Systems and methods for soil classification based on hyperspectral images.

Abstract

The 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 profile for the soil sample based on the received data, wherein the hyperspectral reflectance profile comprises reflectance intensity information; determining one or more soil characteristics based on the determined hyperspectral reflectance profile of the soil sample, wherein the one or more soil characteristics include one or more of: soil-type classification information for the soil sample; chemical composition information for the soil sample; particle size information for the soil sample A machine learning method for classifying soil samples is also disclosed, along with a training model, and associated systems for carrying out methods according to the disclosure. Unlocking insights from Geo- Data, the present invention further relates to improvements in sustainability and environmental developments: together we create a safe and liveable world.

Description

1 / 42 Systems and methods for soil classification based on hyperspectral images. TECHNICAL FIELD

[001] This disclosure relates to methods and systems for predicting characteristic soil information based on representative data from a hyperspectral image of a soil sample. The disclosure also provides a method for training a machine learning model to identify a characteristic of a soil sample, a method for obtaining a soil classification estimate using a machine learning model, and computer-readable systems and means configured to do the same. By unlocking information from Geographic Data, the present invention also relates to improvements in sustainability and environmental development: together, we create a safe and habitable world. BACKGROUND

[002] There is a general and ongoing need for systems and methods to determine or estimate the characteristics of collected or observed soil samples. One of the objectives of soil classification is to quantify or estimate the proportion of soil types that constitute a soil sample. Such information is useful because it can be used as a predictor or indicator of the physical properties of the soil, which are relevant to the planning and design of structures that interact with the soil environment. For example, the physical properties of soil located at a site intended for construction may impose different requirements on the foundations of a structure.

[003] In situ soil classification can be performed in geological surveys. In these surveys, specialists in the field of geology and soil mechanics can observe the visual aspect of the soil, including, for example, soil color and granularity. As this preliminary analysis depends on human perception of the soil, peer consensus is usually necessary for Petition 870250077788, dated 01 / 09 / 2025, page 13 / 73 2 / 42 provide an unbiased assessment of soil classification. In addition to the increasing logistical challenges of gathering sufficient expertise to assess soil in situ, analysis based on human perception is subject to non-quantifiable errors and biases and lacks reproducibility.

[004] Furthermore, soil classification techniques can be affected by the dryness (or other factors) of the sample, since the visual appearance of a sample can vary greatly depending on the saturation and moisture content of the sample; in situ soil classification is vulnerable to variability due to changes in environmental conditions at the sampling site.

[005] 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 with reference samples available in a laboratory. Samples can also be prepared under controlled conditions (e.g., samples can be dried in an oven) before analysis. Soil attributes, such as chemical composition and pH, can be analyzed.

[006] There is still a need for improved soil classification techniques that can classify soil types accurately and reliably, with better reproducibility and without the logistical challenges imposed by current standard techniques. ABSTRACT

[007] This disclosure provides systems and methods for improved soil classification.

[008] According to a first aspect of the disclosure, a method is provided for predicting characteristic soil information, the method comprising: receiving representative data from a hyperspectral image of a soil sample; determining a hyperspectral reflectance profile for the soil sample based on the received data, wherein the reflectance profile Petition 870250077788, dated 01 / 09 / 2025, page 14 / 73 3 / 42 Hyperspectral comprises reflectance intensity information; determining one or more soil characteristics based on the hyperspectral reflectance profile determined from the soil sample, wherein the one or more soil characteristics include one or more of the following: soil type classification information for the soil sample; chemical composition information for the soil sample; particle size information for the soil sample. The method may be a computer-implemented method. The reflectance profile may comprise a reflectance intensity spectrum across the entire hyperspectral range of interest, for example, in the VNIR range. Although the examples described in this document comprise profiles measured in the VNIR range, this disclosure is not limited to that.

[009] Optionally, determining one or more soil characteristics based on the determined hyperspectral reflectance profile of the soil sample involves comparing the determined hyperspectral reflectance profile with one or more known reflectance spectra and / or profiles. The hyperspectral reflectance profile may comprise reflectance intensity information for wavelengths in the range of: 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 example, the spectral profile comprises reflectance intensity information for wavelengths below approximately 600 nm and above approximately 800 nm, optionally below 550 nm and above 850 nm.

[010] The hyperspectral image of a soil sample can be an instantaneous hyperspectral image. In some modalities, the hyperspectral image may comprise a spectral sampling range of approximately 20 nm or less, approximately 15 nm or less, approximately 10 nm or less, or approximately 5 nm or less. The sampling range may be constant and extend over Petition 870250077788, dated 01 / 09 / 2025, page 15 / 73 4 / 42 substantially the entire wavelength range captured. Alternatively, the spectral range may vary across the wavelength range. The sampling interval may be less than 20 nm, optionally less than 10 nm, optionally less than 5 nm.

[011] The method may also include: identifying one or more regions of interest, ROIs, in the hyperspectral image; determining the hyperspectral reflectance profile for one or more ROIs.

[012] A plurality of ROIs can be identified, and the determination of the hyperspectral reflectance profile of the ROIs may involve determining an average hyperspectral reflectance profile for the plurality of ROIs.

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

[014] The method may also include the step of capturing the hyperspectral image.

[015] The image can be captured from a soil sample in situ or the image is captured from a soil sample in a test environment.

[016] Soil type classification information may include a soil classification label, such as clay, silt and / or sand.

[017] The step of determining one or more soil characteristics based on the hyperspectral reflectance profile determined from the soil sample may involve the use of a machine learning model to predict soil characteristics.

[018] In another aspect of the disclosure, a system is provided Petition 870250077788, dated 01 / 09 / 2025, page 16 / 73 5 / 42 to capture hyperspectral reflectance information from a soil sample, the system comprising: a remotely operated vehicle, ROV, comprising one or more hyperspectral cameras configured to capture a hyperspectral image of a soil sample; wherein the one or more hyperspectral cameras are configured to detect a hyperspectral reflectance profile comprising reflectance intensity information for wavelengths in the range of: 400 nm to 2000 nm; 400 nm to 1000 nm, or between 400 nm to 550 nm and between 850 nm to 1000 nm.

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

[020] The system may also include a communication module configured to transmit hyperspectral image data to a remote storage module.

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

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

[023] In this way, the machine learning model is trained to classify reflectance spectra. Petition 870250077788, dated 01 / 09 / 2025, page 17 / 73 6 / 42 hyperspectral patterns are associated with images of soil samples in order to emit (predict) a specific soil characteristic of interest. For example, the machine learning model can be trained to emit a label indicating the primary constituent of the soil sample, e.g., clay, sand, or silt. The accuracy of the emission is evaluated against a base-truth target label, and the model parameters are adjusted to improve the prediction accuracy. Through iterative training and adjusting the model parameters in this way, the machine learning model becomes better at the classification task and can then be used to implement the soil classification methods described above and in more detail here. Therefore, this training method facilitates an automated and reliable mechanism for soil sample classification that addresses the shortcomings of existing classification and analysis methods described above.

[024] The hyperspectral reflectance profile can advantageously comprise reflectance data for wavelengths between 400 and 1000 nm. These wavelength spectra can be easily obtained using relatively inexpensive cameras. Optionally, the range of wavelengths analyzed can be extended to include more distant wavelengths in the infrared and UV ranges, which can aid in classification, although this typically requires more complex cameras.

[025] As noted above, the output of the machine learning model is a determined (predicted) characteristic of the soil sample, which can be compared to a target characteristic (base truth) to aid in training. As noted above, an example of a soil characteristic comprises a label indicating the primary (i.e., most prevalent) constituent of a specific soil sample, typically sand, silt, or clay. Said Petition 870250077788, dated 01 / 09 / 2025, page 18 / 73 7 / 42 Therefore, the training method can focus on any suitable soil characteristic that is of interest. For example, the determined and targeted characteristics of the soil sample may comprise one or more of the following: 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 characteristics can provide useful information about the structure, composition, and characteristics of the soil sample.

[026] The present inventors have identified that the effectiveness of the classifier training process described above can be improved through the use of a pre-training process, performed before the primary training of the classifier. In particular, implementing reconstruction pre-training for the machine learning model (or one or more components thereof) before using the machine learning model for classification training can aid in subsequent classification training. Consequently, the method may comprise, before providing the hyperspectral reflectance profile to the machine learning model, performing a pre-training process on one or more components of the machine learning model. The pre-training process may advantageously comprise: obtaining a target high-resolution hyperspectral reflectance profile; and obtaining a low-resolution hyperspectral reflectance profile.The low-resolution hyperspectral reflectance profile may comprise a subsampled or pooled version of the target high-resolution hyperspectral reflectance profile. The method further comprises: providing the low-resolution hyperspectral reflectance profile to one or more machine learning model components to obtain an output from one or more machine learning model components, the output comprising a high-resolution hyperspectral reflectance profile. Petition 870250077788, dated 01 / 09 / 2025, page 19 / 73 8 / 42 resolution determined; and adjust parameters of one or more components of the machine learning model to reduce an error between the determined high-resolution hyperspectral reflectance profile and the target high-resolution hyperspectral reflectance profile.

[027] In this way, the machine learning model (or one or more of its components) is pre-trained to reconstruct high-resolution hyperspectral reflectance spectra from low-resolution hyperspectral reflectance spectra. Through this pre-training, the pre-trained component(s) of the machine learning model effectively learn(s) to understand the features of hyperspectral reflectance spectra. This acquired understanding assists the machine learning model when it is subsequently subjected to classifier training related to soil sample spectra. A key advantage of reconstruction-based pretraining is that any hyperspectral reflectance spectrum can be used during this pretraining, including unlabeled hyperspectral reflectance spectra that are not related to soil sample images.These data are readily available in large volumes, in contrast to the more specific labeled hyperspectral soil spectra needed for full classifier training. Therefore, pretraining addresses the problem that hyperspectral images and associated spectra of soil samples are scarce, which can limit the possibility of effective training using only classifier training with labeled soil sample input data.

[028] It is important to highlight that, when using pretraining, it is not necessary for all components of the machine learning model to go through this process. Instead, in some implementations, only one or a few components of the model are pretrained. More generally, in other Petition 870250077788, dated 01 / 09 / 2025, page 20 / 73 9 / 42 words, a different model architecture can be used during pre-training and full classifier training.

[029] In one implementation, the machine learning model comprises an encoder. If pretraining is used, the encoder can be pretrained, for example, in an encoder-decoder architecture. After pretraining, the pretrained encoder returns to the original machine learning model architecture and is used in classifier training of the model. The use of an encoder in this way has proven to be a particularly effective classifier for hyperspectral soil sample data.

[030] In one implementation, the machine learning model may comprise an implicit neural representation network, optionally a sinusoidal implicit neural representation network (SIREN). The present inventors have identified that implicit neural representation networks (and in particular SIRENs) are particularly suitable for hyperspectral reconstruction and classification tasks. Advantageously, the machine learning model may comprise a residual network (ResNet), which the present inventors have considered particularly suitable for hyperspectral reconstruction and classification tasks.

[031] The machine learning model can be further trained with hyperspectral image data containing another marker indicative of sample moisture. This can help the machine learning model understand the impact of soil sample moisture on the hyperspectral reflectance profile.

[032] According to another aspect, a method is provided for obtaining a soil classification estimate using a machine learning model trained according to any of the machine learning methods described above, wherein the Petition 870250077788, dated 01 / 09 / 2025, page 21 / 73 The 10 / 42 method comprises: obtaining input data comprising representative data from a hyperspectral image of a soil sample; applying the input data to the machine learning model to obtain a machine learning output such as the estimate.

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

[034] The system may comprise one or more sensor systems, wherein one or more sensor systems may optionally comprise a hyperspectral camera.

[035] According to another aspect of the disclosure, one or more computer-readable means are provided comprising instructions which, when executed by one or more data processing devices, cause the one or more data processing devices to perform operations comprising the steps of any of the methods described herein.

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

[037] The disclosure will be described in more detail with reference to illustrative forms and in connection with the following drawings, in which: Figures 1a to 1c show three different soil samples; Figures 2a to 2c show the hyperspectral reflectance profile for three different soil samples; Figure 3 shows an exemplary system for capturing a hyperspectral image and determining classification information. Petition 870250077788, dated 01 / 09 / 2025, page 22 / 73 11 / 42 of soil based on the image captured according to the disclosure; Figure 4 shows a schematic of an exemplary remotely operated vehicle comprising a hyperspectral imaging device as disclosed; Figure 5 shows a flowchart of a soil classification method according to disclosure; Figure 6 schematically shows a method for training a machine learning model to identify a characteristic of a soil sample according to the disclosure; Figure 7 shows, schematically, a particular implementation of the method from Figure 6; Figure 8 shows, schematically, a pre-training method according to the information provided; Figures 9a to 9c show examples of inputs and outputs for the machine learning model according to the disclosure; Figures 10a and 10b show an exemplary network architecture that can be used during pre-training according to the disclosure; Figure 11 shows a computer system for carrying out various dissemination methods. DETAILED DESCRIPTION OF THE DRAWINGS

[038] The following detailed description is merely illustrative and is not intended to limit the application and its uses. Furthermore, there is no intention to be bound by any theory expressed or implied in the preceding technical field, history, brief summary, or the detailed description below. As used herein, the term module refers to any hardware, software, firmware, electronic control component, processing logic, and / or processing device, individually or in any combination, including, without limitation: application-specific integrated circuit (ASIC), a Petition 870250077788, dated 01 / 09 / 2025, page 23 / 73 12 / 42 electronic circuit, a processor (shared, dedicated or grouped) and memory that executes one or more software or firmware programs, a combinational logic circuit and / or other suitable components that provide the described functionality.

[039] 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 may be implemented by any number of hardware, software, and / or firmware components configured to perform the specified functions. For example, an exemplary embodiment of the present disclosure may employ various integrated circuit components, such as memory elements, digital signal processing elements, logic elements, lookup tables, or the like, which may perform a variety of functions under the control of one or more microprocessors or other control devices. Furthermore, those skilled in the art will understand that the embodiments of the present disclosure may be implemented in conjunction with any number of systems and that the systems described herein are merely exemplary embodiments of the present disclosure.

[040] For the sake of brevity, conventional techniques compared to signal processing, data transmission, signaling, control, and other functional aspects of systems (and of the individual operational components of systems) may not be described in detail here. Furthermore, the connecting lines shown in the various figures contained herein are intended to represent examples of 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 be present in an embodiment of the present disclosure.

[041] The systems and methods described herein refer, in a way Petition 870250077788, dated 01 / 09 / 2025, page 24 / 73 13 / 42 In general, systems and methods for predicting characteristic soil information, such as soil type, based on received hyperspectral data. The received data may be a hyperspectral image of a soil sample. Determining soil type (e.g., clay, silt, sand) allows soil samples to be classified into a labeled group that can provide important information about the physical properties of the soil, which can be predictive of soil behavior during and after the construction of a structure.

[042] This disclosure generally provides methods and systems for determining a soil type based on an observed sample. The systems and methods described herein provide automated (or semi-automated) soil type determination. The automated systems are configured to receive representative data from a hyperspectral image of a soil sample and to determine, by comparing the hyperspectral reflectance profile with reference spectra and / or values, a soil type for the sample based on the comparison.

[043] The systems and methods described here can also make use of a machine learning model to determine the type of soil based on received hyperspectral information, as will become apparent from the detailed description of the modalities below.

[044] Figures 1a to 1c show each of the soil samples 100a, 100b and 100c, with the enlarged portions B showing, schematically, a visually different appearance for each of the samples. The visually different characteristics of the soil samples may include color, particle shape, particle size and a combination thereof. In addition to the different visual attributes of the soil samples, each of the different samples may have a different hyperspectral profile.

[045] Figures 2a to 2c show the hyperspectral reflectance profile of a soil sample. Figures 2a to 2c show Petition 870250077788, dated 01 / 09 / 2025, page 25 / 73 14 / 42 The reflectance intensity for different wavelengths of light incident on a soil sample. In the example shown in Figures 2a to 2c, the reflectance profile covers a wavelength range from approximately 400 nm to approximately 1000 nm. In other words, the hyperspectral reflectance profile is in the visual and near-infrared (VNIR) range. However, the present disclosure is not limited to the wavelength range shown in Figures 2a to 2c. For example, the range in which reflectance can be measured can be extended to the mid-infrared or beyond. In at least one implementation, a reflectance profile can be collected in a wavelength range of 400 nm to 2500 nm, optionally a wavelength range of 400 nm to 200 nm, optionally a wavelength range of 400 to 1500 nm, and optionally a wavelength range of 400 to 100 nm.

[046] Figure 2a shows a 200a hyperspectral reflectance profile for a soil sample classified as clay. As shown in Figure 2a, the sample shows an identifiable primary reflectance peak, approximately 0.36, at a wavelength of approximately 960 nm. As shown in Figure 2a, the 200a reflectance profile for the clay sample shows a (relatively) small linear increase in reflectance from approximately 400 nm to approximately 900 nm, when a sharp increase in reflectance intensity is observed. The reflectance peaks at a wavelength of approximately 960 nm and drops sharply to approximately 1000 nm. This profile, including the shape of the profile, the gradient of reflectance across all or parts of the spectrum, the number and / or location (within the wavelength range) of identified peaks, can be characteristic of a specific soil type, in this case clay. Petition 870250077788, dated 01 / 09 / 2025, p. 26 / 73 15 / 42

[047] Figure 2b shows a 200b hyperspectral reflectance profile for a soil sample classified as sand. As shown in Figure 2b, the sample shows an identifiable reflectance peak, approximately 0.56, at a wavelength of 950 nm. As shown in Figure 2b, the 200b profile does not exhibit linear growth (as in Figure 2a). Instead, there is a steeper increase in reflectance between 400 nm and 500 nm. Between 600 nm and 900 nm there is a substantially linear increase, before the reflectance increases abruptly from approximately 900 nm. The peak at 950 nm is observed, before the reflectance drops again abruptly to 1000 nm. This profile, including the shape of the profile, the reflectance gradient across all or parts of the spectrum, the number and / or location (within the wavelength range) of the identified peaks, can be characteristic of a specific soil type, in this case sand.

[048] Figure 2c shows a 200c hyperspectral reflectance profile for a soil sample classified as silt. As shown in Figure 2b, the sample shows an identifiable reflectance peak, approximately 0.52, at approximately 950 nm. A secondary peak, at approximately 880 nm, can be observed, with a reflectance of approximately 0.44. The reflectance is generally stable between 400 nm and 450 nm, with substantially linear growth between 450 nm and 850 nm. The reflectance drops between the first and second peaks (approximately 850 nm and 950 nm) and again drops sharply from the primary peak at approximately 950 nm towards 1000 nm. This profile, including the profile shape, the reflectance gradient across all or parts of the spectrum, the number and / or location (within the wavelength range) of identified peaks, may be characteristic of a specific soil type, in this case, silt.

[049] In the spectra shown in Figures 2a to 2c, the Petition 870250077788, dated 01 / 09 / 2025, page 27 / 73 16 / 42 Reflectance intensity was measured in a spectral sampling range of 4 nm. However, it is possible to select other spectral sampling ranges. The spectral range is preferably less than 20 nm, preferably less than 15 nm, preferably less than 10 nm, and preferably less than 5 nm.

[050] The spectral range in Figures 2a to 2c is also constant throughout the measured range (between 400 nm and 100 nm). However, it should be noted that the spectral range may vary along the measured range. For example, the spectral sampling range may be larger in spectral regions that provide few characteristic peaks relevant to the soil type than in others. For example, as can be seen in the spectra from Figures 2a to 2c, identifiable spectral profile features can be found in the wavelength range below approximately 600 nm and at wavelengths above approximately 800 nm for the three soil types exemplified.Therefore, the sampling frequency can be higher (resulting in a smaller sampling interval) in spectral regions of interest, for example, below approximately 600 nm or approximately 550 nm, and above 800 or 850 nm, than in regions of less interest, such as the region between 600 nm or 800 nm or between 550 nm and 850 nm.

[051] In the discussion above, it will be observed that hyperspectral reflectance profiles for a soil sample can be indicative of soil type. In the examples provided in Figures 2a to 2c, the maximum reflectance is indicated in each profile on the y-axis. However, it will be observed that the maximum reflectance value for a sample is less relevant for characterizing the soil type than the location and relative intensity between peaks. This approach to identifying soil samples can be particularly advantageous, since the profile is largely unaltered by the moisture content of the sample. The inventors Petition 870250077788, dated 01 / 09 / 2025, page 28 / 73 17 / 42 of the present application observed that, although the visual appearance of a sample changes with the degree of water saturation of the sample (e.g., the wetter the sample, the darker it appears), and although the intensity of reflected radiation varies with humidity, the shape of the profile and the location of the peaks do not vary. Systems and methods according to the disclosure may therefore provide further improvements over traditional methods of soil type determination, which may require additional steps to

[052] Returning to Figure 3, an exemplary system for determining a hyperspectral profile will be described below. As shown in Figure 3, system 300 comprises a hyperspectral image capture device 304. The device can be configured to capture an instantaneous hyperspectral image. The hyperspectral image capture device may comprise 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 electromagnetic radiation (reflected) is recorded for a plurality of wavelengths for each pixel of the image, within the visible range and beyond. The hyperspectral image, therefore, provides spatial information (as in a conventional RGB image), in addition to information about the reflectance intensity for the plurality of wavelengths. As mentioned earlier, the range of interest may comprise the VNIR range.The hyperspectral imaging device can also be configured to measure reflectance intensity in the Shortwave Infrared (SWIR) range.

[053] Although in the example described above with reference to Figure 3, a single hyperspectral image capture device is shown, the present disclosure also includes embodiments in which a plurality of image capture devices are used. Petition 870250077788, dated 01 / 09 / 2025, page 29 / 73 18 / 42 hyperspectral scanners are used, configured to determine hyperspectral images for distinct (and optionally overlapping) wavelength ranges.

[054] As shown in Figure 3, the image capture device 304 (or plurality of image capture devices) is in operational 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 the captured image data. The captured image data may comprise a data cube, representing a two-dimensional array of pixels, with intensity information for the plurality of wavelengths provided through multiple channels for each pixel. The data cube representing the hyperspectral image may comprise the information necessary to determine one or more hyperspectral profiles for the imaged sample, as will be explained in more detail below.The image capture device's processor can be configured to perform some or all of the data processing steps described below. Alternatively, the image capture device can be configured to export the image data to an external system for processing, as illustrated in Figure 3.

[055] In modes in which a plurality of image capture devices is used to capture a plurality of overlapping or non-overlapping wavelength bands, the data may be stored in a plurality of data cubes or may be consolidated into a single data cube.

[056] It should be understood that, in the context of this disclosure, a hyperspectral profile can be determined from a single pixel of a captured hyperspectral image by determining the reflectance intensity for each interval. Petition 870250077788, dated 01 / 09 / 2025, page 30 / 73 19 / 42 wavelength across the measured wavelength range. Alternatively, a hyperspectral profile can be determined based on a plurality of pixels, for example, by averaging the reflectance intensity for a plurality of selected pixels. The selected pixels may comprise all pixels within a selected region of interest (ROI) of the hyperspectral image. A plurality of ROIs can be identified, and a spectral profile determined for each ROI, based, for example, on the average intensity of the pixels of an individual ROI. The hyperspectral profiles of each ROI can then be compared to each other to identify inconsistencies between the spectral profiles. This comparison step can be used to exclude pixels from a non-representative ROI from the hyperspectral image determined for the sample.Alternatively, hyperspectral profiles for multiple ROIs can be determined from a single hyperspectral image to identify samples of mixed types, where a first ROI is classifiable as a first soil type and a second ROI is classifiable as a second soil type.

[057] In another example, a hyperspectral profile can be determined based on all pixels of the captured hyperspectral image or on all pixels of a hyperspectral image associated with the sample. In one implementation, the image data can be processed to exclude representative data from any pixel regions with reflectance exceeding a threshold value as representative of the background or calibration surface. The hyperspectral profile information can then be determined as an average of the reflectance of all pixels not considered part of the background.

[058] In other modes, individual pixels that exceed the limit values ​​for reflectance may be excluded as not representative of the sample.

[059] In the preceding paragraphs, steps were described. Petition 870250077788, dated 01 / 09 / 2025, page 31 / 73 20 / 42 image processing examples, in which one or more hyperspectral profiles representative of a soil sample can be determined. It is important to note that the 300 system can be configured to perform some or all of these image processing steps. Alternatively, the 300 system can be configured to store the hyperspectral image data and export the raw data to an external system for processing.

[060] As shown in Figure 3, system 300 can be a stand-alone capture system, configured to capture a hyperspectral image and store the image data for subsequent processing on an external system. In such embodiments, system 300 may comprise a wired or wireless connection to an external data processing system.

[061] In the embodiment shown in Figure 3, system 300 is in operational communication with a network device 310. The network device 310 is in operational communication with a first terminal 314, which comprises a personal computer, and a second terminal 312, which comprises a mobile device. The end-user terminals may further comprise additional processing means to process the hyperspectral image data and determine a soil type based on the captured image.

[062] Although the 300 system described above is configured for wireless communication via a 310 network device, it is important to note that one or more connections may be wired.

[063] Although not shown, the 300 system may still comprise or be configured for use with a calibration surface. The calibration surface may comprise a surface with known hyperspectral reflectance properties, such as a bright white matte plate. The calibration surface may be configured to be within the field of Petition 870250077788, dated 01 / 09 / 2025, page 32 / 73 21 / 42 view of camera 304 when an image of a sample is captured.

[064] Processor 306 and / or a processor communicating with system 300 (such as in one of the terminal devices) can be configured to determine one or more regions of interest (ROIs) for the soil sample. An ROI might be, for example, near an edge of the soil sample, adjacent to a region of a calibration surface in the field of view of camera 304. Identifying an ROI by the image capture device or by a processor communicating with system 300 can be advantageous, since the volume of data required for upload to a later system for further processing steps can be reduced.

[065] Although ROIs can be automatically determined by a processor, it will be considered that a human operator can identify one or more ROIs from a hyperspectral image. It will also be considered that an ROI does not need to be a sub-region of the image, but can be the complete image of the soil sample. However, selecting one or more sub-regions of a hyperspectral image as an ROI can be advantageous in terms of reducing data volume for processing and selecting ROIs representative of the sample, avoiding anomalous material that is not representative of the wider sample.

[066] Returning now to Figure 4, in at least one exemplary embodiment, an image capture device is provided on a remotely operated vehicle (ROV). The ROV may comprise a terrestrial ROV, an aerial ROV (e.g., an unmanned aerial vehicle or UAV), or a surface or submersible ROV. The remotely operated vehicle 400 comprises a body 402, which comprises a hyperspectral image capture device 404, a processor 406, and a storage device 408. The ROV may further comprise a light source 410, for illuminating a sample, and / or a calibration surface 412 with hyperspectral reflectance properties. Petition 870250077788, dated 01 / 09 / 2025, page 33 / 73 22 / 42 known. One or more light sources 410 and the calibration surface 412 can be mounted on an articulated arm, for example, a robotic arm, which allows repositioning of the light source and the calibration surface relative to a sample.

[067] The ROV 400 may also comprise propulsion means 414, configured to move the ROV. For example, propulsion means 414 may take the form of one or more propellers for a submersible ROV. Propulsion means 414 may take the form of one or more rotating blades for an unmanned aerial vehicle.

[068] The ROV can be configured to capture a hyperspectral image of a soil sample, in situ, at 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.

[069] In the preceding paragraphs, and with reference to Figures 1 to 4, exemplary systems for capturing one or more hyperspectral images of soil samples to be classified were described.

[070] Moving on to Figure 5, a method for classifying a soil sample will be described. As shown in Figure 5, the method comprises an optional preparatory step, 500, in which a hyperspectral image of a soil sample is captured. In step 502, the method comprises receiving representative data from a hyperspectral image of a soil sample. In step 504, a hyperspectral reflectance profile is determined for the soil sample based on the received data, with the hyperspectral reflectance profile comprising reflectance intensity data for a plurality of wavelengths. In steps 506 and 508, one or more soil characteristics are determined based on the hyperspectral reflectance profile of the soil sample, comparing the Petition 870250077788, dated 01 / 09 / 2025, page 34 / 73 23 / 42 Hyperspectral reflectance profile determined from one or more known reflectance spectra. The one or more soil characteristics include one or more pieces of soil type classification information for the soil sample; chemical composition information for the soil sample; and / or particle size information for the soil sample. The soil type classification may be a soil classification label, such as clay, silt, and / or sand.

[071] The hyperspectral image of the ground can be a snapshot image. The snapshot image can comprise a data cube, in which each pixel provides intensity information for a plurality of wavelengths.

[072] The hyperspectral reflectance profile comprises reflectance data for wavelengths in the range of: 400 nm to 2000 nm; 400 nm to 1000 nm, or between 400 nm to 550 nm and between 850 nm to 1000 nm.

[073] The hyperspectral image comprises a spectral sampling range 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 range optionally extends over one or more of the above ranges.

[074] The method may also comprise, in step 503, the identification of one or more regions of interest, ROIs, in the hyperspectral image, and the determination of the hyperspectral reflectance profile in one or more ROIs. The determination of one or more ROIs may comprise the selection of an ROI adjacent to a calibration surface. In such implementations, the method further comprises the inclusion of a calibration surface or a portion thereof in the field of view of the hyperspectral imaging device when capturing an image of the sample.

[075] If multiple ROIs are identified, the method also includes determining the reflectance profile. Petition 870250077788, dated 01 / 09 / 2025, page 35 / 73 24 / 42 hyperspectral average for the ROIs.

[076] In step 502, a hyperspectral image is captured using a hyperspectral image capture device. The image capture device can be positioned at a distance D from the soil sample during the image capture step. The distance D is preferably approximately 100 m or less, approximately 25 m or less, optionally approximately 10 m or less, optionally approximately 5 m or less, optionally approximately 1 m or less.

[077] Step 502 may involve capturing an image of a sample in situ, or of a sample in a laboratory or test environment.

[078] The image acquisition step 502 may also comprise directing a light source toward a soil sample and, optionally, toward a calibration surface. An optional calibration step may comprise comparing the hyperspectral reflectance profile of the calibration surface with a measured hyperspectral reflectance profile for the soil sample. In one exemplary embodiment, calibration may be performed before each image acquisition. For example, a hyperspectral camera may be arranged so that the camera's field of view is filled by the soil sample to be imaged and a calibration surface. In one example, a calibration surface may be arranged so that it fills the camera's field of view, and the soil sample may be positioned on the calibration surface. A light source may be arranged to illuminate the sample and the calibration surface.The hyperspectral image captured by the camera can therefore provide reflectance intensity for the (regions of interest of the) soil sample and reflectance intensity for the calibration surface. This calibration approach may be particularly suitable for hyperspectral cameras. Petition 870250077788, dated 01 / 09 / 2025, page 36 / 73 25 / 42 configured to separate spectra by light decomposition (such as the Hinalea VNIR 4500). It will be appreciated that the calibration step can be omitted in some embodiments, or other calibration methods can be used.

[079] In addition to determining a soil type based on the hyperspectral reflectance profile of a soil sample, the embodiments of this disclosure can also identify additional characteristics of the soil sample using hyperspectral image analysis. Such additional characteristics may include, for example, particle size. Determining particle size can be a useful additional reference point for predicting the behavioral properties of a soil substrate during engineering and construction projects.

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

[081] As can be seen, the 500 method described above and with reference to Figure 5 may also include the use of a machine learning model to predict soil characteristics. The machine learning model, a training model and a pre-training model will now be described with reference to Figures 6 to 9.

[082] Returning now to Figure 6, a method for training a machine learning model to identify a feature of a soil sample is shown schematically. This training method can be used in the context of soil analysis described above. In other words, Petition 870250077788, dated 01 / 09 / 2025, page 37 / 73 26 / 42 A machine learning model trained using the method in Figure 6 can be used to perform the soil sample analysis described above, particularly with reference to Figure 5.

[083] The method in Figure 6 begins, in step 602, with obtaining input data on soil characteristics. This is data obtained experimentally from a set of soil samples, from which some aspect or characteristic of the soil samples (e.g., their composition) can be determined. As described above, a particularly useful form of input data on soil characteristics is a hyperspectral reflectance profile, which indicates the reflectance behavior of a given soil sample at a variety of wavelengths and which can be indicative of material properties (e.g., soil type). These hyperspectral reflectance spectra can be obtained from hyperspectral images obtained by hyperspectral cameras, as described above.

[084] In step 604, the method proceeds with obtaining target soil characteristic data. Target soil characteristic data comprises labels or indicators related to a specific characteristic of each sample in the set of soil samples from which the input data provided in step 602 were obtained. For example, target soil characteristic data might comprise a label indicating the primary constituent of each soil sample (e.g., clay, silt, sand, etc.). This target data can be used in the training method as a basis truth for each soil sample that the machine learning model is attempting to replicate.

[085] 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. Petition 870250077788, dated 01 / 09 / 2025, page 38 / 73 27 / 42 provided in step 604. In particular, input soil characteristic data for a given soil sample are provided to the machine learning model. The machine learning model then predicts, based on the input data, a characteristic of the soil sample. It is then evaluated whether the determined (predicted) soil characteristic matches the actual soil characteristic specified in the respective target soil characteristic data provided in step 604. For example, in one implementation, the machine learning network might predict whether the primary constituent in a soil sample is clay, silt, or sand. This determination is then evaluated against the basic truth label provided for that soil sample to facilitate training.

[086] Training can be performed by adjusting model parameters of the machine learning model to reduce the error between the model outputs and the target data, as is known in the technique. This process is repeated until a sufficiently well-trained model is obtained, measured by a stopping criterion, such as a convergence criterion, an error criterion, or a predefined number of epochs. Several programming languages ​​and libraries are available to implement this process using a wide variety of machine learning models. An experienced reader will realize that a suitable model, based on considerations such as available data and computation, and the type of data to be processed based on routine considerations, can be chosen. The precise details of the training process in step X06 will vary based on these implementation details, as will also be perceived by an experienced reader.The end result of the method in Figure 6 is a machine learning model trained to identify soil features based on soil input data, such as hyperspectral reflectance profiles of soil samples. A trained machine learning model. Petition 870250077788, dated 01 / 09 / 2025, page 39 / 73 28 / 42 in this way can therefore be used to implement the classification methods described above, in particular the method in Figure 5.

[087] It is important to highlight that the specific type of soil data provided as input and destination data in steps 602 and 604 will depend on the soil characteristic being studied. A specific implementation based on determining soil characteristics from hyperspectral reflectance spectra data is shown in Figure 7.

[088] The training method in Figure 7 was developed by the present inventors and provides particularly effective training to enable a machine learning model to classify soil features based on hyperspectral reflectance spectra received from soil samples. The training method in Figure 7, therefore, begins, in step 702, by obtaining hyperspectral reflectance profiles associated with a soil sample. This data is used as input data in the subsequent training process. Therefore, it is important to note that step 702 corresponds to step 602 of Figure 6.

[089] In step 704, the process obtains a target feature associated with the soil sample. This target feature may comprise soil type classification information for the soil sample, such as a label or indicator of the primary constituent (e.g., a clay label). The target feature may alternatively or additionally comprise more detailed information about the proportions of one or more soil types present in the soil sample, information about the chemical composition of the soil sample, and / or particle size information of the soil sample. The target feature is used as the target or base truth in the training process. Therefore, it is important to note that step 704 corresponds to step 604 in Figure 6. Steps 706 to 712 provide a more detailed description. Petition 870250077788, dated 01 / 09 / 2025, page 40 / 73 29 / 42 detailed description of the training process for step 606 in Figure 6.

[090] From step 706, the hyperspectral reflectance profile obtained in step 702 is fed to a machine learning model. In response to this input, an output from the machine learning model is obtained in step 708. In particular, the output of the machine learning model comprises a specific characteristic of the soil sample, which is determined based on the input data. In other words, the machine learning model receives a hyperspectral reflectance profile as input and produces as output a classification of some characteristic of the soil sample. For example, the machine learning model might output a predicted primary constituent of the soil (e.g., clay), predicted proportions of constituents (e.g., 10% clay, 90% sand), predictions of chemical properties, and so on.

[091] In step 710, an error is determined between the determined characteristic of the soil sample and the target characteristic of the soil sample. This can be done in any suitable way by comparing the output obtained in step 708 with the target data provided in step 704. In response to this determination, the method proceeds in step 712 to adjust the machine learning model parameters in order to reduce the error determined in step 710.

[092] The process in Figure Y can then be repeated for other soil samples until a stopping criterion is met. For example, the method can be repeated N times, for a predefined number of epochs, or until a convergence criterion is met. Example of implementation of Figure 7

[093] To aid understanding, a specific implementation of the training process just described will be provided with reference to Figure 7. The method in Figure 7 was Petition 870250077788, dated 01 / 09 / 2025, p. 41 / 73 30 / 42 implemented by the present inventors as follows: using a Jupyter Notebook and VS Code running PyTorch. The laptop had 32 GB of RAM with an Intel® Core™ i9-9980HK CPU at 2.40 GHz. A dataset containing 276 hyperspectral images of soil samples was obtained. Each image was 596 x 968 px. From these images, reflectance spectra for the set of soil samples were extracted and loaded into a data loader. The spectra were split into training, validation, and test data in a 60%, 20%, 20% ratio. After random shuffling of the data, approximately 15 million training spectra and 2.5 million validation and test spectra were obtained.An Adaptive Momentum Estimation (ADAM) optimizer with a weight decay of 10⁻⁵ and cross-entropy loss was used, with a batch size of 256 and a learning rate of 10⁻⁴. The network architecture comprised 299 inputs per channel, with four Fully Connected layers and Rectified Linear Unit (ReLU) activations to add non-linearity. The training objective was to predict the primary soil constituent, which could be clay, sand, or silt. Therefore, a final output layer of the network contained three outputs, one for each class, from which the maximum was considered the predicted class. After training for only a few epochs, the method converged to very small losses and high accuracies in both training and validation. The results can be seen in Table 1. Table 1: Classification accuracy of each soil class in % Sand Silt Clay 1 a -88 -80 -56

[094] One problem with implementing the Figure 7 method for soil classification is the scarcity of suitable hyperspectral images of soil samples that can be used to provide Petition 870250077788, dated 01 / 09 / 2025, page 42 / 73 31 / 42 paired classifier training data (comprising hyperspectral reflectance spectra). As noted, in the specific implementation described above, only 276 hyperspectral images were available. This is significantly less than the thousands of images typically used during the generalized learning of a neural network classifier. To address this problem, the present inventors have identified that classifier training results can be improved if a pretraining process is first used to pretrain the machine learning model (or a component thereof) through a reconstruction task in which it learns to “understand” hyperspectral images, and specifically hyperspectral reflectance spectra, better. This pretraining method is shown schematically in Figure 8.

[095] The pretraining method in Figure 8 is effectively a reconstruction task, where the machine learning model is trained to generate (or reconstruct) high-resolution target data from low-resolution input data. In this specific example, since the goal is to train the model to understand hyperspectral reflectance spectra, both 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 only indicate that the high-resolution data has a higher resolution than the low-resolution data. For example, high-resolution hyperspectral reflectance spectra might have 66 channels, while low-resolution hyperspectral reflectance spectra might have 33 or 22 channels. The example below, for simplicity, assumes that the entire machine learning model undergoes pretraining.However, as noted above, this is not essential and, in some cases, pre-training may be preferable. Petition 870250077788, dated 01 / 09 / 2025, page 43 / 73 32 / 42 only one or a few components of the machine learning model, such as an encoder.

[096] In step 802, the method begins by obtaining a high-resolution hyperspectral reflectance profile, such as by processing a hyperspectral image. In step 804, a corresponding low-resolution hyperspectral reflectance profile is obtained. This low-resolution hyperspectral reflectance profile is typically obtained by subsampling or binning the high-resolution hyperspectral reflectance profile obtained in step 802. For example, if the high-resolution profile comprises 66 channels, the low-resolution profile can be obtained by binning the profile to a lower resolution, such as 33 or 22 channels.

[097] In step 806, the low-resolution hyperspectral reflectance profile is provided to the machine learning model as input. In response to this input, an output from the machine learning model is obtained in step 808. In particular, the output of the machine learning model comprises a determined high-resolution hyperspectral reflectance profile, which the machine learning model predicts based on the input low-resolution profile.

[098] In step 810, an error is determined between the determined high-resolution hyperspectral reflectance profile and the target high-resolution hyperspectral reflectance profile. This can be done in any suitable way by comparing the output obtained in step 808 with the target data provided in step 802. In response to this determination, the method proceeds in step 812 to adjust the machine learning model parameters in order to reduce the error determined in step 810.

[099] The process in Figure 8 can then be repeated for other hyperspectral reflectance spectra until a stopping criterion is met. For example, the method can be repeated N times, for a predefined number of epochs, or until Petition 870250077788, dated 01 / 09 / 2025, page 44 / 73 33 / 42 a convergence criterion is met.

[100] A key benefit of the Figure 8 pretraining method is that the input and target data obtained in steps 802 and 804 can be any suitable hyperspectral reflectance profile data, including unlabeled data. Fundamentally, the present inventors have identified that this data does not need to be related to soil sample images to enable effective pretraining. Instead, hyperspectral reflectance spectra associated with any suitable hyperspectral images, such as satellite hyperspectral images, can be used. This data is widely available in large volumes; for example, the PRISMA dataset contains hundreds of publicly available satellite hyperspectral images. This contrasts with classifier training, which requires hyperspectral spectra from soil samples, which are scarce due to a lack of suitable hyperspectral images.Therefore, the pre-training method in Figure 8 addresses the problem of the scarcity of labeled hyperspectral soil spectra, of the type needed for traditional classifier training.

[101] The present inventors have identified that pretraining a machine learning model using the method in Figure 8 is an effective way to prepare the machine learning model (or one or more components thereof) to subsequently perform the classification training in Figure 7 and, finally, the classification task in Figure 5. Consequently, the steps in Figure 7 can follow the steps in Figure 8, and both Figures 7 and 8 can precede the steps in Figure 5. As noted above, in some examples, the entire machine learning model pretrained using the method in Figure 8 is subsequently used in the classifier training in Figure Y and the classification task in Figure 5. Alternatively, only a subset of the components of the Petition 870250077788, dated 01 / 09 / 2025, page 45 / 73 The 34 / 42 machine learning model is carried forward for classifier training and eventual classification. Example of implementation of Figure 8

[102] To facilitate understanding, a specific implementation of the newly described pre-training process will be presented below with reference to Figure 8. The method in Figure 8 was implemented by the present inventors using an Implicit Neural Representation Network. In particular, a Sinusoidal Implicit Neural Representation Network (SIREN) with Periodic Activation Functions was used to implement the pre-training. This network architecture is known in the art, but has not been used in the context of soil classification.

[103] The SIREN model was pre-trained on unlabeled hyperspectral reflectance spectra obtained from hyperspectral images. In this example, the hyperspectral images were obtained from the PRISMA dataset of satellite hyperspectral images. The PRISMA satellite is a single satellite positioned in proper Lower Earth Orbit (LEO) and Sun-Synchronous Orbit (SSO) characterized by a repeat cycle of approximately 29 days. Its payload contains an imaging spectrometer (hyperspectral camera) capable of capturing images in the VNIR and SWIR wavelength range from 400 to 2500 nm. The swath or Field of View (FOV) is 30 km or 2.77 degrees. The VNIR range camera contains 66 bands from 400 to 1010 nm. The SWIR camera contains 173 bands ranging from 920 to 2500 wavelengths — a slight overlap in the NIR range from 920 to 1010 nm wavelengths.Hyperspectral reflectance spectra associated with each image obtained from this dataset were generated.

[104] The hyperspectral reflectance spectra obtained were then grouped (reduced (“downsampled”)) from 66 channels to 33 channels to provide the low-resolution input data needed for training. The spectra of Petition 870250077788, dated 01 / 09 / 2025, page 46 / 73 Low-resolution reflectance (33 channels) data were provided to the SIREN model, and the model attempted to reconstruct the high-resolution version (66 channels). Through iteration and adjustment of model parameters, the SIREN model was gradually trained to reconstruct high-resolution hyperspectral reflectance spectra. Example inputs and outputs of this process are shown in Figures 9a-9c. Figure 9a shows a reduced low-resolution reflectance profile (33 channels) provided as input to the machine learning model during pre-training. Figure 9b shows the output of the machine learning model, which is a determined (predicted) high-resolution version of the reflectance profile. Figure 9c represents the base truth, i.e., the original real high-resolution reflectance profile before reduction.Through training, the machine learning model iteratively improves at reconstructing the high-resolution profile of the base truth from the low-resolution input profile.

[105] In the present example implementation, the training set contained ~10 million pixels and the corresponding spectra in the VNIR range. The hyperparameters used were a learning rate of 10-5, weight decay of 10-4, with an ADAM optimizer. Finally, both the Lower Bound of Evidence (ELBO) loss and a combination of Mean Squared Error (MSE) and L1 loss were considered. The batch sizes were 32 for training and validation and 16 for the test set. The network architecture was then adapted with convolutional layers, slowly increasing the number of parameters and layers, resulting in a CNN with residual layers.

[106] The network architecture used during pretraining in this example can be described in two parts, as shown in Figure 10a. The network consists of a 1002 encoder with convolutional layers and a 1004 decoder, therefore the network was analogous to an autoencoder. Pretraining Petition 870250077788, dated 01 / 09 / 2025, page 47 / 73 36 / 42 using this network architecture continued for 25 Epochs, following the steps described above in relation to Figure 8. Using a pre-trained encoder to perform classification training.

[107] As noted above, one or more components pretrained using the method in Figure 8 can subsequently be used to implement the classification training in Figure 7. This process was implemented by the present inventors as follows.

[108] After completing the pretraining as described, the pretrained encoder 1002 was implemented in a classification architecture, shown in Figure Kb. Encoder 1002 was then paired with a classification head K06, composed of fully connected layers (FC) and with three outputs (again corresponding to the soil labels clay, silt, and sand, in this case). This architecture was then used to implement a classification training regime, as described above in relation to Figure Y, to train the model to classify hyperspectral reflectance spectra into one of the three soil categories: clay, silt, and sand.

[109] The same hyperparameters used in pretraining were maintained: ADAM optimizer with a learning rate of 10-5 and weight decay of 10-4. Batch sizes of 32 for the training and validation set and 16 for the test set. Samples were split into 60% - 20% - 20% proportions for the training, validation, and test sets. In the present example, the weights in the pretrained encoder were initially frozen during classifier training, allowing the first layer and head classification to train a few iterations before unfreezing the weights and allowing the backpass to change all weights end-to-end across the network. Training continued for 20 Epochs. Petition 870250077788, dated 01 / 09 / 2025, page 48 / 73 37 / 42

[110] The present inventors identified that the pre-training process exemplified by Figure 8 improved the ability of the machine learning model to subsequently classify hyperspectral reflectance spectra in relation to soil characteristics. Specifically, the results proved quite promising in the use of reconstruction as an unsupervised pre-training prior to soil classification.

[111] The present inventors also identified that a particularly effective network architecture for use during classifier training comprises a convolutional neural network (CNN) with residual layers, also known as ResNet, where the encoder was preferably pre-trained according to the method in Figure 8 described above.

[112] With reference to Figure 11, a suitable 1100 computing device for executing the methods described above will be described below. Figure 11 shows a block diagram of an implementation of an 1100 processing system in the form of a computing device, within which a set of instructions can be implemented to cause the computing device to perform one or more of the methodologies discussed here. In alternative implementations, the computing device can be connected (e.g., networked) to other machines on a Local Area Network (LAN), an intranet, an extranet, or the Internet. The computing device can operate in the capacity of a server or a client machine in a client-server network environment, or as a peer machine in a peer-to-peer (or distributed) network environment.A computing device can be a personal computer (PC), a tablet, a set-top box (STB), a Personal Digital Assistant (PDA), a mobile phone, a web device, a server, a network router, switch or bridge, or any machine capable of executing a set of instructions (sequential or not) that specify actions to be performed. Petition 870250077788, dated 01 / 09 / 2025, page 49 / 73 38 / 42 taken by this machine. Furthermore, although only a single computing device is illustrated, the term "computing device" should also be understood as any set of machines (e.g., computers) that, individually or collectively, execute a set (or multiple sets) of instructions to perform one or more of the methodologies discussed here.

[113] The example processing system 1100 includes a processor 1102, a main memory 1104 (e.g., read-only memory (ROM), flash memory, dynamic random-access memory (DRAM), such as 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., a data storage device 1118), which communicate with each other by means of a bus 1130.

[114] The 1102 processor represents one or more general-purpose processors, such as a microprocessor, a central processing unit, or similar. More specifically, the 1102 processor 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 that implements other instruction sets, or processors that implement a combination of instruction sets. The 1102 processor 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 similar. The 1102 processor is configured to execute the processing logic (1122 instructions) to perform the operations and steps discussed herein.

[115] The 1100 processing system may also include a Petition 870250077788, dated 01 / 09 / 2025, pp. 50 / 73 39 / 42 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 touch screen), a cursor control device 1114 (e.g., a mouse or touch screen), and an audio device 1016 (e.g., a loudspeaker).

[116] It will be evident that some features of the processing system 1100 shown in Figure 10 may be absent. For example, the processing system 1100 may not require the display device 1110 (or any associated adapters). This may be the case, for example, for certain server-side computing devices, which are used only for their processing capabilities and do not need to display information to users. Similarly, user input device 1112 may not be required. In its simplest form, processing system 1100 comprises processor 1102 and main memory 1104.

[117] The data storage device 1118 may include one or more machine-readable storage media (or, more specifically, one or more computer-readable non-transient storage media) 1128 in which one or more instruction sets 1122 are stored that incorporate one or more of the methodologies or functions described herein. The instructions 1122 may also reside, wholly or at least partially, in main memory 1104 and / or in the processor 1002 during their execution by the processing system 1100, with main memory 1104 and the processor 1102 also constituting the computer-readable storage media 1128.

[118] The various methods described above can be implemented by a computer program. The program of Petition 870250077788, dated 01 / 09 / 2025, pp. 51 / 73 40 / 42 Computer code may include organized computer code to instruct a computer to perform the functions of one or more of the various methods described above. The computer program and / or code for performing such methods may be provided to an apparatus, such as a computer, in one or more computer-readable media or, more generally, in a computer program product. The computer-readable media may be transient or non-transient. The one or more computer-readable media may be, for example, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, or a propagation medium for data transmission, for example, for downloading the code over the internet.Alternatively, one or more computer-readable media may take the form of one or more computer-readable physical media, such as semiconductor or solid-state memory, magnetic tape, a removable computer floppy disk, random access memory (RAM), read-only memory (ROM), a magnetic hard disk, and an optical disk, such as a CD-ROM, CD-RW, or DVD.

[119] The computer program is executable by processor 1102 to perform the functions of the systems and methods described herein.

[120] In one implementation, the modules, components and other resources described herein may be implemented as discrete components or integrated into the functionality of hardware components, such as ASICs, FPGAs, DSPs or similar devices.

[121] A “hardware component” is a tangible (i.e., non-transient) physical component (e.g., an assembly of one or more processors) capable of performing certain operations and which can be configured or arranged in a particular physical manner. A hardware component may include dedicated circuitry or logic, permanently configured to perform certain operations. Petition 870250077788, dated 01 / 09 / 2025, pp. 52 / 73 41 / 42 A hardware component may be or include a special-purpose processor, such as an FPGA (Field Programmable Gate Array) or an ASIC. A hardware component may also include programmable logic or circuits, temporarily configured by software to perform specific operations.

[122] Thus, the term “hardware component” should be understood as encompassing a tangible entity that can be physically constructed, permanently configured (e.g., physically connected) or temporarily configured (e.g., programmed) to operate in a particular manner or to perform certain operations described herein.

[123] In addition, modules and components can be implemented as firmware or functional circuits in hardware devices. Furthermore, modules and components can be implemented in any combination of hardware devices and software components, or only in software (e.g., stored code or otherwise embedded in a machine-readable medium or in a transmission medium).

[124] Unless specifically stated otherwise, as will appear in the discussion below, it is understood that, throughout the description, discussions using terms such as receive, determine, compare, enable, maintain, identify, receive, provide or similar, refer to the actions and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (electronic) quantities within the computer system's registers and memories into other data similarly represented as physical quantities within the computer system's memories or registers or other information storage, transmission or display devices.

[125] In the examples given above, the systems and methods are configured to determine classification information of Petition 870250077788, dated 01 / 09 / 2025, pages 53 / 73 42 / 42 soil, such as soil type. However, it should be noted that the systems and methods described herein may additionally or alternatively be configured to provide information on the particle size and / or chemical composition of a soil sample. For example, in addition to determining a soil type, e.g., clay, silt, or sand, the embodiments of this disclosure may also be configured to determine a secondary descriptor, such as particle size. Furthermore, in addition to determining soil type based on a hyperspectral reflectance profile, the systems and methods of this disclosure may be configured to identify or estimate one or more chemical components of the sample.

[126] It should be understood that the above description is intended to be illustrative and not restrictive. Many other implementations will become apparent to those skilled in the art after reading and understanding the above description. Although the present disclosure has been described with reference to specific exemplary implementations, it is recognized that the disclosure is not limited to the implementations described and may be practiced with modifications and alterations within the scope of the appended claims. Consequently, the specification and drawings should be considered in an illustrative and not restrictive sense. The scope of the disclosure should therefore be determined with reference to the appended claims.

[127] Although at least one exemplary form has been presented in the detailed description above, it should be understood that there is a vast number of variations. It should also be understood that the exemplary form or forms are only examples and are not intended to limit dissemination in any way. Petition 870250077788, dated 01 / 09 / 2025, pp. 54 / 73

Claims

1 / 7 CLAIMS 1. A method for predicting soil characteristic information, the method characterized in that it comprises: receiving representative data from a hyperspectral image of a soil sample; determining a hyperspectral reflectance profile for the soil sample based on the received data, wherein the hyperspectral reflectance profile comprises reflectance intensity information; determining one or more soil characteristics based on the determined hyperspectral reflectance profile of the soil sample, wherein one or more soil characteristics include one or more of the following: soil type classification information for the soil sample; chemical composition information for the soil sample; particle size information for the soil sample.

2. A method according to claim 1, characterized in that determining one or more soil characteristics based on the determined hyperspectral reflectance profile of the soil sample comprises comparing the determined hyperspectral reflectance profile with one or more known reflectance spectra.

3. Method, according to claim 1 or 2, characterized in that the hyperspectral reflectance profile comprises reflectance intensity information for wavelengths in the range of: - 400 nm to 2000 nm; - 400 nm to 1000 nm, or - between 400 nm to 550 nm and between 850 nm to 1000 nm.

4. Method, according to claim 1, 2 or 3, characterized in that the hyperspectral image of a soil sample is an instantaneous hyperspectral image.

5. A method according to any one of claims 1 to 4, characterized in that the hyperspectral image comprises a spectral sampling range 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 range optionally extends over one or more of the ranges of claim 2.

6. A method according to any one of claims 1 to 5, characterized in that the method further comprises: identifying one or more regions of interest, ROIs, in the hyperspectral image; determining the hyperspectral reflectance profile in one or more ROIs.

7. A method according to claim 6, characterized in that a plurality of ROIs is identified, and wherein the determination of the hyperspectral reflectance profile of the ROIs comprises determining an average hyperspectral reflectance profile for the plurality of ROIs.

8. Method, according to any one of claims 1 to 7, characterized in that the hyperspectral image is captured at a distance D from the soil sample, wherein D is approximately 100 m or less, optionally approximately 25 m or less, optionally approximately 10 m or less, optionally approximately 5 m or less, optionally approximately 1 m or less.

9. A method according to any one of claims 1 to 8, characterized in that the method further comprises the step of capturing the hyperspectral image.

10. Method, according to claim 9, characterized in that the image is captured from a soil sample in situ or in that the image is captured from a soil sample in a test environment. Petition 870250077788, dated 01 / 09 / 2025, p. 56 / 73 3 / 7 11. A method according to any one of claims 1 to 10, characterized in that the soil type classification information comprises a soil classification label, such as clay, silt and / or sand.

12. A method, according to any one of claims 1 to 11, characterized in that the step of determining one or more soil characteristics based on the determined hyperspectral reflectance profile of the soil sample comprises the use of a machine learning model to predict the soil characteristics.

13. Method according to claim 12, characterized in that it further comprises the steps of: training the machine learning model to identify a feature of a soil sample as defined in any of claims 14 to 22.

14. A method for training a machine learning model to identify a feature of a soil sample, the method characterized in that it comprises: obtaining a hyperspectral reflectance profile associated with a soil sample; obtaining a target feature associated with the soil sample; providing the hyperspectral reflectance profile to a machine learning model to obtain a machine learning model output, the machine learning model output comprising a determined feature of the soil sample; and adjusting machine learning model parameters to reduce an error between the determined feature of the soil sample and the target feature associated with the soil sample.

15. Method, according to claim 14, characterized in that the hyperspectral reflectance profile comprises reflectance data for wavelengths in the range of: Petition 870250077788, dated 01 / 09 / 2025, page 57 / 73 4 / 7 - 400 nm to 2000 nm; - 400 nm to 1000 nm and between 400 nm to 550 nm and between 850 nm to 1000 nm.

16. Method, according to claim 14 or 15, characterized in that the determined and targeted characteristics of the soil sample comprise one or more of the following: soil type classification information for the soil sample; chemical composition information for the soil sample; particle size information for the soil sample.

17. A method according to any one of claims 14 to 16, characterized in that it further comprises, before providing the hyperspectral reflectance profile to the machine learning model, performing a pretraining process on one or more components of the machine learning model, the pretraining process comprising: obtaining a target high-resolution hyperspectral reflectance profile; obtaining a low-resolution hyperspectral reflectance profile; providing the low-resolution hyperspectral reflectance profile to one or more components of the machine learning model to obtain an output from one or more components of the machine learning model, the output comprising a determined high-resolution hyperspectral reflectance profile;and adjust parameters of one or more components of the machine learning model to reduce an error between the determined high-resolution hyperspectral reflectance profile and the target high-resolution hyperspectral reflectance profile.

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

19. Method of any of claims 14 to 18, Petition 870250077788, dated 01 / 09 / 2025, p. 58 / 73 5 / 7 characterized in that the machine learning model comprises an implicit neural representation network.

20. Method according to claim 19, characterized in that the machine learning model comprises a sinusoidal implicit neural representation network.

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

22. A method, according to any one of claims 14 to 21, characterized in that the machine learning model is additionally trained with hyperspectral image data comprising an additional sample moisture indicator label.

23. Method for obtaining a soil classification estimate using a trained machine learning model, according to any one of claims 14 to 22, the method characterized in that it comprises: obtaining input data comprising representative data from a hyperspectral image of a soil sample; applying the input data to the machine learning model to obtain a machine learning output as an estimate.

24. A system characterized in that it comprises: one or more processors; one or more memories having stored therein computer-readable instructions configured to cause the one or more processors to perform operations comprising the steps as defined in any of the preceding claims.

25. System according to claim 24, characterized in that it further comprises one or more sensors, wherein the one or more sensors optionally comprise Petition 870250077788, dated 01 / 09 / 2025, page 59 / 73 6 / 7 a hyperspectral camera.

26. One or more computer-readable means, characterized in that it comprises instructions which, when executed by one or more data processing devices, cause the one or more data processing devices to perform operations comprising the steps as defined in any one of claims 1 to 23.

27. Machine learning model stored on one or more computer-readable media, characterized in that the model was trained according to the method as defined in claims 14 to 22 or the method as defined in claim 13.

28. System for capturing hyperspectral reflectance information from a soil sample, the system characterized in that it comprises: a remotely operated vehicle (ROV) comprising one or more hyperspectral cameras configured to capture a hyperspectral image of a soil sample; wherein one or more hyperspectral cameras are configured to detect a hyperspectral reflectance profile comprising reflectance intensity information for wavelengths in the range of: - 400 nm to 2000 nm; - 400 nm to 1000 nm; or - between 400 nm and 550 nm and between 850 nm and 1000 nm.

29. System according to claim 28, characterized in that the ROV comprises: a light source and, optionally, a calibration surface.

30. System according to claim 28 or 29, characterized in that it further comprises: a communication module configured to transmit hyperspectral image data to a remote storage module. Petition 870250077788, dated 01 / 09 / 2025, pp. 60 / 73 7 / 7 31. System, according to any one of claims 28 to 30, characterized in that the system further comprises one or more processors configured to execute the steps as defined in any one of claims 1 to 13 or 23. Petition 870250077788, dated 01 / 09 / 2025, pp. 61 / 73

Images