Hyperspectral oil spill monitoring identification method, model training method and storage medium
By using the CatBoost algorithm to screen feature bands and construct an oil spill monitoring model in hyperspectral remote sensing technology, the problem of low oil film identification accuracy in marine oil spill monitoring by hyperspectral remote sensing technology has been solved, achieving higher identification accuracy and environmental adaptability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEBEI SAILHERO ENVIRONMENTAL PROTECTION HIGH TECH
- Filing Date
- 2023-03-10
- Publication Date
- 2026-06-05
Smart Images
Figure CN116385872B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of oil spill monitoring technology, and in particular to a hyperspectral oil spill monitoring identification method, model training method, and storage medium. Background Technology
[0002] Oil spills at sea are a major source of marine pollution. Because water is fluid, oil spills can spread rapidly if they are not detected and dealt with promptly. Therefore, understanding the location and spread trend of oil spills is crucial for monitoring and controlling them at sea.
[0003] Hyperspectral remote sensing technology, characterized by high spectral resolution, multiple bands, and rich information, is widely used for monitoring marine oil spills. Images obtained through hyperspectral remote sensing exhibit superior detail representation, better distinguishing between seawater and oil slicks, and enabling more realistic and objective monitoring of oil spill information.
[0004] However, images obtained by hyperspectral remote sensing are easily affected by the environment, resulting in lower accuracy in oil film identification. Summary of the Invention
[0005] This invention provides a hyperspectral oil spill monitoring identification method, model training method, and storage medium to solve the problem of low accuracy in current oil film identification.
[0006] In a first aspect, embodiments of the present invention provide a method for identifying hyperspectral oil spills, comprising:
[0007] Acquire the reflectance spectrum of the hyperspectral remote sensing image of the area to be identified in the monitoring area; wherein the reflectance spectrum corresponds to the reflectance image of the hyperspectral remote sensing image to be identified.
[0008] The reflectance of a preset band combination is extracted from the reflectance spectrum as the target dataset, and the target dataset is input into a pre-trained hyperspectral oil spill monitoring model to determine whether there is an oil spill area on the hyperspectral remote sensing image to be identified.
[0009] The hyperspectral oil spill monitoring model is trained on a pre-built ensemble learning algorithm model using training samples of reflectance based on a preset band combination. The preset band combination includes multiple feature bands, each of which is selected from the reflectance spectra of multiple hyperspectral remote sensing images containing oil films based on the ensemble learning algorithm.
[0010] In one possible implementation, the target dataset is input into a pre-trained hyperspectral oil spill monitoring model to determine whether an oil spill area exists on the hyperspectral remote sensing image to be identified, including:
[0011] The target dataset is input into a pre-trained hyperspectral oil spill monitoring model to perform oil spill identification processing on the target dataset;
[0012] Based on the results of oil spill identification processing, the oil spill areas in the reflectivity image are identified;
[0013] The identified portion of the reflectance image is classified and processed to determine whether an oil spill area exists on the hyperspectral image to be identified based on the results of the classification and processing; wherein, the classification and processing includes one or more of the following: classification merging, filtering or clustering.
[0014] In one possible implementation, before inputting the target dataset into the pre-trained hyperspectral oil spill monitoring model, the following steps are also included:
[0015] The reflectance of each feature band in the target dataset is re-assigned according to a preset weight to obtain the re-assigned target dataset;
[0016] The assigned target dataset is input into a pre-trained hyperspectral oil spill monitoring model to determine whether there is an oil spill area on the hyperspectral remote sensing image to be identified.
[0017] The hyperspectral oil spill monitoring model is obtained by re-assigning the reflectance of each characteristic band according to a preset rule, and training a pre-built ensemble learning algorithm model based on training samples of the reflectance of the preset band combination after re-assignment.
[0018] In one possible implementation, the ensemble learning algorithm is the CatBoost algorithm, and the hyperspectral oil spill monitoring model is a model based on the CatBoost algorithm.
[0019] The feature bands are selected from all bands of the reflectance spectrum based on the CatBoost algorithm, according to the frequency of each band appearing in the reflectance spectra of multiple hyperspectral remote sensing images containing oil films.
[0020] Secondly, embodiments of the present invention provide a training method for a hyperspectral oil spill monitoring model, comprising:
[0021] Obtain the first training sample; wherein the first training sample includes multiple hyperspectral remote sensing images containing oil films, and each hyperspectral remote sensing image is provided with corresponding identification information, which is used to identify the oil film area of each hyperspectral remote sensing image.
[0022] Obtain the reflectance spectrum corresponding to the first training sample, and select multiple feature bands from the reflectance spectrum based on the ensemble learning algorithm; among them, the multiple feature bands are combined into a preset band combination;
[0023] The reflectance of a preset band combination is extracted from the reflectance spectrum of each hyperspectral remote sensing image as the target dataset, and a second training sample corresponding to the first training sample is formed. Each of the second training samples has identification information corresponding to the first training sample.
[0024] The pre-built ensemble learning algorithm model is trained based on the second training sample and its corresponding identification information to obtain a trained hyperspectral oil spill monitoring model.
[0025] In one possible implementation, the reflectance spectrum corresponding to the first training sample is obtained, and multiple feature bands are selected from the reflectance spectrum based on an ensemble learning algorithm, including:
[0026] Obtain the reflectance spectrum corresponding to the first training sample;
[0027] The reflectance corresponding to each band in each reflectance spectrum is normalized to obtain the processed reflectance corresponding to each band.
[0028] Based on the ensemble learning algorithm and the frequency of each band in all reflectance spectra, the reflectance corresponding to each band after processing is filtered to select multiple feature bands from the reflectance spectrum.
[0029] In one possible implementation, a pre-built ensemble learning algorithm model is trained based on a second training sample and its corresponding identification information to obtain a trained hyperspectral oil spill monitoring model, including:
[0030] The reflectance of the target feature band in the second training sample is reassigned according to the preset weight corresponding to the target feature band to obtain the new reflectance of the target feature band after reassignment, and a third training sample corresponding to the second training sample is formed; wherein, the target feature band is any one of the feature bands in the preset band combination, and the third training sample is equipped with identification information corresponding to the second training sample.
[0031] The pre-built ensemble learning algorithm model is trained based on the third training sample and its corresponding identification information to obtain a trained hyperspectral oil spill monitoring model.
[0032] In one possible implementation, the ensemble learning algorithm is the CatBoost algorithm, and the hyperspectral oil spill monitoring model is a CatBoost-based model.
[0033] Thirdly, embodiments of the present invention provide an identification device for hyperspectral oil spill monitoring, comprising:
[0034] The acquisition module is used to acquire the reflectance spectrum of the hyperspectral remote sensing image of the monitoring area to be identified; wherein the reflectance spectrum corresponds to the reflectance image of the hyperspectral remote sensing image to be identified;
[0035] The identification module is used to extract the reflectance of a preset band combination from the reflectance spectrum as the target dataset, and input the target dataset into a pre-trained hyperspectral oil spill monitoring model to determine whether there is an oil spill area on the hyperspectral remote sensing image to be identified.
[0036] The hyperspectral oil spill monitoring model is trained on a pre-built ensemble learning algorithm model using training samples of reflectance based on a preset band combination. The preset band combination includes multiple feature bands, each of which is selected from the reflectance spectra of multiple hyperspectral remote sensing images containing oil films based on the ensemble learning algorithm.
[0037] In one possible implementation, an identification module is used to input the target dataset into a pre-trained hyperspectral oil spill monitoring model and perform oil spill identification processing on the target dataset;
[0038] Based on the results of oil spill identification processing, the oil spill areas in the reflectivity image are identified;
[0039] The identified portion of the reflectance image is classified and processed to determine whether an oil spill area exists on the hyperspectral image to be identified based on the results of the classification and processing; wherein, the classification and processing includes one or more of the following: classification merging, filtering or clustering.
[0040] In one possible implementation, the identification module is used to re-assign the reflectance of each feature band in the target dataset according to a preset weight, so as to obtain the re-assigned target dataset.
[0041] The assigned target dataset is input into a pre-trained hyperspectral oil spill monitoring model to determine whether there is an oil spill area on the hyperspectral remote sensing image to be identified.
[0042] The hyperspectral oil spill monitoring model is obtained by re-assigning the reflectance of each characteristic band according to a preset rule, and training a pre-built ensemble learning algorithm model based on training samples of the reflectance of the preset band combination after re-assignment.
[0043] In one possible implementation, the ensemble learning algorithm is the CatBoost algorithm, and the hyperspectral oil spill monitoring model is a model based on the CatBoost algorithm.
[0044] The feature bands are selected from all bands of the reflectance spectrum based on the CatBoost algorithm, according to the frequency of each band appearing in the reflectance spectra of multiple hyperspectral remote sensing images containing oil films.
[0045] Fourthly, embodiments of the present invention provide a training device for a hyperspectral oil spill monitoring model, comprising:
[0046] The sample acquisition module is used to acquire the first training sample; wherein, the first training sample includes multiple hyperspectral remote sensing images containing oil films, and each hyperspectral remote sensing image is provided with corresponding identification information, which is used to identify the oil film area of each hyperspectral remote sensing image.
[0047] The band selection module is used to obtain the reflectance spectrum corresponding to the first training sample, and select multiple feature bands from the reflectance spectrum based on the ensemble learning algorithm; wherein, the multiple feature bands are combined into a preset band combination;
[0048] A training sample formation module is used to extract the reflectance of a preset band combination from the reflectance spectrum of each hyperspectral remote sensing image as a target dataset, and form a second training sample corresponding to the first training sample. Each second training sample is equipped with identification information corresponding to the first training sample.
[0049] The training module is used to train the pre-built ensemble learning algorithm model based on the second training sample and its corresponding identification information to obtain a trained hyperspectral oil spill monitoring model.
[0050] In one possible implementation, a band filtering module is used to obtain the reflectance spectrum corresponding to the first training sample;
[0051] The reflectance corresponding to each band in each reflectance spectrum is normalized to obtain the processed reflectance corresponding to each band.
[0052] Based on the ensemble learning algorithm and the frequency of each band in all reflectance spectra, the reflectance corresponding to each band after processing is filtered to select multiple feature bands from the reflectance spectrum.
[0053] In one possible implementation, the training module is used to reassign the reflectance of the target feature band in the second training sample according to the preset weight corresponding to the target feature band, to obtain the new reflectance of the target feature band after reassignment, and to form a third training sample corresponding to the second training sample; wherein, the target feature band is any one of the feature bands in the preset band combination, and the third training sample is provided with identification information corresponding to the second training sample.
[0054] The pre-built ensemble learning algorithm model is trained based on the third training sample and its corresponding identification information to obtain a trained hyperspectral oil spill monitoring model.
[0055] In one possible implementation, the ensemble learning algorithm is the CatBoost algorithm, and the hyperspectral oil spill monitoring model is a CatBoost-based model.
[0056] Fifthly, embodiments of the present invention provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method as described in the first aspect or the second aspect or any possible implementation of the first aspect or the second aspect above.
[0057] In a sixth aspect, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method as described in the first aspect or the second aspect, or any possible implementation of the first aspect or the second aspect.
[0058] This invention provides a method for identifying hyperspectral oil spills, a model training method, and a storage medium for monitoring. First, the reflectance spectrum of the hyperspectral remote sensing image of the monitoring area to be identified is obtained. Then, the reflectance of a preset band combination is extracted from the reflectance spectrum as a target dataset, and the target dataset is input into a pre-trained hyperspectral oil spill monitoring model to finally determine whether there is an oil spill area on the hyperspectral remote sensing image to be identified.
[0059] By selecting bands from a pre-defined combination of bands from all bands of the reflectance spectrum, the reflectance spectral feature space of the hyperspectral remote sensing image to be identified is optimized, thereby reducing the influence of other bands on the identification process and improving the accuracy of the hyperspectral oil spill monitoring model. Attached Figure Description
[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0061] Figure 1 This is a flowchart illustrating the implementation of the training method for the hyperspectral oil spill monitoring model provided in this embodiment of the invention.
[0062] Figure 2This is a schematic diagram illustrating the importance of 240 bands determined using the CatBoost algorithm, provided in an embodiment of the present invention.
[0063] Figure 3 This is a schematic diagram of the 16 characteristic bands determined by the CatBoost algorithm according to an embodiment of the present invention;
[0064] Figure 4 This is a flowchart illustrating the implementation of the hyperspectral oil spill monitoring identification method provided in this embodiment of the invention.
[0065] Figure 5 This is a schematic diagram of the identification result using the spectral angle threshold classification method provided in an embodiment of the present invention;
[0066] Figure 6 This is a schematic diagram of the recognition result obtained using the recognition method provided by this invention, as provided in an embodiment of the invention;
[0067] Figure 7 This is a schematic diagram of the structure of the identification device for hyperspectral oil spill monitoring provided in an embodiment of the present invention;
[0068] Figure 8 This is a schematic diagram of the structure of the training device for the hyperspectral oil spill monitoring model provided in an embodiment of the present invention;
[0069] Figure 9 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0070] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.
[0071] To make the objectives, technical solutions, and advantages of the present invention clearer, specific embodiments will be described below in conjunction with the accompanying drawings.
[0072] Hyperspectral analysis is a classification method based on the physical properties of ground objects. It uses spectral curves that reflect the physical and optical properties of ground objects to identify them.
[0073] Ground cover types in images can be identified using known spectral data from spectral libraries and matching algorithms. Matching algorithms based on minimum inter-spectral distance calculate the distance between unknown and reference spectral values and then classify them according to the least squares principle. The inter-spectral distance can be Euclidean, Mahalanobis, or Bachian distance, etc. However, this method is highly sensitive to noise, which can affect the accuracy of identification.
[0074] Spectral angle mapping (SAM) treats the spectrum as a multidimensional vector, calculates the generalized angle between two spectral vectors, and considers the smaller the angle to be the more similar the spectra. Unknown spectra are then classified according to a given similarity threshold. However, this method is only suitable for coarse classification and identification, and cannot perform fine-grained classification.
[0075] Hyperspectral remote sensing images are characterized by multiple bands, high resolution, and rich information. They are also more outstanding in terms of detail representation, and can better distinguish the differences between seawater and oil slicks, thus displaying oil spill information more objectively.
[0076] However, hyperspectral remote sensing images are easily affected by incident light intensity and noise, which can affect the accuracy of oil film identification.
[0077] To address the problems of existing technologies, embodiments of the present invention provide a method for identifying hyperspectral oil spills, a model training method, and a storage medium.
[0078] Before introducing the identification method for hyperspectral oil spill monitoring provided in the embodiments of the present invention, it is necessary to first introduce the training method of the hyperspectral oil spill monitoring model. Once the hyperspectral oil spill monitoring model is trained, it can be used to identify whether there is an oil film in the monitoring area.
[0079] See Figure 1 The flowchart illustrating the training method of the hyperspectral oil spill monitoring model provided in this embodiment of the invention is described in detail below:
[0080] Step S110: Obtain the first training sample.
[0081] The first training sample includes multiple hyperspectral remote sensing images containing oil films, and each hyperspectral remote sensing image is associated with a label, which is used to identify the oil film area in each hyperspectral remote sensing image.
[0082] In this embodiment, multiple hyperspectral remote sensing images containing oil films acquired at different times using a hyperspectral imaging oil spill telemetry system can be used to form the first training sample. For example, the XHGGP-90C tower-type hyperspectral imaging oil spill telemetry system can be used. This system can be installed in the field, allowing for unattended long-term automatic monitoring, and possesses large-area monitoring and spatial imaging capabilities. It can be flexibly configured with a "point-line-area" working mode according to on-site needs, covering a water area of hundreds to tens of thousands of square meters. A single device can acquire the spatial distribution of water quality at different scales.
[0083] For example, the XHGGP-90C tower-type hyperspectral imaging oil spill telemetry system can be used to acquire multiple hyperspectral remote sensing images of oil films with varying volumes or thicknesses over different time periods. Each acquired hyperspectral remote sensing image is labeled with information indicating the oil film area. This labeling information can be visually interpreted to identify the oil film and non-oil film areas in each hyperspectral remote sensing image. Each hyperspectral remote sensing image is divided into positive and negative sample areas; the positive sample area represents the oil film area, and the negative sample area represents the non-oil film area.
[0084] Step S120: Obtain the reflectance spectrum corresponding to the first training sample, and select multiple feature bands from the reflectance spectrum based on the ensemble learning algorithm.
[0085] Among them, multiple characteristic bands are combined to form a preset band combination.
[0086] Because hyperspectral remote sensing images contain a large number of bands—for example, a tower-type hyperspectral image has 240 bands for ground features—the increased number of bands results in hyperspectral data being 1-2 orders of magnitude larger than traditional data, significantly increasing data storage and computer processing load. Furthermore, processing all bands would lead to feature redundancy, affecting identification accuracy. Therefore, it is necessary to extract reflectance from a predefined combination of bands from the reflectance spectrum.
[0087] In some embodiments, the reflectivity of oil films collected at different time periods varies, and this difference can affect the identification results. Therefore, before screening feature bands, the reflectivity needs to be processed first. The specific processing steps are as follows:
[0088] Step S1201: Obtain the reflectance spectrum corresponding to the first training sample.
[0089] Obtain the reflectance spectrum of each pixel in each hyperspectral remote sensing image containing an oil film in the first training sample. Each hyperspectral remote sensing image containing an oil film corresponds to a reflectance spectrum.
[0090] Step S1202: Normalize the reflectance corresponding to each band in each reflectance spectrum to obtain the processed reflectance corresponding to each band.
[0091] The normalization formula is:
[0092]
[0093] Among them, R i Let R be the reflectance of each pixel in band i, R be the normalized reflectance, and n be the total number of bands.
[0094] Step S1203: Based on the ensemble learning algorithm and the frequency of each band in all reflectance spectra, the reflectance corresponding to each band after processing is screened to select multiple feature bands from the reflectance spectrum.
[0095] In this embodiment, by normalizing the reflectance using steps S1201-S1203, the problem of poor generalization ability in oil film recognition caused by varying illumination at different times can be solved. An ensemble learning algorithm is used to calculate the importance of each band by improving the performance metric at each band's feature split point. Multiple feature bands are then selected from the reflectance spectrum based on the importance of each band, thereby reducing the influence of other interfering bands and improving recognition accuracy. Furthermore, by compressing the data and optimizing the spectral feature space, the dimensionality of feature extraction is reduced, effectively improving the speed of data processing.
[0096] In some embodiments, the ensemble learning algorithm can be the CatBoost algorithm. By employing the CatBoost algorithm to perform statistical analysis on all bands in the reflectance spectrum corresponding to the first training sample, the importance of each band is determined based on the frequency of each band appearing in all reflectance spectra, thereby selecting multiple feature bands from the reflectance spectrum based on the importance of each band.
[0097] like Figure 2 and 3 As shown, it illustrates the use of the CatBoost algorithm to determine the importance of 240 bands based on the frequency of each band's occurrence, and then selecting the top 16 important bands from these 240 bands as feature bands. Among these, Figure 2 and Figure 3 The horizontal axis represents the waveband, and the vertical axis represents the importance of the feature.
[0098] Step S130: Extract the reflectance of a preset band combination from the reflectance spectrum of each hyperspectral remote sensing image as the target dataset, and form a second training sample corresponding to the first training sample.
[0099] Each of the second training samples contains identification information corresponding to that of the first training sample.
[0100] The reflectance of a preset band combination is extracted from the reflectance spectrum of each hyperspectral remote sensing image as the target dataset, and the target datasets of all hyperspectral remote sensing images are combined to form a second training sample corresponding to the first training sample.
[0101] Step S140: Train the pre-built ensemble learning algorithm model based on the second training sample and its corresponding identification information to obtain the trained hyperspectral oil spill monitoring model.
[0102] In some embodiments, the pre-built ensemble learning algorithm model is also a CatBoost algorithm-based model, and the hyperspectral oil spill monitoring model is a CatBoost algorithm model trained with training samples.
[0103] In different embodiments, different training samples can be used to train the pre-built ensemble learning algorithm model.
[0104] In one embodiment, a pre-built ensemble learning algorithm model can be trained directly by selecting a second training sample and its corresponding identification information. By comparing the identification result of the second training sample by the pre-built ensemble learning algorithm model with the identification information of the second training sample, the final trained hyperspectral oil spill monitoring model can be determined.
[0105] In another embodiment, due to the small reflectance value and the varying importance of reflectance across different bands, to improve identification accuracy, the reflectance of the target feature band in the second training sample can first be reassigned according to a preset weight corresponding to the target feature band, resulting in a new reflectance for the target feature band, which forms a third training sample corresponding to the second training sample. Then, based on the third training sample and its corresponding identification information, a pre-constructed ensemble learning algorithm model is trained to obtain a trained hyperspectral oil spill monitoring model. Here, the target feature band is any one of the preset band combinations, and each of the third training samples has identification information corresponding to the second training sample.
[0106] It should be noted that the reassignment here can either simultaneously increase the reflectance of all preset band combinations by N times, or it can set different weights based on the importance of each characteristic band, thereby highlighting the role of important characteristic bands and further improving the accuracy of identification.
[0107] Once the hyperspectral oil spill monitoring model has been trained, it can be used for hyperspectral oil spill monitoring.
[0108] The identification method for hyperspectral oil spill monitoring provided in the embodiments of the present invention will be described below.
[0109] See Figure 4 The flowchart illustrating the implementation of the hyperspectral oil spill monitoring identification method provided in this embodiment of the invention is described in detail below:
[0110] Step S410: Obtain the reflectance spectrum of the hyperspectral remote sensing image of the monitoring area to be identified.
[0111] The reflectance spectrum corresponds to the reflectance image of the hyperspectral remote sensing image to be identified.
[0112] Because hyperspectral remote sensing images are easily affected by varying incident light intensities and noise, the spectral images of oil films can vary significantly. Directly using hyperspectral remote sensing images to identify the presence of oil films often fails to accurately predict their presence under different spatiotemporal conditions. However, reflectance images do not change with the light source; the reflectance of each pixel is only related to the material of the sample. For example, the reflectance of a stool remains constant under different environments; its reflectance image can be used to accurately identify the stool without being affected by the environment. The reflectance image is obtained by calculating the reflectance spectrum of each pixel in the hyperspectral remote sensing image to be identified.
[0113] Step S420: Extract the reflectance of a preset band combination from the reflectance spectrum as the target dataset, and input the target dataset into the pre-trained hyperspectral oil spill monitoring model to determine whether there is an oil spill area on the hyperspectral remote sensing image to be identified.
[0114] As mentioned in the training process of the hyperspectral oil spill monitoring model, different hyperspectral oil spill monitoring models will be obtained by training the pre-built ensemble learning algorithm model with different training samples.
[0115] Hyperspectral oil spill monitoring models trained on different training samples also require the use of data corresponding to the training data when in use.
[0116] In one implementation, the hyperspectral oil spill monitoring model is trained on a pre-constructed ensemble learning algorithm model using training samples of reflectance from a preset band combination. The preset band combination includes multiple feature bands, each of which is selected from the reflectance spectra of multiple hyperspectral remote sensing images containing oil films using an ensemble learning algorithm.
[0117] When using this hyperspectral oil spill monitoring model for identification, the reflectance of a preset band combination can be extracted from the reflectance spectrum as the target dataset. Then, the target dataset can be directly input into the pre-trained hyperspectral oil spill monitoring model to determine whether there is an oil spill area on the hyperspectral remote sensing image to be identified.
[0118] In another embodiment, the hyperspectral oil spill monitoring model is obtained by training a pre-built ensemble learning algorithm model based on training samples of the reflectance of the pre-built ensemble learning algorithm model, which are based on the reassignment of the reflectance of each feature band according to a preset rule and the reassignment of the pre-set band combination.
[0119] When using this hyperspectral oil spill monitoring model for identification, the reflectance of a preset combination of bands can be extracted from the reflectance spectrum as the target dataset. Then, the reflectance of each feature band in the target dataset is reassigned according to a preset weight. Finally, the reassigned target dataset is input into the pre-trained hyperspectral oil spill monitoring model to determine whether an oil spill area exists on the hyperspectral remote sensing image to be identified.
[0120] In this embodiment, by reassigning the reflectance of each feature band in the target dataset according to a preset weight, the importance of the feature bands can be amplified, thereby improving the accuracy of identification. This reassignment according to the preset weight can mean simultaneously increasing the reflectance of all preset band combinations by a factor of N. Alternatively, different weights can be set according to the importance of each feature band, thereby highlighting the role of important feature bands and further improving the accuracy of identification.
[0121] In some embodiments, after identification by the hyperspectral oil spill monitoring model, sporadic isolated spots or small holes are found in the identified reflectance image, which can affect the accuracy of the identification. Therefore, further processing of the identified and labeled reflectance image is required to improve the accuracy of the identification. The specific processing steps are as follows:
[0122] Step S4201: Input the target dataset into the pre-trained hyperspectral oil spill monitoring model and perform oil spill identification processing on the target dataset.
[0123] Step S4202: Based on the results of the oil spill identification process, the oil spill area in the reflectivity image is identified.
[0124] The results of oil spill identification processing using the hyperspectral oil spill monitoring model indicate that if an oil spill area exists on the reflectance image, it will be marked accordingly to show that an oil spill exists in that area. However, it can be observed that some of the identified oil spill areas have scattered isolated dots on the outside and some holes on the inside, that is, there are local misclassified pixels, making accurate identification difficult. Therefore, further processing of the reflectance image is required.
[0125] Step S4203: Perform classification post-processing on the identified portion of the reflectance image to determine whether there is an oil spill area on the hyperspectral image to be identified based on the results of the classification post-processing.
[0126] The post-classification processing includes one or more of the following: classification merging, filtering, or clustering. By performing post-classification processing on the labeled reflectance image, the recognition results can be made more accurate, and the influence of holes and isolated points can be removed.
[0127] In this embodiment, by classifying and processing the labeled reflectance image, the influence of holes and isolated dots on recognition can be removed, thereby improving the accuracy of recognition.
[0128] The identification method provided by the present invention first obtains the reflectance spectrum of the hyperspectral remote sensing image of the monitoring area to be identified, then extracts the reflectance of a preset band combination from the reflectance spectrum as the target dataset, and inputs the target dataset into a pre-trained hyperspectral oil spill monitoring model to finally determine whether there is an oil spill area on the hyperspectral remote sensing image to be identified.
[0129] By selecting bands from a pre-defined combination of bands from all bands of the reflectance spectrum, the reflectance spectral feature space of the hyperspectral remote sensing image to be identified is optimized, thereby reducing the influence of other bands on the identification process and improving the accuracy of the hyperspectral oil spill monitoring model.
[0130] The following uses the hyperspectral oil spill monitoring identification method and spectral angle threshold classification identification method provided by this invention to identify eight remote sensing images of different time phases in the 240-band spectrum. In this invention, 16 characteristic bands were selected from the 240-band spectrum. The identification results are as follows: Figure 5 and Figure 6 As shown, the recognition accuracy and time are illustrated in Tables 1 and 2:
[0131] Table 1. Recognition accuracy and time taken by the spectral angle threshold classification method.
[0132]
[0133] Table 2. Identification method for hyperspectral oil spill monitoring provided by the present invention.
[0134]
[0135] As can be seen from the data in Tables 1 and 2, the recognition method provided by this invention can handle categorical features more reasonably and obtain good recognition results without parameter tuning. It effectively solves the problem of prediction offset caused by gradient bias and multi-temporal spectral differences, reduces overfitting, and improves the accuracy and generalization ability of recognition.
[0136] from Figure 5 and Figure 6 It can be seen that the identification method provided by this invention has high identification accuracy. Among other things, Figure 5 and Figure 6 The white area in the middle represents the identified oil film area.
[0137] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.
[0138] Based on the identification method and training method of the hyperspectral oil spill monitoring model provided in the above embodiments, the present invention also provides specific implementation methods of the identification device and training device of the hyperspectral oil spill monitoring model applied to the identification method and training method of the hyperspectral oil spill monitoring model. Please refer to the following embodiments.
[0139] like Figure 7 As shown, a hyperspectral oil spill monitoring identification device 700 is provided, the device comprising:
[0140] The acquisition module 710 is used to acquire the reflectance spectrum of the hyperspectral remote sensing image of the monitoring area to be identified; wherein the reflectance spectrum corresponds to the reflectance image of the hyperspectral remote sensing image to be identified;
[0141] The identification module 720 is used to extract the reflectance of a preset band combination from the reflectance spectrum as the target dataset, and input the target dataset into a pre-trained hyperspectral oil spill monitoring model to determine whether there is an oil spill area on the hyperspectral remote sensing image to be identified.
[0142] The hyperspectral oil spill monitoring model is trained on a pre-built ensemble learning algorithm model using training samples of reflectance based on a preset band combination. The preset band combination includes multiple feature bands, each of which is selected from the reflectance spectra of multiple hyperspectral remote sensing images containing oil films based on the ensemble learning algorithm.
[0143] In one possible implementation, the identification module 720 is used to input the target dataset into a pre-trained hyperspectral oil spill monitoring model and perform oil spill identification processing on the target dataset;
[0144] Based on the results of oil spill identification processing, the oil spill areas in the reflectivity image are identified;
[0145] The identified portion of the reflectance image is classified and processed to determine whether an oil spill area exists on the hyperspectral image to be identified based on the results of the classification and processing; wherein, the classification and processing includes one or more of the following: classification merging, filtering or clustering.
[0146] In one possible implementation, the identification module 720 is used to re-assign the reflectance of each feature band in the target dataset according to a preset weight, so as to obtain the re-assigned target dataset.
[0147] The assigned target dataset is input into a pre-trained hyperspectral oil spill monitoring model to determine whether there is an oil spill area on the hyperspectral remote sensing image to be identified.
[0148] The hyperspectral oil spill monitoring model is obtained by re-assigning the reflectance of each characteristic band according to a preset rule, and training a pre-built ensemble learning algorithm model based on training samples of the reflectance of the preset band combination after re-assignment.
[0149] In one possible implementation, the ensemble learning algorithm is the CatBoost algorithm, and the hyperspectral oil spill monitoring model is a model based on the CatBoost algorithm.
[0150] The feature bands are selected from all bands of the reflectance spectrum based on the CatBoost algorithm, according to the frequency of each band appearing in the reflectance spectra of multiple hyperspectral remote sensing images containing oil films.
[0151] like Figure 8 As shown, a training device 800 for a hyperspectral oil spill monitoring model is provided, the device comprising:
[0152] The sample acquisition module 810 is used to acquire the first training sample; wherein, the first training sample includes multiple hyperspectral remote sensing images containing oil films, and each hyperspectral remote sensing image is provided with corresponding identification information, which is used to identify the oil film area of each hyperspectral remote sensing image.
[0153] The band selection module 820 is used to obtain the reflectance spectrum corresponding to the first training sample and select multiple feature bands from the reflectance spectrum based on the ensemble learning algorithm; wherein, the multiple feature bands are combined into a preset band combination;
[0154] The training sample forming module 830 is used to extract the reflectance of a preset band combination from the reflectance spectrum of each hyperspectral remote sensing image as the target dataset, and form a second training sample corresponding to the first training sample. The second training sample is equipped with identification information corresponding to the first training sample.
[0155] The training module 840 is used to train the pre-built ensemble learning algorithm model based on the second training sample and its corresponding identification information to obtain a trained hyperspectral oil spill monitoring model.
[0156] In one possible implementation, the band filtering module 820 is used to obtain the reflectance spectrum corresponding to the first training sample;
[0157] The reflectance corresponding to each band in each reflectance spectrum is normalized to obtain the processed reflectance corresponding to each band.
[0158] Based on the ensemble learning algorithm and the frequency of each band in all reflectance spectra, the reflectance corresponding to each band after processing is filtered to select multiple feature bands from the reflectance spectrum.
[0159] In one possible implementation, the training module 840 is used to reassign the reflectance of the target feature band in the second training sample according to the preset weight corresponding to the target feature band, to obtain the new reflectance of the target feature band after reassignment, and to form a third training sample corresponding to the second training sample; wherein, the target feature band is any one of the feature bands in the preset band combination, and the third training sample is provided with identification information corresponding to the second training sample.
[0160] The pre-built ensemble learning algorithm model is trained based on the third training sample and its corresponding identification information to obtain a trained hyperspectral oil spill monitoring model.
[0161] In one possible implementation, the ensemble learning algorithm is the CatBoost algorithm, and the hyperspectral oil spill monitoring model is a CatBoost-based model.
[0162] Figure 9 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. For example... Figure 9 As shown, the electronic device 9 of this embodiment includes: a processor 90, a memory 91, and a computer program 92 stored in the memory 91 and executable on the processor 90. When the processor 90 executes the computer program 92, it implements the steps in the embodiments of the above-described hyperspectral oil spill monitoring identification methods and hyperspectral oil spill monitoring model training methods, for example... Figure 1 Steps 110 to 140 and / or shown Figure 4 Steps 410 to 420 are shown. Alternatively, when the processor 90 executes the computer program 92, it implements the functions of each module in the above-described device embodiments, for example... Figure 7 Modules 710 to 720 and / or shown Figure 8 The functions of modules 810 to 840 are shown.
[0163] For example, the computer program 92 can be divided into one or more modules, which are stored in the memory 91 and executed by the processor 90 to complete the present invention. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program 92 in the electronic device 9. For example, the computer program 92 can be divided into... Figure 7 Modules 710 to 720 and / or shown Figure 8 Modules 810 to 840 are shown.
[0164] The electronic device 9 may include, but is not limited to, a processor 90 and a memory 91. Those skilled in the art will understand that... Figure 9 This is merely an example of electronic device 9 and does not constitute a limitation on electronic device 9. It may include more or fewer components than shown, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.
[0165] The processor 90 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.
[0166] The memory 91 can be an internal storage unit of the electronic device 9, such as a hard disk or memory. The memory 91 can also be an external storage device of the electronic device 9, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory 91 can include both internal and external storage units of the electronic device 9. The memory 91 is used to store the computer program and other programs and data required by the electronic device. The memory 91 can also be used to temporarily store data that has been output or will be output.
[0167] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.
[0168] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0169] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.
[0170] In the embodiments provided by this invention, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.
[0171] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0172] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0173] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of the above embodiments of the identification method and training method for the hyperspectral oil spill monitoring model. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.
[0174] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.
Claims
1. A method for identifying oil spills using hyperspectral imaging, characterized in that, include: Acquire the reflectance spectrum of the hyperspectral remote sensing image of the monitoring area to be identified; wherein the reflectance spectrum corresponds to the reflectance image of the hyperspectral remote sensing image to be identified; The reflectance of a preset band combination is extracted from the reflectance spectrum as a target dataset, and the target dataset is input into a pre-trained hyperspectral oil spill monitoring model to perform oil spill identification processing on the target dataset; based on the results of the oil spill identification processing, the oil spill areas in the reflectance image are marked; the marked parts of the reflectance image are subjected to classification post-processing to determine whether there is an oil spill area on the hyperspectral remote sensing image to be identified based on the results of the classification post-processing; wherein, the classification post-processing includes one or more of the following processes: classification merging, filtering, or clustering processing; Before inputting the target dataset into the pre-trained hyperspectral oil spill monitoring model, the method further includes: The reflectance of each feature band in the target dataset is re-weighted according to a preset weight to obtain the re-weighted target dataset; the re-weighted target dataset is input into a pre-trained hyperspectral oil spill monitoring model to determine whether there is an oil spill area on the hyperspectral remote sensing image to be identified; The hyperspectral oil spill monitoring model is obtained by re-assigning the reflectance of each of the feature bands according to a preset rule, and training a pre-constructed ensemble learning algorithm model based on training samples of the reflectance of the preset band combination after re-assignment. The preset band combination includes multiple feature bands, each of which is selected from the reflectance spectra of multiple hyperspectral remote sensing images containing oil films based on the ensemble learning algorithm, and each feature band is selected based on the frequency of each band appearing in all reflectance spectra.
2. The identification method as described in claim 1, characterized in that, The ensemble learning algorithm is the CatBoost algorithm, and the hyperspectral oil spill monitoring model is a model based on the CatBoost algorithm. The characteristic bands are selected from all bands of the reflectance spectrum based on the CatBoost algorithm, according to the frequency of each band appearing in the reflectance spectra of multiple hyperspectral remote sensing images containing oil films.
3. A training method for a hyperspectral oil spill monitoring model, applied in the hyperspectral oil spill monitoring identification method of claim 1 or 2, characterized in that, include: Obtain a first training sample; wherein the first training sample includes multiple hyperspectral remote sensing images containing oil films, and each hyperspectral remote sensing image is provided with corresponding identification information, the identification information being used to identify the oil film area of each hyperspectral remote sensing image; Obtain the reflectance spectrum corresponding to the first training sample, and based on the ensemble learning algorithm, select multiple feature bands from the reflectance spectrum; wherein, the multiple feature bands are combined into a preset band combination; The reflectance of a preset band combination is extracted from the reflectance spectrum of each hyperspectral remote sensing image as a target dataset, and a second training sample corresponding to the first training sample is formed. Each of the second training samples is equipped with identification information corresponding to the first training sample. The pre-constructed ensemble learning algorithm model is trained based on the second training sample and its corresponding identification information to obtain a trained hyperspectral oil spill monitoring model.
4. The training method as described in claim 3, characterized in that, The step of obtaining the reflectance spectrum corresponding to the first training sample, and selecting multiple feature bands from the reflectance spectrum based on an ensemble learning algorithm, includes: Obtain the reflectance spectrum corresponding to the first training sample; The reflectance corresponding to each band in each reflectance spectrum is normalized to obtain the processed reflectance corresponding to each band. Based on the ensemble learning algorithm and the frequency of each band in all reflectance spectra, the reflectance corresponding to each band after processing is filtered to select multiple feature bands from the reflectance spectrum.
5. The training method as described in claim 4, characterized in that, The step of training a pre-constructed ensemble learning algorithm model based on the second training samples and their corresponding identification information to obtain a trained hyperspectral oil spill monitoring model includes: The reflectance of the target feature band in the second training sample is re-assigned according to the preset weight corresponding to the target feature band to obtain the new reflectance of the target feature band after re-assignment, and a third training sample corresponding to the second training sample is formed; wherein, the target feature band is any one of the feature bands in the preset band combination, and the third training sample is provided with identification information corresponding to the second training sample. The pre-constructed ensemble learning algorithm model is trained based on the third training sample and its corresponding identification information to obtain a trained hyperspectral oil spill monitoring model.
6. The training method according to any one of claims 3-5, characterized in that, The ensemble learning algorithm is the CatBoost algorithm, and the hyperspectral oil spill monitoring model is a model based on the CatBoost algorithm.
7. An electronic device, characterized in that, The method includes a memory and a processor, the memory being used to store a computer program, and the processor being used to call and run the computer program stored in the memory to perform the method as described in any one of claims 1 to 6.
8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 6.