Methods for Sea Ice Classification and Concentration Inversion Using UAV Optical and Thermal Infrared Data

Abstract

This application relates to a method for sea ice classification and concentration inversion using UAV optical and thermal infrared data. The method includes: first, obtaining registered UAV optical and thermal infrared images through preprocessing techniques such as orbit correction, radiometric calibration, and geometric correction, providing high-quality data for subsequent sea ice information extraction; second, constructing an Ice-Unet deep learning model to obtain high-precision sea ice classification results; third, selecting surface temperature values ​​corresponding to open water and fixed ice in the sea ice classification to construct coordinate point values ​​for the thermal infrared data, and constructing a linear inversion model based on a radiative transfer model to obtain sea ice concentration inversion results based on thermal infrared data. This solves the problems of existing technologies in achieving rapid reporting and on-site support of sea ice information in Arctic waters.

Description

Technical Field

[0001] This application relates to the field of remote sensing image processing technology, and in particular to a method for sea ice classification and concentration inversion based on UAV optical and thermal infrared data. Background Technology

[0002] As a crucial component of the marine environment, Arctic sea ice covers approximately 2% of the global ocean area. On one hand, the rapid changes in polar sea ice affect the ocean's heat and water balance, making it a significant monitoring factor for global climate change. On the other hand, the spatiotemporal distribution of polar sea ice also influences shipping route planning and the development of polar resources, making it a current research hotspot. Real-time monitoring and analysis of sea ice distribution, concentration, and thickness allow for better understanding and prediction of ice conditions in shipping lanes, enabling the adoption of appropriate navigation strategies. This not only ensures navigational safety but also provides valuable data for understanding ecological and environmental changes in the Arctic region, thus predicting the impacts of these changes on global climate and ecosystems. Therefore, Arctic sea ice monitoring has significant scientific and economic value.

[0003] In sea ice classification, hyperspectral data can be used to construct a complete framework for sea ice feature extraction and analysis using spectral-spatial-joint feature methods to obtain high-precision sea ice category information. High-resolution SAR (Synthetic Aperture Radar) data can be used to obtain detailed sea ice distribution information. Combined with traditional sea ice concentration products, richer multi-scale ice condition maps can be constructed to support polar sea ice monitoring and route optimization. Regarding classification model selection, the main methods include support vector machines, Bayesian algorithms, neural networks, and principal component analysis. Support Vector Machines (SVMs) combined with texture feature analysis and using gray-level co-occurrence matrices to extract feature values ​​improve classification accuracy and reliability. However, training time increases significantly with the amount of data, and the selection of kernel functions and regularization parameters has a significant impact on the results. The optimization process is complex, and it is inefficient when dealing with multi-class classification problems. Bayesian algorithms calculate the posterior probability of each class by statistically analyzing the prior and conditional probabilities of different classes and using Bayes' theorem, selecting the class with the highest posterior probability as the classification result. Naive Bayes assumes feature independence, but the spectral, texture, and spatial features of sea ice are often highly correlated, and this algorithm is sensitive to prior distributions and cannot handle complex distributions. Neural networks are classifiers built based on neuron simulations, suitable for handling complex nonlinear problems. However, they are prone to overfitting when training samples are insufficient or data noise is high, and the selection of hyperparameters has a significant impact on the results. The optimization process is complex and time-consuming. Principal Component Analysis (PCA) compresses high-dimensional spectral or texture data into a low-dimensional space to extract key features. However, data dimensionality reduction may result in the loss of some useful information, affecting classification performance and model interpretability. Furthermore, principal component analysis assumes that the data has a linear structure, making it difficult to handle complex nonlinear relationships.

[0004] In the field of sea ice concentration inversion, current research mainly includes methods based on passive microwave concentration inversion algorithms, optical data inversion algorithms, and multi-source data fusion, as detailed below:

[0005] 1) Single-frequency inversion methods based on polarization ratio, such as the Bootstrap method, use polarization ratio information of vertical and horizontal brightness temperatures at 19 GHz or 37 GHz to invert density. However, the polarization ratio is weakly responsive to snow cover and melting ice, and is prone to large errors during the summer melting period. Secondly, the resolution of passive microwave brightness temperature is usually on the order of tens of kilometers, which is difficult to reflect small-scale sea ice distribution.

[0006] 2) Single-polarization methods based on frequency gradients describe the distribution of sea ice and seawater within a resolution cell by using the differences in electromagnetic scattering of microwaves at different frequencies to perform density inversion, such as the CalVal and NORSEX algorithms. These algorithms rely on empirical models, and the threshold of the frequency gradient and model parameters need to be adjusted according to the region and season, thus having limited versatility.

[0007] 3) Dense inversion algorithms that combine polarization ratio and frequency gradient, such as the ASI (ARTIST Sea Ice, Arctic Radiation and Turbulence Exchange Research Sea Ice) algorithm and the NASATeam algorithm. These methods have high algorithm complexity, require processing multiple channel data simultaneously, and have high computational costs.

[0008] 4) Dense inversion algorithm based on multi-channel data;

[0009] 5) Sea ice concentration inversion algorithms based on multi-source data fusion, such as sea ice concentration inversion methods based on neural network models. These methods establish neural network models using features such as passive microwave polarization ratio and SAR backscattering. Based on feature fusion, hidden layer units of the network are constructed, and iterative updates are performed through error backpropagation. The maximum likelihood constraint criterion is used to obtain the concentration inversion results of the time series.

[0010] In summary, existing technologies still have problems such as the trade-off between resolution and coverage, poor adaptability to complex environments, and insufficient model versatility.

[0011] With the continuous advancement and updating of detection methods, drones are increasingly being used in sea ice research both domestically and internationally. Compared with other traditional methods such as satellite remote sensing, aerial remote sensing, ship-based and helicopter observation, satellite remote sensing can cover a wider area, but its data acquisition usually requires waiting for satellites to pass overhead, and may not be effective in adverse weather conditions. Drones, on the other hand, can provide real-time data, reducing reliance on satellite and aerial remote sensing data, thereby lowering costs and improving the timeliness of data acquisition. In terms of spatial resolution and data accuracy, drones can provide up to sub-meter spatial resolution, enabling them to identify and measure the physical properties of sea ice in detail, such as ice surface roughness and melt pool information. In addition, drones can be deployed quickly, operate flexibly, and can enter sparsely populated areas to provide first-hand data. Although drone operation is greatly limited by weather conditions, such as strong winds and severe weather, which may restrict its flight capabilities, drones can provide more detailed local data, and their continuity can be ensured through more frequent flight missions.

[0012] However, existing technologies are insufficient for rapid reporting and on-site support of sea ice information in Arctic waters, and this issue urgently needs to be addressed. Summary of the Invention

[0013] This application provides a method for classifying and inverting sea ice concentration using optical and thermal infrared data from unmanned aerial vehicles (UAVs), in order to solve the problems of existing technologies being unable to achieve rapid reporting of sea ice information and on-site support in Arctic seawater.

[0014] The first aspect of this application provides a method for sea ice classification and concentration inversion using optical and thermal infrared data from a UAV, comprising the following steps: acquiring optical and thermal infrared data of a target UAV, and preprocessing and orthorectifying the optical and thermal infrared data to obtain an orthorectified image corresponding to the optical and thermal infrared data; constructing a sea ice classification training dataset based on the orthorectified image and a preset SAM model, training a pre-constructed Ice-Unet model using the sea ice classification training dataset, and inputting the orthorectified image into the trained Ice-Unet model to output sea ice category information corresponding to the orthorectified image; determining the target reference temperature for the target type of water area and the target type of sea ice based on the sea ice category information, and constructing a sea ice concentration inversion model based on the target reference temperature, so as to quantitatively evaluate the sea ice concentration inversion accuracy using the sea ice concentration inversion model.

[0015] Optionally, in one embodiment of this application, acquiring the optical and thermal infrared data of the target UAV, and preprocessing and orthorectifying the optical and thermal infrared data to obtain an orthorectified image corresponding to the optical and thermal infrared data, includes: performing geometric correction and radiometric calibration on the optical and thermal infrared data to obtain corresponding calibration results; constructing a corresponding digital elevation model based on the calibration results, and performing orthorectification on the optical and thermal infrared data according to the digital elevation model to obtain corresponding orthorectified data; and performing orthorectified stitching on the orthorectified data to obtain the orthorectified image.

[0016] Optionally, in one embodiment of this application, training the pre-constructed Ice-Unet model using the sea ice classification training dataset includes: constructing the Ice-Unet model based on a preset backbone feature extraction network, an enhanced feature extraction network, and a prediction network; inputting training data from the sea ice classification training dataset into the backbone feature extraction network of the Ice-Unet model to perform multiple convolution and max pooling operations on the training data to generate multiple preliminary effective feature layers; inputting the multiple preliminary effective feature layers into the enhanced feature extraction network to perform upsampling and feature fusion processing on the multiple preliminary effective feature layers to obtain a target fused effective feature layer; and using a target convolutional layer to perform channel adjustment operations on the target fused effective feature layer to obtain the trained Ice-Unet model.

[0017] Optionally, in one embodiment of this application, the step of determining the target type of water area and the target type of sea ice based on the sea ice category information, and constructing a sea ice concentration inversion model based on the target reference temperature, so as to quantitatively evaluate the sea ice concentration inversion accuracy using the sea ice concentration inversion model, includes: determining the target pixel grid corresponding to each pixel in the orthorectified image, and dividing the target pixel grid into multiple sub-grids; selecting the target percentile sea ice surface temperature in each of the multiple sub-grids, and determining the preliminary sea ice reference temperature of the target type of sea ice based on the target percentile sea ice surface temperature; and determining the preliminary sea ice reference temperature based on the preliminary sea ice reference temperature and a preset linear... A regression strategy is used to determine the final reference temperature of the target type of sea ice. Based on a preset pixel-by-pixel sliding window strategy, multiple overlay operations are performed on each pixel to obtain the final reference temperature of the target type of sea ice corresponding to each overlay operation. The target reference temperature of the target type of sea ice corresponding to each pixel is calculated based on the final reference temperature of the target type of sea ice corresponding to each overlay operation. The sea ice concentration corresponding to each pixel is calculated based on the target reference temperature of the target type of water area and the target type of sea ice and the sea ice concentration inversion model. The sea ice concentration is then projected to obtain the corresponding projection result, and the accuracy of the sea ice concentration inversion is quantitatively evaluated through the projection result.

[0018] Optionally, in one embodiment of this application, the mathematical expression of the sea ice concentration inversion model is:

[0019]

[0020] Wherein, SIC represents the sea ice concentration; t pwate The target reference temperature of the target type of water body is indicated by t. pice The target reference temperature represents the target type of sea ice; IST represents the surface temperature of the sea ice.

[0021] A second aspect of this application provides a device for classifying and inverting sea ice concentration based on UAV optical and thermal infrared data, comprising: a preprocessing module for acquiring optical and thermal infrared data of a target UAV, and performing preprocessing and orthorectification stitching operations on the optical and thermal infrared data to obtain an orthorectified image corresponding to the optical and thermal infrared data; a classification module for constructing a sea ice classification training dataset based on the orthorectified image and a preset SAM model, training a pre-constructed Ice-Unet model using the sea ice classification training dataset, and inputting the orthorectified image into the trained Ice-Unet model to output sea ice category information corresponding to the orthorectified image; and an inversion module for determining the target type water area and the target type sea ice based on the sea ice category information, and constructing a sea ice concentration inversion model based on the target reference temperature to quantitatively evaluate the sea ice concentration inversion accuracy using the sea ice concentration inversion model.

[0022] Optionally, in one embodiment of this application, the preprocessing module includes: a radiometric calibration unit, used to perform geometric correction and radiometric calibration operations on the optical and thermal infrared data to obtain corresponding calibration results; an orthorectification unit, used to construct a corresponding digital elevation model based on the calibration results, and perform orthorectification operations on the optical and thermal infrared data according to the digital elevation model to obtain corresponding orthorectified data; and an orthorectification stitching unit, used to perform orthorectification stitching processing on the orthorectified data to obtain the orthorectified image.

[0023] Optionally, in one embodiment of this application, the classification module includes: a modeling unit, used to construct the Ice-Unet model based on a preset backbone feature extraction network, an enhanced feature extraction network, and a prediction network; a backbone feature extraction unit, used to input training data from the sea ice classification training dataset into the backbone feature extraction network of the Ice-Unet model to perform multiple convolution and max pooling operations on the training data to generate multiple preliminary effective feature layers; an enhanced feature extraction unit, used to input the multiple preliminary effective feature layers into the enhanced feature extraction network to perform upsampling and feature fusion processing on the multiple preliminary effective feature layers to obtain a target fused effective feature layer; and a prediction unit, used to perform channel adjustment operations on the target fused effective feature layer using a target convolutional layer to obtain the trained Ice-Unet model.

[0024] Optionally, in one embodiment of this application, the inversion module includes: a partitioning unit, configured to determine the target pixel grid corresponding to each pixel in the orthorectified image, and divide the target pixel grid into multiple sub-grids; a selection unit, configured to select the target percentile sea ice surface temperature in each of the multiple sub-grids, and determine the preliminary sea ice reference temperature of the target type of sea ice based on the target percentile sea ice surface temperature; a determination unit, configured to determine the final reference temperature of the target type of sea ice based on the preliminary sea ice reference temperature and a preset linear regression strategy; and a coverage unit, configured to cover the image based on a preset pixel-by-pixel sliding... The windowing strategy performs multiple coverage operations on each pixel to obtain the final reference temperature of the target type of sea ice corresponding to each coverage operation, and calculates the target reference temperature of the target type of sea ice corresponding to each pixel based on the final reference temperature of the target type of sea ice corresponding to each coverage operation; the quantitative evaluation unit is used to calculate the sea ice concentration corresponding to each pixel based on the target reference temperature of the target type of water area and the target type of sea ice and the sea ice concentration inversion model, and projects the sea ice concentration to obtain the corresponding projection result, and quantitatively evaluates the sea ice concentration inversion accuracy through the projection result.

[0025] Optionally, in one embodiment of this application, the mathematical expression of the sea ice concentration inversion model is:

[0026]

[0027] Wherein, SIC represents the sea ice concentration; t pwate The target reference temperature of the target type of water body is indicated by t. pice The target reference temperature represents the target type of sea ice; IST represents the surface temperature of the sea ice.

[0028] A third aspect of this application provides 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 program to implement the UAV optical and thermal infrared data sea ice classification and concentration inversion method as described in the above embodiments.

[0029] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for classifying and inverting sea ice concentrations from UAV optical and thermal infrared data.

[0030] A fifth aspect of this application provides a computer program product, including a computer program that is executed to implement the above-described method for classifying and inverting sea ice concentrations from UAV optical and thermal infrared data.

[0031] Therefore, the embodiments of this application have the following beneficial effects:

[0032] The embodiments of this application acquire optical and thermal infrared data of a target UAV, preprocess and orthorectify the data to obtain orthorectified images. Based on the orthorectified images and a pre-defined SAM model, a sea ice classification training dataset is constructed. A pre-built Ice-Unet model is trained using this dataset, and the orthorectified images are input into the trained Ice-Unet model to output sea ice category information corresponding to the orthorectified images. The target reference temperature for the target type of water area and sea ice is determined based on the sea ice category information. A sea ice concentration inversion model is then constructed based on the target reference temperature to quantitatively evaluate the accuracy of sea ice concentration inversion. This allows for the acquisition of high-precision sea ice category and concentration data, enabling simple route planning and providing reliable reference data for regional Arctic sea ice changes. This solves the problems of existing technologies in achieving rapid reporting and on-site support of sea ice information in Arctic waters.

[0033] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0034] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0035] Figure 1 This is a flowchart illustrating a method for classifying and inverting sea ice concentration using UAV optical and thermal infrared data, provided according to an embodiment of this application.

[0036] Figure 2 A schematic diagram of the execution logic of a UAV optical and thermal infrared data sea ice classification and concentration inversion method provided for one embodiment of this application;

[0037] Figure 3 A schematic diagram of ISAT-SAM dataset annotation software and annotation results is provided as an embodiment of this application;

[0038] Figure 4 A schematic diagram of an Ice-Unet sea ice classification model provided for one embodiment of this application;

[0039] Figure 5 A schematic diagram illustrating the sea ice classification results of a different method provided for one embodiment of this application;

[0040] Figure 6 A schematic diagram illustrating the verification of sea ice results using a different method, as provided in one embodiment of this application;

[0041] Figure 7 A schematic diagram of sea ice concentration inversion results provided for one embodiment of this application;

[0042] Figure 7 Image (a) is a schematic diagram of optical data provided in one embodiment of this application;

[0043] Figure 7 (b) is a schematic diagram of thermal infrared data provided in an embodiment of this application;

[0044] Figure 7 (c) is a schematic diagram of an optical data sea ice concentration inversion result provided in an embodiment of this application;

[0045] Figure 7 (d) is a schematic diagram of the sea ice concentration inversion result from thermal infrared data provided in an embodiment of this application;

[0046] Figure 8 A schematic diagram illustrating the verification of sea ice concentration results is provided as an embodiment of this application;

[0047] Figure 8 (a) is a schematic diagram comparing optical and ASI density according to an embodiment of this application;

[0048] Figure 8 (b) is a schematic diagram showing a comparison of thermal infrared and ASI density according to an embodiment of this application;

[0049] Figure 8 (c) is a schematic diagram of optical and thermal infrared density comparison provided in an embodiment of this application;

[0050] Figure 9 This is an example diagram of a UAV optical and thermal infrared data sea ice classification and concentration inversion device according to an embodiment of this application;

[0051] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0052] Among them, 10-UAV optical and thermal infrared data sea ice classification and concentration inversion device; 100-preprocessing module, 200-classification module, 300-inversion module; 1001-memory, 1002-processor, 1003-communication interface. Detailed Implementation

[0053] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0054] The following describes, with reference to the accompanying drawings, a method for classifying and inverting sea ice concentration using UAV optical and thermal infrared data according to embodiments of this application. To address the problems mentioned in the background, this application provides a method for sea ice classification and concentration inversion using UAV optical and thermal infrared data. This method acquires optical and thermal infrared data from a target UAV, preprocesses and orthorectifies the data to obtain orthorectified images. Based on the orthorectified images and a pre-defined SAM model, a sea ice classification training dataset is constructed. A pre-built Ice-Unet model is trained using this dataset, and the orthorectified images are input into the trained Ice-Unet model to output sea ice category information corresponding to the orthorectified images. The target reference temperature for the target type of water body and sea ice is determined based on the sea ice category information. A sea ice concentration inversion model is then constructed based on the target reference temperature to quantitatively evaluate the accuracy of sea ice concentration inversion. This allows for the acquisition of high-precision sea ice category and concentration data, facilitating simple route planning and providing reliable reference data for regional Arctic sea ice changes. This solves the problems of existing technologies' difficulty in rapidly reporting and providing on-site support for sea ice information in Arctic waters.

[0055] Specifically, Figure 1 A flowchart illustrating a method for classifying and inverting sea ice concentration using UAV optical and thermal infrared data, provided in this application embodiment.

[0056] like Figure 1 As shown, the method for sea ice classification and concentration inversion based on UAV optical and thermal infrared data includes the following steps:

[0057] In step S101, optical and thermal infrared data of the target UAV are acquired, and the optical and thermal infrared data are preprocessed and orthorectified to obtain orthorectified images corresponding to the optical and thermal infrared data.

[0058] The embodiments of this application first need to acquire remote sensing data (i.e., optical and thermal infrared data) from UAVs and satellites, and improve the reliability of the data through orthorectification, geometric correction, and radiometric calibration. In view of the influence of geometric factors such as flight trajectory and terrain undulation on UAV and remote sensing satellite imagery, terrain correction is performed on UAV imagery based on ground control points and carrier pose to obtain orthorectified imagery.

[0059] In actual implementation, the embodiments of this application can perform radiometric calibration on UAV infrared images and use the radiative transfer equation to eliminate the influence of absorption and scattering of radiation signals between ground objects and sensors, calculate the true surface temperature, and at the same time, combine the placed standard radiation source to correct the measurement error of the sensor caused by changes in ambient temperature and humidity.

[0060] To address the issue of spatial misalignment in multi-source data, embodiments of this application can perform coarse spatial alignment based on high-precision pose data from satellites and UAVs combined with preprocessed orthophotos. Subsequently, based on computer vision strategies, corresponding points on multimodal data are extracted to construct feature matching and find geometric transformation models between multi-source images. Based on the transformation models, the multi-source images are mapped to the same spatial coordinate system. Furthermore, embodiments of this application also require fine-tuning of local registration based on mutual information registration principles to obtain registered UAV optical and thermal infrared images, thereby providing high-quality data for subsequent sea ice information extraction.

[0061] Optionally, in one embodiment of this application, acquiring optical and thermal infrared data of the target UAV, and preprocessing and orthorectifying the optical and thermal infrared data to obtain an orthorectified image corresponding to the optical and thermal infrared data includes: performing geometric correction and radiometric calibration operations on the optical and thermal infrared data to obtain corresponding calibration results; constructing a corresponding digital elevation model based on the calibration results, and performing orthorectification operations on the optical and thermal infrared data according to the digital elevation model to obtain corresponding orthorectified data; and performing orthorectified stitching processing on the orthorectified data to obtain an orthorectified image.

[0062] It should be noted that the specific steps for preprocessing and orthorectifying the optical and thermal infrared data in the embodiments of this application are as follows:

[0063] Step 1, Geometric Correction

[0064] First, embodiments of this application can establish parameter models of ground control points and image sensors based on ground control points deployed for on-site sea ice observation, thereby determining the accurate location of image pixels in geographic space. Second, embodiments of this application can obtain camera exterior orientation parameters, such as position and attitude angles, through GPS and IMU data from UAVs, to establish a precise relationship between images and ground coordinates. Finally, embodiments of this application can eliminate offset errors caused by terrain undulations through terrain correction.

[0065] In the specific implementation process, embodiments of this application may choose a second-order polynomial to realize the coordinate transformation between the original image and the corrected image. After the coordinate transformation, the pixel center position usually changes. Therefore, embodiments of this application need to resample the original image according to certain rules based on the position of each pixel in the output image in the original image, and establish a new raster matrix by recalculating the raster values. Embodiments of this application may choose bilinear interpolation to perform resampling to assign grayscale values ​​to the output image pixels of the distorted image to realize the geometric correction of the UAV image data, and use it for subsequent sea ice feature extraction and parameter inversion.

[0066] Step 2, Radiation Calibration

[0067] Radiometric calibration of UAV imagery involves converting the raw digital quantization (DN) values ​​into radiance, as detailed below:

[0068] (1) The thermal infrared sensor carried by the UAV senses the infrared photons radiated by ground objects, and converts them into electrical signals and quantizes them into DN values ​​based on the calibration coefficients of the sensor, including radiation gain and offset.

[0069] (2) Using the radiative transfer equation, the effects of absorption and scattering of radiation signals between ground objects and sensors are eliminated to obtain the radiance of the sea ice surface;

[0070] (3) Based on the radiative transfer model, the radiance is converted into luminance temperature, and the actual surface temperature is calculated. The measurement error caused by changes in ambient temperature and humidity of the sensor is corrected by combining the standard radiation source placed during the field observation, and the final calibration result is output.

[0071] Step 3: Construct the DEM (Digital Elevation Model)

[0072] After acquiring UAV data and performing steps 1 and 2, embodiments of this application can select image pairs with sufficient overlap, extract feature points from the images using the SIFT algorithm in the feature extraction algorithm, find corresponding feature points in adjacent images using a matching algorithm, and calculate the three-dimensional spatial coordinates of these feature points using triangulation based on the principles of photogrammetry, thereby obtaining point cloud data and constructing point cloud data with sufficient density and accuracy to reflect the undulations of the terrain.

[0073] Step 4, Orthorectification

[0074] Orthorectification is based on the photogrammetric principle of central projection. It converts a centrally projected image into an orthophoto by eliminating geometric distortions in the image, as described below:

[0075] (1) The principle of geometric correction in step 1 can be adopted. Based on the UAV image, a collinearity equation can be established. The sensor parameters calculated by geometric correction in step 1 can be substituted into the equation. At the same time, the DEM and planar coordinates obtained in step 3 can be combined to link the image point coordinates with the actual geographic coordinates of the ground feature points.

[0076] (2) Perform image point coordinate reprojection and grayscale resampling. For each image point in the image, based on the established geometric correction model and DEM data, use bilinear interpolation to obtain a relatively smooth image, and finally calculate its new coordinates under orthophoto projection.

[0077] Step 5, Orthophoto splicing

[0078] Based on the above data preprocessing steps, embodiments of this application can import the processed drone data into corresponding software. For example, after correcting the data obtained by Pegasus V500, the data can be imported into the drone management software, and image data and corresponding POS data can be added. Then, an orthorectified image can be generated based on the feature extraction results and the generated point cloud.

[0079] In step S102, a sea ice classification training dataset is constructed based on the orthorectified image and the preset SAM model. The pre-constructed Ice-Unet model is then trained using the sea ice classification training dataset. The orthorectified image is then input into the trained Ice-Unet model to output the sea ice category information corresponding to the orthorectified image.

[0080] Furthermore, embodiments of this application can utilize the orthophoto data obtained above, construct a training dataset using SAM (SegmentAnything Model), and build an Ice-Unet network based on this dataset, such as... Figure 2 As shown, this is to achieve high-precision sea ice classification using optical images.

[0081] Optionally, in one embodiment of this application, training a pre-constructed Ice-Unet model using a sea ice classification training dataset includes: constructing an Ice-Unet model based on a preset backbone feature extraction network, an enhanced feature extraction network, and a prediction network; inputting training data from the sea ice classification training dataset into the backbone feature extraction network of the Ice-Unet model to perform multiple convolution and max pooling operations on the training data to generate multiple preliminary effective feature layers; inputting the multiple preliminary effective feature layers into the enhanced feature extraction network to perform upsampling and feature fusion processing on the multiple preliminary effective feature layers to obtain a target fused effective feature layer; and using a target convolutional layer to perform channel adjustment operations on the target fused effective feature layer to obtain the trained Ice-Unet model.

[0082] As one possible approach, embodiments of this application construct a sea ice classification training dataset based on orthorectified images and a pre-defined SAM model, and the specific steps for training the pre-constructed Ice-Unet model using the sea ice classification training dataset are as follows:

[0083] Step 1: Sample selection based on the SAM model

[0084] The embodiments of this application can employ an interactive sea ice annotation method based on the SAM segmentation model to construct a sea ice classification dataset. Those skilled in the art should understand that ISAT-SAM is an interactive semi-automatic image segmentation and annotation tool based on SAM, supporting both manual polygon drawing and semi-automatic annotation modes. In semi-automatic annotation mode, left-clicking the area to be annotated serves as a prompt for ISAT-SAM, which automatically identifies similar regions with generally good recognition results. If an incorrect identification occurs, right-clicking can cancel the annotation of that point. Finally, multiple clicks are used to complete the annotation of the entire image. Figure 3 As shown; this tool generates a JSON tag file by default, and then uses its format conversion function to convert the JSON format to PNG format to obtain the required tag file;

[0085] Step 2: Ice-Unet Model Construction

[0086] The Ice-Unet model in this application embodiment is as follows: Figure 4 As shown, the model can be divided into three parts: a backbone feature extraction network, a reinforcement feature extraction network, and a prediction network, as described below:

[0087] (1) The first part is the backbone feature extraction network. This part is similar to the VGG16 network structure. It obtains five preliminary effective feature layers through multiple convolutions and max pooling, which serve as the input to strengthen the feature extraction network. The specific process is as follows:

[0088] 1) The input sea ice image has 3 channels. First, perform two 3×3 convolutions with 64 output channels to obtain the first preliminary effective feature layer.

[0089] 2) Perform max pooling and two 3×3 convolutions with 128 output channels in sequence to obtain the second preliminary effective feature layer;

[0090] 3) Perform max pooling and three 3×3 convolutions with 256 output channels to obtain the third preliminary effective feature layer;

[0091] 4) Perform max pooling and three 3×3 convolutions with 512 output channels to obtain the fourth preliminary effective feature layer;

[0092] 5) Perform max pooling and three 3×3 convolutions with 512 output channels to obtain the fifth preliminary effective feature layer;

[0093] In the above process, the kernel and stride of the max pooling convolution are both set to 2, so that the width and height of the image are halved while the number of channels remains unchanged.

[0094] (2) The second part is to strengthen the feature extraction network. This part upsamples and fuses the five preliminary effective feature layers to obtain the effective feature layer that fuses all features (i.e., the target fusion effective feature layer). The fifth preliminary effective feature layer is upsampled by 2 times and then concatenated with the fourth preliminary effective feature layer, and then convolved twice to obtain the first strengthened feature layer. The first strengthened feature layer is upsampled by 2 times and then concatenated with the third preliminary effective feature layer, and then convolved twice to obtain the second strengthened feature layer. The second strengthened feature layer is upsampled by 2 times and then concatenated with the second preliminary effective feature layer, and then convolved twice to obtain the third strengthened feature layer. The third strengthened feature layer is upsampled by 2 times and then concatenated with the first preliminary effective feature layer, and then convolved twice to obtain the final effective feature layer, which has the same size as the original input image and has 64 channels.

[0095] (3) The third part is the prediction network. This part uses a 1×1 convolution to adjust the channels of the final effective feature layer so that the number of channels equals the number of targets to be classified (in this embodiment, the number of targets to be classified is 4, namely ice, meltwater pool, water, and boat). This is equivalent to classifying each feature point of the feature layer, such as... Figure 5 As shown, this enables pixel-by-pixel semantic segmentation of sea ice images.

[0096] Compared with other classification algorithms, such as Figure 6 As shown, the Ice-Unet model in this application embodiment has superior sea ice classification accuracy.

[0097] In step S103, the target reference temperature of the target type of water area and the target type of sea ice is determined according to the sea ice category information. Based on the target reference temperature, a sea ice concentration inversion model is constructed to quantitatively evaluate the accuracy of sea ice concentration inversion.

[0098] Furthermore, embodiments of this application can use the sea ice category information obtained above as auxiliary information for reference temperature selection, obtain reference temperatures for open water and sea ice, and construct a sea ice concentration inversion model based on thermal infrared data, thereby quantitatively evaluating the accuracy of sea ice concentration inversion.

[0099] Therefore, the embodiments of this application introduce the Ice-Unet deep learning model to achieve high-precision sea ice classification, providing auxiliary information for the selection of thermal infrared sea ice system points. Based on this, the embodiments of this application establish a sea ice concentration inversion model based on thermal infrared data to achieve high-precision sea ice concentration inversion. Thus, the sea ice information extraction research based on unmanned observation proposed in the embodiments of this application can obtain high-precision sea ice category and concentration data, thereby enabling simple route planning and providing reference data for regional Arctic sea ice changes.

[0100] Optionally, in one embodiment of this application, the target reference temperature for the target type of water area and the target type of sea ice is determined based on sea ice category information, and a sea ice concentration inversion model is constructed based on the target reference temperature to quantitatively evaluate the sea ice concentration inversion accuracy. This includes: determining the target pixel grid corresponding to each pixel in the orthorectified image and dividing the target pixel grid into multiple sub-grids; selecting the target percentile sea ice surface temperature in each of the multiple sub-grids, and determining the preliminary sea ice reference temperature for the target type of sea ice based on the target percentile sea ice surface temperature; and determining the preliminary sea ice reference temperature based on the preliminary sea ice reference temperature and a preset reference temperature. A linear regression strategy is used to determine the final reference temperature of the target type of sea ice. Based on a preset pixel-by-pixel sliding window strategy, multiple overlay operations are performed on each pixel to obtain the final reference temperature of the target type of sea ice corresponding to each overlay operation. The target reference temperature of the target type of sea ice for each pixel is then calculated based on the final reference temperature of the target type of sea ice for each overlay operation. The sea ice concentration for each pixel is calculated based on the target reference temperature of the target type of water area and the target type of sea ice, as well as the sea ice concentration inversion model. The sea ice concentration is then projected to obtain the corresponding projection result, and the accuracy of the sea ice concentration inversion is quantitatively evaluated based on the projection result.

[0101] In the embodiments of this application, the calculation of sea ice concentration mainly uses linear equations to solve for pixels. A crucial parameter is surface temperature, which is divided into the reference temperature of open water (i.e., the target reference temperature of the target type of water) t. pwater and the reference temperature of sea ice (i.e., the target reference temperature for the target type of sea ice) t pice , which is the temperature when the current pixel is completely covered by water or sea ice, where the location of seawater and sea ice can be obtained using a sea ice classification method based on the Ice-Unet model.

[0102] In practical implementation, embodiments of this application can select the temperature corresponding to the seawater category in thermal infrared data from the sea ice category information, and select the minimum mean square error of the temperature as the reference temperature for the seawater category; however, ice surface temperature is greatly affected by air temperature and varies considerably, therefore it is not possible to select a fixed Arctic temperature value as a reference. To cope with local variations, each pixel is assigned an independent t pice .

[0103] In the specific processing, in order to obtain the density inversion result of the target resolution from the high-resolution raw UAV data, the embodiments of this application can divide the raw image into a 20×20 pixel grid. Each grid represents a pixel of the target resolution density inversion result. In order to determine the sea ice reference temperature of each pixel, each grid can be further subdivided into 5×5 blocks (i.e., multiple sub-grids). In each block, the 25th percentile sea ice surface temperature value (i.e., the target percentile sea ice surface temperature) is selected as the preliminary sea ice reference temperature. Then, the final reference temperature is determined by a linear regression formula, as shown in the following formula:

[0104] t pice (x,y)=a·x+b·y+c

[0105] Where x and y are the coordinates of each pixel, and a, b, and c are regression coefficients.

[0106] Optionally, in one embodiment of this application, the mathematical expression for the sea ice concentration inversion model is:

[0107]

[0108] Where SIC represents sea ice concentration; t pwate Indicates the target reference temperature of the target type of water body; t pice The target reference temperature indicates the target type of sea ice; IST indicates the sea ice surface temperature.

[0109] Subsequently, embodiments of this application may employ a pixel-by-pixel sliding window method, covering each pixel 20 times; then, the average value of the 20 iterations is selected as t. pice The Sea Ice Concentration (SIC) was estimated using a linear model, and the mathematical expression of this sea ice concentration inversion model is as follows:

[0110]

[0111] Where SIC represents sea ice concentration; t pwate Indicates the target reference temperature of the target type of water body; t pice The target reference temperature indicates the target type of sea ice; IST indicates the sea ice surface temperature.

[0112] Finally, embodiments of this application can project the SIC results onto a 1m grid and combine the data from the same orbit to generate the final sea ice density inversion product.

[0113] It should be noted that, Figure 7 and Figure 8 These are schematic diagrams showing the sea ice concentration inversion results and the sea ice concentration result verification results, respectively. Figure 7 (a) in the diagram is a schematic diagram of optical data. Figure 7 (b) in the diagram is a schematic diagram of thermal infrared data. Figure 7 (c) in the figure is a schematic diagram of the sea ice concentration inversion results from optical data. Figure 7 (d) in the figure is a schematic diagram of the sea ice concentration inversion results from thermal infrared data. Figure 8 (a) in the diagram is a comparison of optical and ASI density. Figure 8 (b) in the diagram is a comparison of thermal infrared and ASI density. Figure 8 (c) in the diagram is a comparison of optical and thermal infrared density.

[0114] like Figure 7 (a)-(d) and Figure 8 As shown in (a)-(c), the embodiments of this application are capable of reliable sea ice concentration inversion and can achieve quantitative assessment of the accuracy of sea ice concentration inversion.

[0115] In summary, the embodiments of this application first obtain registered UAV optical and thermal infrared images through preprocessing techniques such as orbit correction, radiometric calibration, and geometric correction, providing high-quality data for subsequent sea ice information extraction to improve the reliability of multimodal data. Simultaneously, based on a deep learning model-based sea ice feature learning method, the Ice-Unet model is introduced to achieve high-precision sea ice classification results, providing auxiliary information for selecting thermal infrared sea ice system points. This allows for the selection of surface temperature values ​​corresponding to open water and fixed ice in the sea ice classification to construct system point values ​​for thermal infrared data. Based on this, a sea ice concentration inversion model based on thermal infrared data is established to obtain sea ice concentration inversion results, achieving high-precision sea ice concentration inversion and quantitatively assessing the uncertainty of sea ice concentration.

[0116] Therefore, the embodiments of this application address the problems of sea ice scene classification and density inversion. Based on high-resolution UAV data, and by fusing visible light and thermal infrared data, a deep learning feature extraction and fusion model is constructed to achieve sea ice classification and density inversion, thereby solving the problems of rapid reporting of sea ice information and on-site support in Arctic seawater.

[0117] The method for sea ice classification and concentration inversion based on UAV optical and thermal infrared data proposed in this application involves acquiring optical and thermal infrared data of a target UAV, preprocessing the data, and performing orthorectified stitching to obtain orthorectified images. Based on the orthorectified images and a pre-defined SAM model, a sea ice classification training dataset is constructed. A pre-built Ice-Unet model is then trained using this dataset. The orthorectified images are input into the trained Ice-Unet model to output sea ice category information corresponding to the orthorectified images. The target reference temperature for the target type of water body and sea ice is determined based on the sea ice category information. A sea ice concentration inversion model is then constructed based on the target reference temperature to quantitatively evaluate the accuracy of sea ice concentration inversion. This allows for the acquisition of high-precision sea ice category and concentration data, facilitating simple route planning and providing reliable reference data for regional Arctic sea ice changes.

[0118] Secondly, with reference to the accompanying drawings, a device for classifying and inverting sea ice concentration based on UAV optical and thermal infrared data according to an embodiment of this application is described.

[0119] Figure 9 This is a block diagram of a UAV optical and thermal infrared data sea ice classification and concentration inversion device according to an embodiment of this application.

[0120] like Figure 9 As shown, the UAV optical and thermal infrared data sea ice classification and concentration inversion device 10 includes: a preprocessing module 100, a classification module 200, and an inversion module 300.

[0121] The preprocessing module 100 is used to acquire optical and thermal infrared data of the target UAV, and to perform preprocessing and orthorectification stitching operations on the optical and thermal infrared data to obtain orthorectified images corresponding to the optical and thermal infrared data.

[0122] The classification module 200 is used to construct a sea ice classification training dataset based on orthorectified images and a preset SAM model, and to train a pre-constructed Ice-Unet model using the sea ice classification training dataset. The orthorectified images are then input into the trained Ice-Unet model to output sea ice category information corresponding to the orthorectified images.

[0123] The inversion module 300 is used to determine the target reference temperature of the target type of water area and the target type of sea ice based on the sea ice category information, and to construct a sea ice concentration inversion model based on the target reference temperature, so as to quantitatively evaluate the accuracy of sea ice concentration inversion using the sea ice concentration inversion model.

[0124] Optionally, in one embodiment of this application, the preprocessing module 100 includes: a radiometric calibration unit, an orthorectification unit, and an orthorectification stitching unit.

[0125] The radiometric calibration unit is used to perform geometric correction and radiometric calibration operations on optical and thermal infrared data to obtain the corresponding calibration results.

[0126] The orthorectification unit is used to construct the corresponding digital elevation model based on the calibration results, and to perform orthorectification operations on the optical and thermal infrared data according to the digital elevation model to obtain the corresponding orthorectified data.

[0127] The orthorectified stitching unit is used to perform orthorectified stitching processing on orthorectified data to obtain orthorectified images.

[0128] Optionally, in one embodiment of this application, the classification module 200 includes: a modeling unit, a backbone feature extraction unit, an enhanced feature extraction unit, and a prediction unit.

[0129] The modeling unit is used to construct the Ice-Unet model based on a preset backbone feature extraction network, enhanced feature extraction network, and prediction network.

[0130] The backbone feature extraction unit is used to input the training data from the sea ice classification training dataset into the backbone feature extraction network in the Ice-Unet model, so as to perform multiple convolution and max pooling operations on the training data to generate multiple preliminary effective feature layers.

[0131] The enhanced feature extraction unit is used to input multiple preliminary effective feature layers into the enhanced feature extraction network to perform upsampling and feature fusion processing on the multiple preliminary effective feature layers to obtain the target fused effective feature layer.

[0132] The prediction unit is used to perform channel adjustment operations on the target fusion effective feature layer using the target convolutional layer to obtain the trained Ice-Unet model.

[0133] Optionally, in one embodiment of this application, the inversion module 300 includes: a partitioning unit, a selection unit, a determination unit, a coverage unit, and a quantitative evaluation unit.

[0134] The partitioning unit is used to determine the target pixel grid corresponding to each pixel in the orthorectified image and to divide the target pixel grid into multiple sub-grids.

[0135] Select a cell to select the target percentile sea ice surface temperature in each of multiple subgrids, and determine the preliminary sea ice reference temperature for the target type of sea ice based on the target percentile sea ice surface temperature.

[0136] The determination unit is used to determine the final reference temperature of the target type of sea ice based on the preliminary sea ice reference temperature and a preset linear regression strategy.

[0137] The overlay unit is used to perform multiple overlay operations on each pixel based on a preset pixel-by-pixel sliding window strategy to obtain the final reference temperature of the target type of sea ice corresponding to each overlay operation, and to calculate the target reference temperature of the target type of sea ice corresponding to each pixel based on the final reference temperature of the target type of sea ice corresponding to each overlay operation.

[0138] The quantitative assessment unit is used to calculate the sea ice concentration corresponding to each pixel based on the target reference temperature and sea ice concentration inversion model of the target type water area and the target type sea ice, and to project the sea ice concentration to obtain the corresponding projection result. The accuracy of the sea ice concentration inversion is quantitatively assessed through the projection result.

[0139] Optionally, in one embodiment of this application, the mathematical expression for the sea ice concentration inversion model is:

[0140]

[0141] Where SIC represents sea ice concentration; t pwate Indicates the target reference temperature of the target type of water body; t pice The target reference temperature indicates the target type of sea ice; IST indicates the sea ice surface temperature.

[0142] It should be noted that the foregoing explanation of the embodiment of the UAV optical and thermal infrared data sea ice classification and concentration inversion method also applies to the UAV optical and thermal infrared data sea ice classification and concentration inversion device of this embodiment, and will not be repeated here.

[0143] The UAV optical and thermal infrared data sea ice classification and concentration inversion device proposed in this application includes a preprocessing module for acquiring optical and thermal infrared data of the target UAV and performing preprocessing and orthorectification stitching operations on the optical and thermal infrared data to obtain orthorectified images corresponding to the optical and thermal infrared data; a classification module for constructing a sea ice classification training dataset based on the orthorectified images and a preset SAM model, training a pre-constructed Ice-Unet model using the sea ice classification training dataset, and inputting the orthorectified images into the trained Ice-Unet model to output sea ice category information corresponding to the orthorectified images; and an inversion module for determining the target type water area and the target type sea ice based on the sea ice category information, and constructing a sea ice concentration inversion model based on the target reference temperature to quantitatively evaluate the sea ice concentration inversion accuracy using the sea ice concentration inversion model. This allows for the acquisition of high-precision sea ice category and concentration data, enables simple route planning, and provides reliable reference data for regional Arctic sea ice changes.

[0144] Figure 10 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0145] The memory 1001, the processor 1002, and the computer program stored on the memory 1001 and capable of running on the processor 1002.

[0146] When the processor 1002 executes the program, it implements the UAV optical and thermal infrared data sea ice classification and concentration inversion method provided in the above embodiments.

[0147] Furthermore, electronic devices also include:

[0148] Communication interface 1003 is used for communication between memory 1001 and processor 1002.

[0149] The memory 1001 is used to store computer programs that can run on the processor 1002.

[0150] The memory 1001 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0151] If the memory 1001, processor 1002, and communication interface 1003 are implemented independently, then the communication interface 1003, memory 1001, and processor 1002 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized into address buses, data buses, control buses, etc. For ease of representation, Figure 10 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0152] Optionally, in a specific implementation, if the memory 1001, processor 1002, and communication interface 1003 are integrated on a single chip, then the memory 1001, processor 1002, and communication interface 1003 can communicate with each other through an internal interface.

[0153] The processor 1002 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0154] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for classifying and inverting sea ice concentration based on UAV optical and thermal infrared data.

[0155] This application also provides a computer program product, including a computer program, which, when executed, is used to implement the above-described method for classifying and inverting sea ice concentration based on UAV optical and thermal infrared data.

[0156] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0157] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0158] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0159] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0160] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0161] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0162] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0163] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. An unmanned aerial vehicle optical and thermal infrared data sea ice classification and concentration retrieval method, characterized in that, The method comprises the following steps: Obtaining optical and thermal infrared data of a target unmanned aerial vehicle, and pre-processing and ortho-mosaicking the optical and thermal infrared data to obtain ortho-rectified images corresponding to the optical and thermal infrared data; Wherein, obtaining optical and thermal infrared data of a target unmanned aerial vehicle, and pre-processing and ortho-mosaicking the optical and thermal infrared data to obtain ortho-rectified images corresponding to the optical and thermal infrared data, comprises: Performing geometric correction and radiation calibration on the optical and thermal infrared data to obtain corresponding calibration results; Based on the calibration results, a corresponding digital elevation model is constructed, and ortho-rectification is performed on the optical and thermal infrared data according to the digital elevation model to obtain corresponding ortho-rectified data; Ortho-mosaic processing is performed on the ortho-rectified data to obtain the ortho-rectified images; Based on the ortho-rectified images and a pre-set SAM model, a sea ice classification training data set is constructed, and a pre-constructed Ice-Unet model is trained through the sea ice classification training data set, and the ortho-rectified images are input into the trained Ice-Unet model to output sea ice category information corresponding to the ortho-rectified images; Training a pre-constructed Ice-Unet model through the sea ice classification training data set comprises: Based on a pre-set backbone feature extraction network, an enhanced feature extraction network and a prediction network, the Ice-Unet model is constructed; The training data in the sea ice classification training data set is input into the backbone feature extraction network in the Ice-Unet model to perform multiple convolution and maximum pooling operations on the training data to generate multiple preliminary effective feature layers; The multiple preliminary effective feature layers are input into the enhanced feature extraction network to perform up-sampling and feature fusion processing on the multiple preliminary effective feature layers to obtain target fusion effective feature layers; The target fusion effective feature layers are subjected to channel adjustment operation by a target convolution layer to obtain the trained Ice-Unet model; According to the sea ice category information, the target reference temperature of the target type water area and the target type sea ice is determined, and based on the target reference temperature, a sea ice density inversion model is constructed to quantitatively evaluate the sea ice density inversion accuracy by using the sea ice density inversion model; According to the sea ice category information, the target reference temperature of the target type water area and the target type sea ice is determined, and based on the target reference temperature, a sea ice density inversion model is constructed to quantitatively evaluate the sea ice density inversion accuracy by using the sea ice density inversion model, comprising: Determine the target pixel grid corresponding to each pixel point in the ortho-rectified image, and divide the target pixel grid into multiple sub-grids; In each of the multiple sub-grids, the target percentile sea ice surface temperature is selected, and the preliminary sea ice reference temperature of the target type sea ice is determined according to the target percentile sea ice surface temperature; Based on the preliminary sea ice reference temperature and a pre-set linear regression strategy, the final reference temperature of the target type sea ice is determined; According to the target type water area and the target reference temperature and the sea ice density inversion model of the target type sea ice, the sea ice density corresponding to each pixel is calculated, and the sea ice density is projected to obtain a corresponding projection result, and the quantitative evaluation of the sea ice density inversion accuracy is performed through the projection result. The mathematical expression of the sea ice density inversion model is: It comprises: wherein SIC represents the sea ice concentration; t pwate represents a target reference temperature of the target type water area; t pice represents a target reference temperature of the target type sea ice; IST represents the sea ice surface temperature.

2. An unmanned aerial vehicle optical and thermal infrared data sea ice classification and concentration retrieval apparatus, characterized in that, A preprocessing module is configured to obtain optical and thermal infrared data of a target unmanned aerial vehicle, and to perform preprocessing and orthorectification splicing operations on the optical and thermal infrared data to obtain orthorectified images corresponding to the optical and thermal infrared data. A classification module is configured to construct a sea ice classification training data set based on the orthorectified images and a preset SAM model, and to train a pre-constructed Ice-Unet model through the sea ice classification training data set, and to input the orthorectified images into the trained Ice-Unet model to output sea ice category information corresponding to the orthorectified images. An inversion module is configured to determine target type water areas and target reference temperatures of target type sea ice according to the sea ice category information, and to construct a sea ice density inversion model based on the target reference temperatures to quantitatively evaluate the sea ice density inversion accuracy using the sea ice density inversion model. The preprocessing module comprises: A radiation calibration unit is configured to perform geometric correction and radiation calibration operations on the optical and thermal infrared data to obtain corresponding calibration results. An orthorectification unit is configured to construct a corresponding digital elevation model based on the calibration results, and to perform orthorectification operations on the optical and thermal infrared data according to the digital elevation model to obtain corresponding orthorectified data. An orthorectification splicing unit is configured to perform orthorectification splicing processing on the orthorectified data to obtain the orthorectified images. The classification module comprises: A modeling unit is configured to construct the Ice-Unet model based on a preset backbone feature extraction network, a strengthened feature extraction network, and a prediction network. A backbone feature extraction unit is configured to input training data in the sea ice classification training data set into a backbone feature extraction network in the Ice-Unet model to perform multiple convolution and maximum pooling operations on the training data to generate multiple preliminary effective feature layers. A strengthened feature extraction unit is configured to input the multiple preliminary effective feature layers into a strengthened feature extraction network to perform upsampling and feature fusion processing on the multiple preliminary effective feature layers to obtain target fusion effective feature layers. A prediction unit is configured to use a target convolution layer to perform channel adjustment operations on the target fusion effective feature layers to obtain the trained Ice-Unet model. The inversion module comprises: ​ The dividing unit is configured to determine a target pixel grid corresponding to each pixel point in the orthorectified image and divide the target pixel grid into a plurality of sub-grids. The selecting unit is configured to select a target percentile sea ice surface temperature in each of the plurality of sub-grids, and determine a preliminary sea ice reference temperature of the target type sea ice according to the target percentile sea ice surface temperature. The determining unit is configured to determine a final reference temperature of the target type sea ice based on the preliminary sea ice reference temperature and a preset linear regression strategy. The covering unit is configured to perform a plurality of covering operations on each pixel point based on a preset pixel-by-pixel sliding window strategy, to obtain a final reference temperature of the target type sea ice corresponding to each covering operation, and calculate a target reference temperature of the target type sea ice corresponding to each pixel point according to the final reference temperature of the target type sea ice corresponding to each covering operation. The quantitative evaluation unit is configured to calculate a sea ice density corresponding to each pixel according to the target type water area, the target reference temperature of the target type sea ice, and the sea ice density inversion model, project the sea ice density to obtain a corresponding projection result, and quantitatively evaluate the sea ice density inversion accuracy through the projection result. The mathematical expression of the sea ice density inversion model is: wherein SIC represents the sea ice concentration; t pwate represents a target reference temperature of the target type water area; t pice represents a target reference temperature of the target type sea ice; IST represents the sea ice surface temperature.

3. An electronic device, comprising: The method comprises the following steps: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the unmanned aerial vehicle optical and thermal infrared data sea ice classification and density inversion method according to claim 1.

4. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the unmanned aerial vehicle optical and thermal infrared data sea ice classification and density inversion method according to claim 1.

5. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the unmanned aerial vehicle optical and thermal infrared data sea ice classification and density inversion method according to claim 1.

Images