance curve with one or more known reflectance spectra. The one or more soil characteristics include one or more soil type classification information of the soil sample; chemical composition information of the soil sample; and / or particle size information of the soil sample. The soil type classification can be a soil classification label, such as clay, silt, and / or sand.

[0084] The hyperspectral image of soil may be a snapshot image, which may include a data cube in which each pixel provides intensity information at multiple wavelengths.

[0085] The hyperspectral reflectance curve contains reflectance data for wavelengths within the following ranges: 400 nm to 2000 nm; 400 nm to 1000 nm, or between 400 nm to 550 nm and between 850 nm to 1000 nm.

[0086] The hyperspectral image comprises a spectral sampling interval of approximately 20 nm or less, approximately 15 nm or less, approximately 10 nm or less, or approximately 5 nm or less, and wherein the sampling interval optionally spans one or more of the aforementioned ranges.

[0087] The method may further include, in step 503, identifying one or more regions of interest (ROIs) in the hyperspectral image and determining hyperspectral reflectance curves in the one or more ROIs. Determining the one or more ROIs may include selecting an ROI adjacent to a calibration surface. In such an embodiment, the method further includes including the calibration surface, or a portion thereof, in a field of view of the hyperspectral image capture device when capturing an image of the sample.

[0088] Where multiple ROIs are identified, the method further comprises determining the average hyperspectral reflectance profile of these ROIs.

[0089] In step 502, a hyperspectral image is captured using a hyperspectral image capture device. During the image capture step, the image capture device can be located at a distance D from the soil sample. The distance D is preferably about 100 m or less, about 25 m or less, optionally about 10 m or less, optionally about 5 m or less, and optionally about 1 m or less.

[0090] Step 502 may include capturing an image of the sample in situ, or capturing an image of the sample in a laboratory environment or a testing environment.

[0091] Image capture step 502 may also include directing the light source toward the soil sample and, optionally, toward the calibration surface. An optional calibration step may include comparing the hyperspectral reflectance curve of the calibration surface with the hyperspectral reflectance curve of the measured soil sample. In an exemplary embodiment, calibration may be performed before each image acquisition. For example, the hyperspectral camera may be arranged so that the field of view of the camera is filled with the soil sample to be imaged and the calibration surface. In an example, the calibration surface may be arranged so that it fills the field of view of the camera, and the soil sample may be placed on the calibration surface. The light source may be arranged to illuminate the sample and the calibration surface. Therefore, the hyperspectral image captured by the camera can provide the reflectance intensity of the soil sample (region of interest) and the reflectance intensity of the calibration surface. This calibration method may be particularly suitable for a hyperspectral camera (e.g., Hinalea VNIR4500) configured to separate spectra by light decomposition. It should be understood that the calibration step may be omitted in some embodiments, or other calibration methods may be used.

[0092] In addition to determining soil type based on the hyperspectral reflectance curve of a soil sample, embodiments of the present disclosure can also use hyperspectral image analysis to identify additional characteristics of the soil sample. Such additional characteristics may include, for example, particle size. Determining particle size can be an additional useful reference point for predicting the behavior of the soil matrix during engineering design and construction projects.

[0093] Another useful characteristic of a soil sample that can be determined based on hyperspectral imagery is information about the presence of one or more chemical components in the soil sample. This information can be inferred based on, for example, an identified soil type (e.g., sand), or can additionally or alternatively be determined based on a hyperspectral profile determined from the hyperspectral imagery.

[0094] It should be understood that the above reference Figure 5 The illustrated method 500 may also include using a machine learning model to predict soil characteristics. Figure 6 Figures 9 to 9 illustrate machine learning models, training models, and pre-trained models.

[0095] Now go to Figure 6 , which schematically illustrates a method for training a machine learning model to identify features of a soil sample. This training method can be used in the context of soil analysis as described above, in other words, using Figure 6 The machine learning model trained by the method can be used to perform the above soil sample analysis, especially the reference Figure 5 .

[0096] Figure 6 The method begins at step 602, where input soil characteristic data is obtained. This is data experimentally obtained from a set of soil samples, from which certain aspects or characteristics of the soil samples (e.g., their composition) can be determined. As described above, one particularly useful form of input soil characteristic data is a hyperspectral reflectance curve, which indicates the reflectance behavior of a given soil sample across various wavelengths and can indicate material properties (e.g., soil type). As described above, such a hyperspectral reflectance spectrum can be obtained from a hyperspectral image captured by a hyperspectral camera.

[0097] In step 604, the method proceeds by obtaining target soil characteristic data. The target soil characteristic data comprises labels or indicators associated with specific characteristics of each sample in the set of soil samples from which the input data provided in step 602 was obtained. For example, the target soil characteristic data may comprise labels indicating the primary composition of each soil sample (e.g., "clay," "silt," "sand," etc.). This target data may be used in the training method as a "ground truth" for each soil sample that the machine learning model is attempting to replicate.

[0098] In step 606, the method trains a machine learning model using the input soil characteristic data provided in step 602 and the target soil characteristic data provided in step 604. In particular, the input soil characteristic data for a given soil sample is provided to the machine learning model. The machine learning model then predicts the characteristics of the soil sample based on the input data. It is then evaluated whether the determined (predicted) soil characteristics match the actual soil characteristics specified in the corresponding target soil characteristic data provided in step 604. For example, in one embodiment, the machine learning network can predict whether the main component in the soil sample is clay, silt, or sand. This determination is then evaluated based on the ground truth labels provided for the soil sample to facilitate training.

[0099] As known in the art, training can be performed by adjusting the model parameters of the machine learning model to reduce the error between the model output and the target data. Repeat this process until a good enough training model as measured by a stopping criterion is obtained, which is, for example, a convergence criterion, an error criterion, or a preset number of training rounds. There are many programming languages ​​and libraries that can be used to implement this process using a variety of machine learning models. The skilled reader will appreciate that an appropriate model can be selected based on considerations such as available data and calculations, as well as the type of data to be processed based on conventional considerations. The skilled reader will appreciate that the precise details of the training process in step X06 will vary based on these implementation details. Figure 6The end result of the method is a machine learning model that has been trained to identify soil characteristics based on soil input data, such as the hyperspectral reflectance curve of a soil sample. Therefore, the machine learning model trained in this way can be used to implement the above-mentioned classification methods, especially Figure 5 method.

[0100] It will be appreciated that the specific type of soil data provided as input and target data in steps 602 and 604 will depend on the soil characteristics being investigated. Figure 7 A specific embodiment based on determining soil characteristics from hyperspectral reflectance spectrum data is shown in FIG.

[0101] The inventors have developed Figure 7 The present invention provides a training method that provides particularly effective training for enabling a machine learning model to classify soil characteristics based on the hyperspectral reflectance spectrum of a received soil sample. Figure 7 The training method of [ ] begins with step 702 - obtaining a hyperspectral reflectance curve associated with a soil sample. This data is used as input data in subsequent training processes. It should be understood that step 702 is therefore the same as Figure 6 Corresponding to step 602.

[0102] In step 704, the process obtains a target feature associated with the soil sample. The target feature may include soil type classification information for the soil sample, such as a label or indicator of the major component (e.g., a "clay" label). The target feature may alternatively or additionally include more detailed information about the proportion of one or more types of soil present in the soil sample, chemical composition information of the soil sample, and / or particle size information of the soil sample. The target feature is used as a target or ground truth value during the training process. It should be understood that step 704 is therefore similar to Figure 6 Then, steps 706-712 provide for Figure 6 A more detailed description of the training process of step 606 is given below.

[0103] Starting from step 706, the hyperspectral reflectance curve obtained in step 702 is provided to the machine learning model. In response to the input, the output of the machine learning model is obtained in step 708. In particular, the output of the machine learning model includes a determined characteristic of the soil sample, which is determined based on the input data. In other words, the machine learning model takes the hyperspectral reflectance curve as its input and produces a classification of certain characteristics of the soil sample as its output. For example, the machine learning model can output a predicted major component of the soil (e.g., "clay"), a predicted component ratio (e.g., 10% clay, 90% sand), a chemical property prediction, etc.

[0104] In step 710, an error between the determined characteristic of the soil sample and a target characteristic of the soil sample is determined. This may be performed in any suitable manner by comparing the output obtained in step 708 with the target data provided in step 704. In response to this determination, the method adjusts parameters of the machine learning model in step 712 to reduce the error determined in step 710.

[0105] The process of Figure Y can then be repeated for more soil samples until a stopping criterion is met. For example, the method can be repeated N times or a preset number of rounds, or until a convergence criterion is met.

[0106] Figure 7 Exemplary embodiments of

[0107] To help understand, we will now provide a reference to Figure 7 The present inventors implemented the following using Jupyter notebooks and VS code running PyTorch: Figure 7 The laptop has 32GB RAM and uses an Intel(R) Core(TM) i9-9980HK CPU @ 2.40GHz computer processing unit (CPU). A dataset of 276 hyperspectral images of soil samples was obtained. Each image is 596x968 pixels. A set of reflectance spectra of soil samples were extracted from these images and loaded into the data loader. The spectra were divided into training, validation and test data with a ratio of 60%, 20%, and 20%. After randomly shuffling the data, approximately 15 million training spectra and 2.5 million validation and test spectra were obtained. Using a dataset with 10 -5 Adaptive Estimation of Moments (ADAM) optimizer with weight decay and cross entropy loss, with a batch size of 256 and a learning rate of 10 -4 The network architecture consists of 299 inputs per channel, with four fully connected layers and rectified linear unit (ReLU) activations to increase nonlinearity. The goal of training is to predict the main composition of the soil, which can be clay, sand, or silt. Therefore, the final output layer of the network contains three outputs, one for each class, and the maximum value among the outputs is taken as the "predicted class". After only a few rounds of training, the method converged to very small loss and high accuracy in both training and validation. The results can be seen in Table 1:

[0108] Table 1: Classification accuracy for each soil class in %.

[0109] sand silt clay ～88 ～80 ～56

[0110] implement Figure 7One problem with the soil classification methods described herein is that there are few suitable hyperspectral images of soil samples available to provide paired classifier training data (including hyperspectral reflectance spectra). As described above, in the specific embodiment described above, only 276 hyperspectral images were available. This is significantly fewer than the thousands of images typically used during generalization learning of a neural network classifier. To address this issue, the inventors determined that the results of classifier training could be improved if a "pre-training" process was first used to pre-train the machine learning model (or components thereof) through a reconstruction task in which the machine learning model learns to better "understand" the hyperspectral images, and in particular the hyperspectral reflectance spectra. Figure 8 This pre-training method is schematically shown in FIG.

[0111] Figure 8 The pre-training method is actually a reconstruction task in which the machine learning model is trained to generate (or "reconstruct") target high-resolution data from input low-resolution data. In this particular instance, since the goal is to train the model to "understand" hyperspectral reflectance spectra, the input and target data are hyperspectral reflectance spectra. The terms "low" and "high" should be understood as relative rather than absolute. In other words, these labels simply mean that the high-resolution data has a higher resolution than the low-resolution data. For example, a high-resolution hyperspectral reflectance spectrum may have 66 channels, while a low-resolution hyperspectral reflectance spectrum may have 33 or 22 channels. For simplicity, the examples below will assume that the entire machine learning model undergoes pre-training. However, as mentioned above, this is not required, and, in some cases, it may be preferable to pre-train only one component or some components of the machine learning model (e.g., the encoder).

[0112] In step 802, the method begins by obtaining a target high-resolution hyperspectral reflectance curve, for example, by processing a hyperspectral image. In step 804, a corresponding low-resolution hyperspectral reflectance curve is obtained. The low-resolution hyperspectral reflectance curve is typically obtained by downsampling or "binning" the high-resolution hyperspectral reflectance curve obtained in step 802. For example, if the high-resolution curve includes 66 channels, the low-resolution curve can be obtained by binning the curve down to a lower resolution, such as 33 or 22 channels.

[0113] In step 806, the low-resolution hyperspectral reflectance curve is provided as an input to the machine learning model. In response to the input, an output of the machine learning model is obtained in step 808. In particular, the output of the machine learning model includes a determined high-resolution hyperspectral reflectance curve predicted by the machine learning model based on the input low-resolution curve.

[0114] In step 810, an error between the determined high-resolution hyperspectral reflectance curve and a target high-resolution hyperspectral reflectance curve is determined. This can be performed in any suitable manner by comparing the output obtained in step 808 with the target data provided in step 802. In response to this determination, the method adjusts the parameters of the machine learning model in step 812 to reduce the error determined in step 810.

[0115] This can then be repeated for more hyperspectral reflectance spectra Figure 8 The process is repeated until a stopping criterion is met. For example, the method may be repeated N times or a preset number of rounds, or until a convergence criterion is met.

[0116] Figure 8 A key benefit of the pre-training method is that the target and input data obtained in steps 802 and 804 can be any suitable hyperspectral reflectance curve data, including unlabeled data. Crucially, the inventors have determined that this data does not need to be associated with soil sample images in order to achieve effective pre-training. Instead, hyperspectral reflectance spectra associated with any suitable hyperspectral imagery can be used, such as hyperspectral satellite imagery. Such data is widely available in large quantities, for example, the PRISMA dataset contains hundreds of publicly available hyperspectral satellite images. This is in contrast to classifier training, which requires hyperspectra of soil samples, which are in very short supply due to a lack of suitable hyperspectral imagery. Therefore, Figure 8 The pre-training method solves the problem of shortage of labeled hyperspectral soil spectra required for traditional classifier training.

[0117] The inventors determined that using Figure 8 The method of pretraining a machine learning model is to prepare a machine learning model (or one or more of its components) for subsequent execution Figure 7 Classification training and final execution Figure 5 An effective way to solve the classification task. Figure 7 The steps can be Figure 8 After the steps, and Figure 7 and Figure 8 The steps can be Figure 5 As mentioned above, in some examples, the use Figure 8 The entire machine learning model pre-trained by the method is then used to train the classifier of graph Y and Figure 5 Instead, only a subset of the components of the machine learning model are incorporated into the classifier training and final classification.

[0118] Figure 8 Exemplary embodiments of

[0119] To help understand, we will now provide a reference to Figure 8 The present inventors use an implicit neural representation network to implement a specific embodiment of the pre-training process. Figure 8 In particular, a Sinusoidal Implicit Neural Representation Network (SIREN) with a periodic activation function is used for pre-training. This network structure is known in the art, but has not yet been used in the context of soil classification.

[0120] The SIREN model was pre-trained on unlabeled hyperspectral reflectance spectra obtained from hyperspectral imagery. In this example, the hyperspectral imagery was obtained from the PRISMA hyperspectral satellite imagery dataset. The PRISMA satellite is a satellite placed in an appropriate low Earth orbit (LEO) and sun-synchronous orbit (SSO), characterized by a repetition period of approximately 29 days. Its payload contains an imaging spectrometer (hyperspectral camera) capable of capturing images in the VNIR and SWIR wavelength ranges from 400nm to 2500nm. The swath or field of view (FOV) is 30 kilometers or 2.77 degrees. The VNIR range camera contains 66 bands from 400nm to 1010nm. The SWIR camera contains 173 bands from 920nm to 2500nm, with slight overlap in the NIR range from 920nm to 1010nm wavelength. A hyperspectral reflectance spectrum associated with each image obtained from this dataset was generated.

[0121] The obtained hyperspectral reflectance spectrum is then binned (downsampled) from 66 channels to 33 channels to provide the low-resolution input data required for training. The low-resolution (33-channel) reflectance spectrum is provided to the SIREN model, which attempts to reconstruct a high-resolution (66-channel) version. By iterating and adjusting the model parameters, the SIREN model is gradually trained to reconstruct the high-resolution hyperspectral reflectance spectrum. Figures 9a-9c Example inputs and outputs of this process are shown in . Figure 9a Shown are downsampled low-resolution (33 channels) reflectivity curves provided as input to the machine learning model during pre-training. Figure 9b The output of the machine learning model is shown, which is a determined (predicted) high-resolution version of the reflectivity curve. Figure 9c represents the ground truth, i.e., the original true high-resolution reflectance curve before downsampling. Through training, the machine learning model iteratively improves in reconstructing the ground truth high-resolution curve from the low-resolution input curve.

[0122] In this exemplary embodiment, the training set contains approximately 10 million pixels and corresponding spectra in the VNIR range. The hyperparameters used are 10 -5 Learning rate, 10 -4Weight decay was applied, and the ADAM optimizer was used. Finally, the Evidence Lower Bound (ELBO) loss and a combination of mean squared error (MSE) and L1 loss were tested. The batch size for training and validation was 32, and the batch size for the test set was 16. The network architecture was then adapted to the convolutional layers, slowly evolving the number of parameters and layers, resulting in a CNN with residual layers.

[0123] like Figure 10a As shown, in this example, the network architecture used during pre-training can be divided into two parts for illustration. The network consists of an encoder 1002 with convolutional layers and a decoder 1004, so the network is similar to an autoencoder. Pre-training using this network architecture uses the Figure 8 The steps described last for 25 rounds.

[0124] Use pre-trained encoder for classification training

[0125] As mentioned above, use Figure 8 The method of pre-training one or more components can then be used to implement Figure 7 The inventors implemented the process as follows.

[0126] Once pre-training is completed in the manner described above, the pre-trained encoder 1002 is implemented in the classification architecture, as shown in Figure Kb. The encoder 1002 is now paired with a classification head K06, which consists of a fully connected (FC) layer and has three outputs (in this case, again corresponding to the soil labels "clay", "silt", and "sand"). This architecture is then used to implement the classification training mechanism described above with respect to Figure Y to train the model to classify hyperspectral reflectance spectra into one of three soil classes - clay, silt, and sand.

[0127] Keeping the same hyperparameters used in pre-training: with 10 -5 The learning rate and 10 -4 The ADAM optimizer with weight decay was used. The batch size was 32 for the training and validation sets, and 16 for the test set. The samples were split into training, validation, and test sets with a 60%-20%-20% ratio. In this example, the weights in the pre-trained encoder were initially frozen during classifier training, allowing the first layer and classification head to train for a few iterations before unfreezing the weights and allowing the backward pass to change all weights end-to-end throughout the network. Training lasted 20 epochs.

[0128] The present inventors have discovered that Figure 8The pre-training process shown improves the ability of the machine learning model to subsequently classify hyperspectral reflectance spectra based on soil characteristics. Specifically, these results show great promise for using reconstruction as an unsupervised pre-training method prior to soil classification.

[0129] The inventors have also determined that a particularly effective network structure for use during classifier training includes a convolutional neural network (CNN) with residual layers, also known as ResNet, in which the network is preferably trained according to the above-mentioned Figure 8 The encoder is pre-trained using the method.

[0130] Please refer to Figure 11 , a computing device 1100 suitable for executing the above-described method will now be described. Figure 11 A block diagram of an embodiment of a processing system 1100 in the form of a computing device is shown, in which a group of instructions for making the computing device perform any one or more of the methods discussed herein can be performed. In alternative embodiments, the computing device can be connected (e.g., networked) to other machines in a local area network (LAN), an intranet, an extranet, or the internet. The computing device can operate as a server or client in a client-server network environment, or as a peer machine in a peer-to-peer (or distributed) network environment. The computing device can be a personal computer (PC), a tablet computer, a set-top box (STB), a personal digital assistant (PDA), a cellular phone, a network device, a server, a network router, a switch, or a bridge, or any machine that can (in order or otherwise) perform a group of instructions specifying the action to be taken by the machine. In addition, although only a single computing device is shown, the term "computing device" is also understood to include any set of machines (e.g., computers) that perform one or more of the methods discussed herein, either individually or collectively.

[0131] The exemplary processing system 1100 includes a processor 1102, a main memory 1104 (e.g., read-only memory (ROM), flash memory, dynamic random access memory (DRAM) (e.g., synchronous DRAM (SDRAM) or Rambus DRAM (RDRAM)), etc.), a static memory 1106 (e.g., flash memory, static random access memory (SRAM), etc.), and a secondary memory (e.g., data storage device 1118), which communicate with each other via a bus 1130.

[0132] The processor 1102 represents one or more general-purpose processors, such as a microprocessor, a central processing unit, or the like. More specifically, the processor 1102 may be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor implementing other instruction sets, or a processor implementing a combination of instruction sets. The processor 1102 may also be one or more special-purpose processors, such as an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), a network processor, or the like. The processor 1102 is configured to execute processing logic (instructions 1122) for performing the operations and steps discussed herein.

[0133] The processing system 1100 may also include a network interface device 1008. The processing system 1100 may also include a video display unit 1110 (e.g., a liquid crystal display (LCD) or a cathode ray tube (CRT)), an alphanumeric input device 1112 (e.g., a keyboard or a touch screen), a cursor control device 1114 (e.g., a mouse or a touch screen), and an audio device 1016 (e.g., a speaker).

[0134] It will be apparent that some features of the processing system 1100 shown in FIG10 may not be present. For example, the processing system 1100 may not require a display device 1110 (or any associated adapter). This may be the case, for example, for a particular server-side computer device that utilizes only its processing power and does not need to display information to a user. Similarly, the user input device 1112 may not be required. In its simplest form, the processing system 1100 includes a processor 1102 and a main memory 1104.

[0135] The data storage device 1118 may include one or more machine-readable storage media (or more specifically, one or more non-transitory computer-readable storage media) 1128 on which one or more sets of instructions 1122 are stored, the instructions 1122 embodying any one or more of the methodologies or functions described herein. During execution of the instructions 1122 by the processing system 1100, the instructions 1122 may also reside, completely or at least partially, within the main memory 1104 and / or the processor 1102, the main memory 1104 and the processor 1102 also constituting the computer-readable storage medium 1128.

[0136] The various methods described above can be implemented by a computer program. The computer program can include computer code, which is arranged to instruct a computer to perform the function of one or more of the various methods described above. The computer program and / or code for performing this method can be provided to a device, such as a computer, on one or more computer-readable media, or more generally on a computer program product. A computer-readable medium can be temporary or non-temporary. The one or more computer-readable media can be, for example, an electronic, magnetic, optical, electromagnetic, infrared or semiconductor system or a propagation medium for data transmission, such as for downloading code via the Internet. Alternatively, the one or more computer-readable media can take the form of one or more physical computer-readable media, such as semiconductor or solid-state memory, magnetic tape, removable computer disk, random access memory (RAM), read-only memory (ROM), hard disk and optical disk, such as CD-ROM, CD-R / W or DVD.

[0137] The computer program may be executed by processor 1102 to perform the functions of the systems and methods described herein.

[0138] In one embodiment, the modules, components, and other features described herein may be implemented as discrete components or integrated into the functionality of hardware components such as ASICS, FPGAs, DSPs, or similar devices.

[0139] A "hardware component" is a tangible (e.g., non-transitory) physical component (e.g., a collection of one or more processors) that is capable of performing specific operations and may be configured or arranged in a specific physical manner. A hardware component may include dedicated circuitry or logic that is permanently configured to perform certain operations. A hardware component may be or include a dedicated processor, such as a field programmable gate array (FPGA) or an ASIC. A hardware component may also include programmable logic or circuitry that is temporarily configured by software to perform certain operations.

[0140] Thus, the phrase "hardware component" should be understood to encompass a tangible entity capable of being physically constructed, permanently configured (eg, hardwired), or temporarily configured (eg, programmed) to operate in a certain manner or perform certain operations described herein.

[0141] Furthermore, the modules and components may be implemented as firmware or functional circuits within a hardware device. Furthermore, the modules and components may be implemented in any combination of hardware devices and software components, or may be implemented solely in software (e.g., code stored or otherwise included in a machine-readable medium or transmission medium).

[0142] Unless otherwise specifically stated, it will be apparent from the following discussion that throughout this specification, discussions using terms such as "receive," "determine," "compare," "enable," "maintain," "identify," "receive," "provide," etc., should refer to actions and processes of a computer system or similar electronic computing device that manipulate and transform data represented as physical (electronic) quantities in the computer system's registers and memories into other data similarly represented as physical quantities in the computer system's memories or registers or other such information storage, transmission, or display devices.

[0143] In the examples provided above, the systems and methods are configured to determine soil classification information, such as soil type. However, it should be understood that the systems and methods described herein can additionally or alternatively be configured to provide information about the particle size and / or chemical composition of a soil sample. For example, in addition to determining soil types such as clay, silt, or sand, embodiments of the present disclosure can also be configured to determine auxiliary descriptors, such as particle size. Furthermore, in addition to determining soil type based on hyperspectral reflectance curves, the systems and methods of the present disclosure can also be configured to identify or estimate one or more chemical components of a sample.

[0144] It should be understood that the above description is merely illustrative and not restrictive. After reading and understanding the above description, many other embodiments will be apparent to those skilled in the art. Although the present disclosure is described with reference to specific exemplary embodiments, it should be recognized that the present disclosure is not limited to the illustrated embodiments, but can be implemented through modification and variation within the scope of the appended claims. Therefore, the description and drawings should be regarded as illustrative and not restrictive. Therefore, the scope of the present disclosure should be determined with reference to the appended claims.

[0145] Although at least one exemplary embodiment has been described in the above detailed description, it should be understood that there are many variations. It should also be understood that these exemplary embodiments are merely examples and are not intended to limit the present disclosure in any way.

Claims

1. A method for predicting soil characteristic information, the method comprising: receiving data representing a hyperspectral image of a soil sample; determining a hyperspectral reflectance curve for the soil sample based on the received data, wherein the hyperspectral reflectance curve includes reflectance intensity information; determining one or more soil characteristics based on the determined hyperspectral reflectance curve of the soil sample, The one or more soil characteristics include one or more of the following: Soil type classification information of soil samples; Chemical composition information of soil samples; Particle size information of soil samples.

2. The method of claim 1 , wherein determining one or more soil characteristics based on the determined hyperspectral reflectance curve of the soil sample comprises: The determined hyperspectral reflectance curve is compared to one or more known reflectance spectra.

3. The method according to claim 1 or claim 2, wherein the hyperspectral reflectance curve comprises reflectance intensity information at wavelengths within the following ranges: -400nm to 2000nm; -400nm to 1000nm, or - Between 400nm and 550nm and between 850nm and 1000nm.

4. The method according to claim 1, claim 2 or claim 3, wherein the hyperspectral image of the soil sample is a snapshot hyperspectral image.

5. The method of any one of claims 1 to 4, wherein the hyperspectral image comprises a spectral sampling interval of approximately 20 nm or less, approximately 15 nm or less, approximately 10 nm or less, or approximately 5 nm or less, and wherein the sampling interval optionally spans one or more of the ranges of claim 2.

6. The method according to any one of claims 1 to 5, wherein the method further comprises: identifying one or more regions of interest (ROIs) in the hyperspectral image; The hyperspectral reflectance curve in the one or more ROIs is determined. 7 . The method of claim 6 , wherein a plurality of ROIs are identified, and wherein determining the hyperspectral reflectance profile of a ROI comprises determining an average hyperspectral reflectance profile of the plurality of ROIs.

8. The method of any one of claims 1 to 7, wherein the hyperspectral image is captured at a distance D from the soil sample, and wherein D is about 100 m or less, optionally about 25 m or less, optionally about 10 m or less, optionally about 5 m or less, optionally about 1 m or less.

9. The method according to any one of claims 1 to 8, wherein the method further comprises the step of capturing the hyperspectral image.

10. The method of claim 9, wherein the image is captured in situ from a soil sample, or wherein the image is captured from a soil sample in a test environment.

11. The method according to any one of claims 1 to 10, wherein the soil type classification information comprises a soil classification label, such as clay, silt and / or sand.

12. The method of any one of claims 1 to 11, wherein the step of determining one or more soil characteristics based on the determined hyperspectral reflectance curve of the soil sample comprises predicting soil characteristics using a machine learning model.

13. The method according to claim 12, further comprising the steps of: The machine learning model is trained according to any one of claims 14 to 22 to identify features of a soil sample.

14. A method for training a machine learning model to identify characteristics of a soil sample, the method comprising: obtaining a hyperspectral reflectance curve associated with the soil sample; obtaining target characteristics associated with the soil sample; providing the hyperspectral reflectance curve to a machine learning model to obtain an output of the machine learning model, wherein the output of the machine learning model comprises a determined characteristic of the soil sample; and Parameters of the machine learning model are adjusted to reduce the error between the determined characteristics of the soil sample and the relevant target characteristics of the soil sample.

15. The method of claim 14, wherein the hyperspectral reflectance curve comprises reflectance data for wavelengths within the following range: -400nm to 2000nm; -400nm to 1000nm, and Between 400nm and 550nm and between 850nm and 1000nm.

16. The method of claim 14 or claim 15, wherein the determined soil sample characteristics and target soil sample characteristics include one or more of the following characteristics: Soil type classification information of soil samples; Chemical composition information of soil samples; Particle size information of soil samples.

17. The method according to any one of claims 14 to 16, further comprising, before providing the hyperspectral reflectance curve to the machine learning model, performing a pre-training process on one or more components of the machine learning model, the pre-training process comprising: Obtain target high-resolution hyperspectral reflectance curve; Obtain low-resolution high-spectral reflectance curves; providing the low-resolution hyperspectral reflectance curve to one or more components of the machine learning model to obtain an output from the one or more components of the machine learning model, the output comprising a determined high-resolution hyperspectral reflectance curve; and Parameters of one or more components of the machine learning model are adjusted to reduce an error between the determined high-resolution hyperspectral reflectance curve and the target high-resolution hyperspectral reflectance curve.

18. A method according to any one of claims 14 to 17, wherein the machine learning model comprises an encoder.

19. A method according to any one of claims 14 to 18, wherein the machine learning model comprises an implicit neural representation network.

20. The method of claim 19, wherein the machine learning model comprises a sinusoidal implicit neural representation network.

21. A method according to any one of claims 14 to 20, wherein the machine learning model comprises a residual network.

22. The method of any one of claims 14 to 21, wherein the machine learning model is further trained using hyperspectral image data comprising one or more labels indicative of sample moisture.

23. A method for obtaining an estimation result of soil classification using a machine learning model trained according to any one of claims 14 to 22, the method comprising: obtaining input data, the input data comprising data representing a hyperspectral image of a soil sample; The input data is applied to the machine learning model to obtain an output of the machine learning as an estimation result.

24. A system comprising: one or more processors; One or more memories having computer-readable instructions stored thereon, the computer-readable instructions being configured to cause the one or more processors to perform operations including the steps according to any one of the preceding claims.

25. The system of claim 24, further comprising one or more sensors, wherein the one or more sensors optionally comprise a hyperspectral camera.

26. One or more computer-readable media containing instructions that, when executed by one or more data processing apparatuses, cause the one or more data processing apparatuses to perform operations comprising the steps of any one of claims 1 to 23.

27. A machine learning model stored on one or more computer readable media, wherein the model has been trained according to the method of claims 14 to 22 or the method of claim 13.

28. A system for capturing hyperspectral reflectance information of a soil sample, the system comprising: a remotely operated vehicle (ROV) comprising one or more hyperspectral cameras configured to capture hyperspectral images of the soil sample; wherein the one or more hyperspectral cameras are configured to detect a hyperspectral reflectance curve comprising reflectance intensity information at wavelengths within the following range: -400nm to 2000nm; -400nm to 1000nm, or - Between 400nm and 550nm and between 850nm and 1000nm.

29. The system of claim 28, wherein the ROV comprises: a light source, and optionally, a calibration surface.

30. The system according to claim 28 or 29, further comprising: The communication module is configured to transmit the hyperspectral image data to the remote storage module.

31. The system of any one of claims 28 to 30, wherein the system further comprises one or more processors configured to perform the steps of any one of claims 1 to 13 or 23.

Images