Geochemical remote sensing image recognition method, device, equipment, medium and program product

Abstract

The application provides a geochemical remote sensing image recognition method, device, equipment, medium and program product, which can be applied to the technical field of geochemical exploration and remote sensing image processing to solve the technical problem of improving the geochemical remote sensing image recognition efficiency. The geochemical remote sensing image recognition method comprises the following steps: in response to obtaining a hyperspectral image, inputting the hyperspectral image into a remote sensing image recognition model to obtain a remote sensing image recognition result. The remote sensing image recognition model is based on the activation function operation to output a down-sampling feature map, and after the cross-scale feature fusion operation of the down-sampling feature map and the feature map of the same level of the up-sampling operation layer are spliced, the FPN is embedded in different levels, the cross-scale feature fusion is realized, which is suitable for the characteristics of the multi-scale nature of the macroscopic morphology of the mineral vein and the microscopic spectral details of the mineral distribution in geochemical remote sensing, the fusion effect of the multi-source data is enhanced, the weak signal missed detection can be effectively avoided, and the recognition accuracy is improved.

Description

Technical Field

[0001] This application relates to the field of geochemical exploration and remote sensing image processing, and more specifically to a geochemical remote sensing image recognition method, apparatus, equipment, medium, and program product. Background Technology

[0002] Traditional geochemical exploration relies on manual sampling and laboratory analysis, which is costly and time-consuming. For example, methods such as X-ray fluorescence spectroscopy require on-site sampling. Existing remote sensing identification methods are not sensitive enough to weak anomaly signals and are easily affected by cloud cover and terrain noise. Image recognition suffers from problems of low accuracy or efficiency. Summary of the Invention

[0003] In view of the above problems, this application provides geochemical remote sensing image recognition methods, apparatus, equipment, media and program products to improve the efficiency of geochemical remote sensing image recognition.

[0004] According to the first aspect of this application, a geochemical remote sensing image recognition method is provided, comprising: in response to acquiring a hyperspectral image, inputting the hyperspectral image into a remote sensing image recognition model to obtain a remote sensing image recognition result; the remote sensing image recognition model is trained using hyperspectral image samples and labels in the following manner: performing downsampling operations multiple times on the hyperspectral image samples to extract image features and reduce the spatial size to obtain image coding features; the downsampling operation includes sequentially performing convolution operations, activation function operations, and max pooling operations; performing upsampling operations multiple times on the image coding features to restore spatial resolution and obtain image decoding features; the upsampling operation includes: outputting a downsampled feature map based on activation function operations, performing cross-scale feature fusion operations on the downsampled feature map and concatenating it with the feature map of the same level of upsampling operation layer, and performing convolution operations to fuse them to obtain image decoding features; performing convolution and classification operations on the image decoding features output by the last upsampling operation to output an image classification result; wherein, the cross-scale feature fusion operation includes: fusing features of different levels of downsampled feature maps according to the feature complexity of the downsampled feature maps.

[0005] According to an embodiment of this application, the downsampling operation includes: performing convolution, activation function, convolutional block attention module operation and max pooling operation in sequence; and concatenating the downsampled feature map with the feature map of the same level upsampling operation layer after performing cross-scale feature fusion operation, including: concatenating the feature map output by the activation function operation and convolutional block attention module operation in the downsampling operation with the feature map of the same level upsampling operation layer after performing cross-scale feature fusion operation.

[0006] According to an embodiment of this application, the feature fusion of the downsampled feature map based on the feature complexity of the downsampled feature map includes: constructing a multi-scale feature pyramid from the downsampled feature map; calculating the fusion weight of the multi-scale features based on the spectral feature entropy value and the spatial feature entropy value of the downsampled feature map; and fusing the multi-scale features based on the fusion weight.

[0007] According to an embodiment of this application, before performing multiple downsampling operations on hyperspectral image samples to extract image features and reduce spatial size, the method further includes: performing a preprocessing operation on the hyperspectral image, wherein the preprocessing operation includes at least one of atmospheric correction operation, terrain correction operation, and geometric correction operation.

[0008] According to an embodiment of this application, performing multiple downsampling operations on hyperspectral image samples to extract image features and reduce spatial size includes: using the Otsu algorithm to binarize the hyperspectral image and extracting a rough outline of the target region; and performing multiple downsampling operations on the hyperspectral image sample containing the rough outline to extract image features and reduce spatial size.

[0009] According to an embodiment of this application, inputting a hyperspectral image into a remote sensing image recognition model to obtain a remote sensing image recognition result includes: outputting a remote sensing image recognition result with geographic coordinates based on the geographic coordinates and relative attitude parameters at the time of hyperspectral image acquisition.

[0010] The second aspect of this application provides a geochemical remote sensing image recognition device, comprising: an image acquisition module for inputting a hyperspectral image into an image recognition module in response to acquiring the hyperspectral image; and an image recognition module for obtaining a remote sensing image recognition result based on the hyperspectral image and a remote sensing image recognition model. The remote sensing image recognition model is trained using hyperspectral image samples and labels using the following method: performing multiple downsampling operations on the hyperspectral image samples to extract image features and reduce their spatial size to obtain image encoding features; the downsampling operation includes sequentially performing convolution, activation function, and max pooling operations; performing multiple upsampling operations on the image encoding features to restore spatial resolution and obtain image decoding features; the upsampling operation includes: outputting a downsampled feature map based on the activation function operation; performing a cross-scale feature fusion operation on the downsampled feature map and concatenating it with feature maps from the same level of upsampling operation layer, and performing a convolution operation to fuse them to obtain image decoding features; performing convolution and classification operations on the image decoding features output from the last upsampling operation to output an image classification result; and the cross-scale feature fusion operation includes: fusing features from different levels of downsampled feature maps according to the feature complexity in the downsampled feature maps.

[0011] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0012] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0013] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0014] The above one or more embodiments have the following beneficial effects: They improve the efficiency of geochemical remote sensing image recognition through automated image extraction. By embedding FPNs at different levels, cross-scale feature fusion is achieved, which is suitable for the multi-scale characteristics of mineral distribution in geochemical remote sensing, which exhibits both macroscopic vein morphology and microscopic spectral details. This enhances the fusion effect on multi-source data and effectively avoids the missed detection of weak signals. Attached Figure Description

[0015] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0016] Figure 1 A flowchart illustrating a geochemical remote sensing image recognition method according to an embodiment of this application is shown schematically.

[0017] Figure 2 A flowchart illustrating the training of a geochemical remote sensing image recognition model according to an embodiment of this application is shown.

[0018] Figure 3 The diagram illustrates the architecture of a remote sensing image recognition model according to an embodiment of this application.

[0019] Figure 4 This illustration schematically shows a cross-scale feature fusion operation according to an embodiment of this application;

[0020] Figure 5 The diagram schematically illustrates the architecture of another remote sensing image recognition model according to an embodiment of this application;

[0021] Figure 6 This illustration schematically shows an architecture diagram of a convolutional block attention module according to an embodiment of this application;

[0022] Figure 7A schematic diagram illustrating the structure of a geochemical remote sensing image recognition device according to an embodiment of this application is shown; and

[0023] Figure 8 A block diagram schematically illustrates an electronic device suitable for implementing a geochemical remote sensing image recognition method according to an embodiment of this application. Detailed Implementation

[0024] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0025] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0026] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0027] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0028] Embodiments of this application provide a geochemical remote sensing image recognition method, comprising: in response to acquiring a hyperspectral image, inputting the hyperspectral image into a remote sensing image recognition model to obtain a remote sensing image recognition result; the remote sensing image recognition model is trained using hyperspectral image samples and labels according to the following method: performing downsampling operations multiple times on the hyperspectral image samples to extract image features and reduce the spatial size to obtain image coding features; the downsampling operation includes sequentially performing convolution operations, activation function operations, and max pooling operations; performing upsampling operations multiple times on the image coding features to restore spatial resolution and obtain image decoding features; the upsampling operation includes: outputting a downsampled feature map based on activation function operations, performing a cross-scale feature fusion operation on the downsampled feature map and concatenating it with the feature map of the same level of upsampling operation layer, and performing a convolution operation to fuse them to obtain image decoding features; performing convolution and classification operations on the image decoding features to output an image classification result; wherein, the cross-scale feature fusion operation includes: fusing the features of the downsampled feature map according to the feature complexity of the downsampled feature map.

[0029] Figure 1 A flowchart illustrating a geochemical remote sensing image recognition method according to an embodiment of this application is shown.

[0030] like Figure 1 As shown, the geochemical remote sensing image recognition method of this embodiment includes operations S110 to S120, and the transaction processing method can be executed in the following manner.

[0031] In operation S110, in response to acquiring a hyperspectral image, the hyperspectral image is input into the remote sensing image recognition model. In this embodiment, the hyperspectral image data can be acquired by a hyperspectral sensor. The hyperspectral image can contain the 400-2500nm band, and the specific acquisition device can include, for example, a satellite, manned aircraft, drone, or acquisition vehicle. Hyperspectral data typically contains tens to hundreds of continuous bands. In this embodiment, the required key band data can be selected according to different specific tasks. For example, in some embodiments of this application, 10 key bands that cover the characteristic absorption peaks of mineral spectra (such as sensitive bands of iron oxides and clay minerals) can be selected, which reduces the amount of computation while retaining the core spectral information. For example, for iron ore identification, sensitive band data such as 800-900nm and 2200nm can be selected.

[0032] Hyperspectral data contains band information in the spectral dimension and pixel location information in the spatial dimension. Spatial texture features describe the arrangement patterns between pixels, such as the continuity of mineral distribution and boundary roughness, and can be calculated from the spatial dimension of the hyperspectral image using algorithms such as GLCM (Gray-Level Co-occurrence Matrix) and LBP (Local Binary Pattern). Since there is a certain correlation between the spatial texture of minerals and their spectral features—for example, iron ore outcrops may exhibit an absorption peak of 800-900 nm in the spectrum and a "banded" or "clump-like" texture in space—this embodiment improves recognition accuracy by simultaneously extracting key band spectral features and image spatial texture based on key bands from hyperspectral data.

[0033] In operation S120, the remote sensing image recognition result is obtained. This embodiment obtains the remote sensing image recognition result by inputting a hyperspectral image into the remote sensing image recognition model. The remote sensing image recognition model can be a pre-trained model. When using a pre-trained remote sensing image recognition model for image recognition, it is not necessary to retrain the model.

[0034] Remote sensing image recognition models can use hyperspectral image samples and labels to... Figure 2 Obtained through training using Chinese methods.

[0035] Figure 2 A flowchart illustrating the training of a geochemical remote sensing image recognition model according to an embodiment of this application is shown.

[0036] like Figure 2 As shown, the training method for the geochemical remote sensing image recognition model includes operations S210 to S230. The geochemical remote sensing image recognition model in this embodiment is an improvement based on the U-net network, tailored to the characteristics of geochemical remote sensing image recognition tasks.

[0037] In operation S210, the hyperspectral image samples are downsampled multiple times to extract image features and reduce their spatial size, thereby obtaining image coding features. The downsampling operation includes sequentially performing convolution, activation function, and max pooling operations. For example, in this embodiment, four upsampling operations can be performed.

[0038] In operation S220, the image encoding features are upsampled multiple times to restore spatial resolution and obtain image decoding features. The upsampling operation includes: outputting a downsampled feature map based on an activation function operation; performing cross-scale feature fusion on the downsampled feature map; concatenating it with feature maps from the same level of upsampling operation layer; and then performing convolution to fuse them and obtain image decoding features. The cross-scale feature fusion operation includes fusing features from different levels of downsampled feature maps according to the feature complexity of the downsampled feature maps. For example, in this embodiment, when four upsampling operations are performed, four upsampling operations can be performed.

[0039] In operation S230, the image decoding features output from the last upsampling operation are subjected to convolution and classification operations to output the image classification result. For example, when four upsampling operations are performed, the image decoding features output from the fourth upsampling operation are subjected to convolution and classification operations to output the image classification result. The loss function in the embodiment can use a combination of cross-entropy loss. Alternatively, it can combine cross-entropy loss with Dice loss to enhance the focus on small target mineralization areas.

[0040] Figure 3 The diagram illustrates the architecture of a remote sensing image recognition model according to an embodiment of this application.

[0041] refer to Figure 3 The diagram shows the architecture of the remote sensing image recognition model. In operation S210, the hyperspectral image samples are downsampled multiple times to extract image features and reduce their spatial size, thus obtaining image coding features. When a hyperspectral image sample containing 10 key spectral bands and RGB images is used as the model input data, the model input has 13 channels. For example, in a certain task, the 10 key spectral bands include blue band, red band, near-infrared band, SWIR1, SWIR2, SWIR3, thermal infrared band, red-edge band, short-wave infrared extended band, and near-infrared extended band, with an input tensor dimension of [B, C, H, W]. Here, B is the batch size (e.g., default 16), C = 13 (10 spectral bands + 3 RGB channels), and H × W is the image size (e.g., 256 × 256).

[0042] In this embodiment, the downsampling operation includes sequential convolution, activation function, and max pooling operations. First, a level 1 downsampling operation is performed. In the level 1 convolution operation, preliminary feature extraction is performed on the 13-channel input, increasing the number of channels from 13 to 64 through two 3×3 convolutions. Then, downsampling is performed through an activation function operation and one 2×2 max pooling operation, halving the feature map size while doubling the number of channels. The feature map output from the max pooling operation is used as the input for the next downsampling operation. In levels 2-4 of the downsampling operation, two convolution operations, an activation function operation, and one 2×2 max pooling operation are performed using 3×3 convolution kernels. Figure 3 As shown, after four downsampling operations, the number of channels became 64, 128, 256, and 512, respectively.

[0043] In this embodiment, the upsampling operation includes: outputting a downsampled feature map based on an activation function operation, performing a cross-scale feature fusion operation on the downsampled feature map, concatenating it with the feature map of the same level upsampling operation layer, and then performing a convolution operation to fuse them to obtain image decoding features. For example... Figure 3The feature map size is doubled and the number of channels is halved by performing an up-convolution. In this embodiment, after the feature maps are stitched together, two rounds of 3×3 convolution operations and a ReLU operation are used to integrate the stitched features to form the image decoding features for this purpose. The cross-scale feature fusion operation can be implemented using the FPN (Feature Pyramid Network) architecture. Figure 4 The illustration shows a schematic diagram of a cross-scale feature fusion operation according to an embodiment of this application.

[0044] like Figure 4 As shown, the FPN architecture is used for cross-scale feature fusion of feature maps. The working principle of the FPN architecture includes downsampling and convolution operations on the input image to progressively extract features from the bottom layer to the top layer, forming a feature pyramid. The bottom layer features contain rich edge and texture details, while the top layer features contain stronger semantic information about the target category, forming feature maps at different resolution scales, corresponding to the multi-layer features on the left side of the figure. Starting from the high-level, low-resolution, semantically strong feature map, the size is enlarged by upsampling (such as interpolation) to match the resolution of the next higher-level feature map, establishing cross-scale connections, corresponding to the arrows on the right side of the figure that indicate the downward transmission of upper-level features. Lateral connections and fusion: The upsampled high-level features are laterally fused with the next higher-level features of the same resolution in the backbone network through element-wise addition or concatenation.

[0045] In this embodiment, the different levels of FPN in the downsampling operation can be increased or decreased based on the limited computing resources and the balance between target recognition and efficiency. For example, when the encoder depth increases or higher precision small target detection is required, the number of FPN levels can be increased; when computing resources are limited, processing speed is prioritized, or the target region has an abnormally uniform scale, the number of FPN levels can be reduced. Reducing the number of FPN levels can be achieved by directly concatenating the downsampled feature map at a certain level with the feature map of the upsampling operation layer at the same level without undergoing cross-scale feature fusion operations; that is, the downsampled feature maps at some levels can be implemented in the manner of ordinary U-net feature concatenation.

[0046] In this embodiment, images of different scales in the FPN can originate from the outputs of activation function operations at different levels of the U-Net encoder. After fusion using the FPN, these images are concatenated with feature maps from the same level of upsampling operation layer. This design directly utilizes multi-scale features at different levels output by the U-Net encoder (downsampling operation), enabling cross-scale feature fusion and enhancement via the FPN, thus allowing the model to better handle multi-scale targets. In this embodiment, multi-scale feature fusion based on the FPN captures multi-scale patterns ranging from details such as mineral grain texture to the global distribution of mineral belts by extracting features of different resolutions from the bottom up and using a top-down upsampling fusion path. This improves the accuracy of remote sensing image recognition. By embedding the FPN at different levels, cross-scale feature fusion is achieved, suitable for the multi-scale characteristics of mineral distribution in geochemical remote sensing, which exhibits both macroscopic vein morphology and microscopic spectral details. This enhances the fusion effect on multi-source data and effectively avoids the missed detection of weak signals.

[0047] In some embodiments of this application, the downsampling operation includes: sequentially performing convolution, activation function, convolutional block attention module operation, and max pooling operation; and concatenating the downsampled feature map with the feature map of the same level upsampling operation layer after performing cross-scale feature fusion operation, including: concatenating the feature map output by the activation function operation and convolutional block attention module operation in the downsampling operation with the feature map of the same level upsampling operation layer after performing cross-scale feature fusion operation.

[0048] Figure 5 This schematically illustrates the architecture of another remote sensing image recognition model according to an embodiment of this application; as shown... Figure 5 As shown, the features output by the convolutional block attention module in each downsampling operation are... Figure 1 On the one hand, the input is used for cross-scale feature fusion operations. On the other hand, the feature map output from the convolutional block attention module is used as the input for the next downsampling operation after max pooling.

[0049] Figure 6 The diagram schematically illustrates an architecture of a Convolutional Block Attention Module (CBAM) according to an embodiment of this application. Figure 6 As shown, the working principle of the convolutional block attention module operation is as follows.

[0050] The feature map from a certain layer is used as input features to the attention module of the convolutional block. In this embodiment, the feature map is output by the activation function operation in each layer's downsampling operation.

[0051] The convolutional block attention module internally comprises a channel attention module and a spatial attention module. The channel attention module generates channel descriptors through global average pooling and global max pooling, and calculates channel weights using a shared MLP layer. Spatial attention generates spatial descriptors by averaging and maxing the channel dimensions of the feature map, and then generates spatial weights through convolutional layers.

[0052] The formula for calculating channel attention weights can be expressed as: M c (F)=σ(W2δ(W1F avg )+W2'δ(W1'F max Spatial attention weights are generated through 7×7 convolutions. Here, W1, W2, and their variants W1' and W2' are learnable weight matrices, with W1 used to adjust the average pooling result F. avg Dimensionality reduction is performed. W1' is used to perform dimensionality reduction on the max pooling result F. max Dimensionality reduction is performed. W2 is used to restore the dimensionality-reduced features to the original number of channels, and W2' is used to restore the dimensionality-reduced features to the original number of channels.

[0053] The convolutional block attention module, through a dual-path design and a dimensionality reduction-up architecture, enables the model to adaptively focus on important channels and suppress irrelevant channels, thereby improving the representation ability of target features, while keeping the number of parameters under control. After processing by CBAM, key information feature maps are highlighted, and the output features are used for subsequent network layer processing.

[0054] In this embodiment, CBAM uses global pooling and MLP to calculate channel weights, suppressing noisy channels and highlighting key spectral features (such as the 860nm iron ion absorption peak). These enhanced features serve as input to the next layer, allowing subsequent convolutions to focus more on valuable spectral information. A spatial attention module focuses on specific regions, such as the boundaries of alteration zones in remote sensing images, thereby reducing irrelevant background interference. The features received by the next encoder layer already contain spatial focusing information, which helps extract more refined spatial texture features.

[0055] This embodiment embeds a CBAM attention module after each convolutional block to enhance the capture of weakly anomalous spectral features. The first 10 spectral band channels are enhanced with CBAM channel attention to improve the response to key feature bands, while the last 3 RGB channels are enhanced with spatial attention to strengthen texture edges. This embodiment significantly improves the model's ability to capture key features by introducing an adaptive attention mechanism in multi-scale feature processing. The convolutional block attention module operation and the cross-scale feature fusion operation are deeply integrated to form a unified overall scheme. Channel attention suppresses irrelevant channels (such as noise bands) and enhances target-related channels in the features of each scale of the FPN; spatial attention focuses on the spatial region where the target is located and suppresses background interference in the features of each scale of the FPN, thus achieving the ability to capture targets of different sizes and focus on key information at each scale, thereby improving image recognition accuracy.

[0056] In some embodiments of this application, the features of the downsampled feature map are fused according to the feature complexity of the downsampled feature map, including: constructing a multi-scale feature pyramid from the downsampled feature map; calculating the fusion weight of the multi-scale features based on the spectral feature entropy value and the spatial feature entropy value of the downsampled feature map; and fusing the multi-scale features according to the fusion weight.

[0057] In this embodiment, the spectral feature entropy can be calculated based on the spectral characteristics of 10 key bands. Specifically, the probability distribution of the feature values ​​of each spectral band is statistically analyzed to obtain the probability distribution p. i (i=1,2,…,10). Spectral feature entropy is calculated based on the information entropy formula (1). Spectral feature entropy reflects the complexity of spectral features. The higher the entropy value, the richer the spectral information. For example, if the characteristic band differences of altered minerals are significant, the channel contributes more to identification and should be given higher weight.

[0058] (1)

[0059] Spatial feature entropy calculates the gradient magnitude g for each of the three RGB channels based on the input RGB image data. j (j=R,G,B), the gradient distribution q is obtained. j Spatial feature entropy is calculated based on the information entropy formula, and the specific calculation formula is as shown in (2). Spatial feature entropy reflects the complexity of spatial features. The higher the entropy value, the more complex the spatial texture, such as the rich edge features of bare rock areas.

[0060] (2)

[0061] In this embodiment, the spectral feature weights are calculated as shown in equation (3).

[0062] (3)

[0063] When Hs >H v At that time, the spectral characteristics are more complex. As H approaches 1, the weights tilt towards spectral features. s <H v At that time, the spatial characteristics are more complex. As the weight approaches zero, the weights tilt towards spatial features.

[0064] In this embodiment, the spatial feature weights are calculated as shown in equation (4).

[0065] (4)

[0066] In this embodiment, for a certain scale of spectral features F s and spatial features F v The fusion can be performed according to formula (5).

[0067] (5)

[0068] The FPN architecture integrates multi-scale features from the bottom up, extracting features at different scales (1 / 4, 1 / 8, 1 / 16 resolution), and calculates H separately for each scale. s and Multi-scale features are upsampled from top to bottom and laterally connected. The fused features of each scale are superimposed according to weights, and the calculation formula is as shown in (6).

[0069] (6)

[0070] Where n is the number of scales. The weights for each scale, It can be determined through channel entropy or other attention mechanisms. Channel entropy describes the degree of disorder in the information distribution of a channel and can be calculated using the formula for information entropy. Taking hyperspectral images as an example, each pixel corresponds to the reflectance values ​​of 10 bands, forming a 10-dimensional channel feature vector. The feature values ​​of each channel are normalized to obtain the probability distribution of each channel, and then the channel entropy is calculated using the formula for information entropy.

[0071] This embodiment can be further optimized using an adaptive gating mechanism. Specifically, global average pooling is performed on the feature map output by the FPN to obtain the entropy scalar H. global The H is converted using the Sigmoid function. global Mapped to fusion weights , .

[0072] The final fusion formula combining CBAM and FPN features is (7).

[0073] (7)

[0074] in, The spectral features after attention enhancement of the CBAM channel; Multi-scale spatial texture features extracted by FPN.

[0075] This embodiment quantifies feature complexity through information entropy, achieving adaptive fusion of spectral and spatial information. It retains the segmentation advantages of U-Net while enhancing multi-scale representation capabilities through FPN, making it suitable for accurate identification tasks in geochemical remote sensing images.

[0076] In some embodiments of this application, before performing multiple downsampling operations on hyperspectral image samples to extract image features and reduce spatial size, the method further includes: performing preprocessing operations on the hyperspectral image, wherein the preprocessing operations include at least one of atmospheric correction operations, terrain correction operations, and geometric correction operations.

[0077] Atmospheric correction can be performed using the MODTRAN 6.0 model, with input parameters including aerosol optical depth (AOD), water vapor content (which can be measured in real time via a built-in miniature spectrometer), and solar zenith angle. Atmospheric up / down radiation is calculated based on the radiative transfer equation, eliminating the effects of Rayleigh and Mie scattering, and apparent reflectance data is output. The apparent reflectance data obtained through atmospheric correction in this embodiment provides a core data foundation for upgrading geochemical remote sensing from qualitative identification to quantitative analysis, improving identification efficiency and accuracy.

[0078] Terrain correction operations can use SRTMDEM data, combined with slope and aspect, to calculate terrain factors, specifically using the C-correction model.

[0079] Geometric correction can use attitude angles to rotate and correct images, eliminating image distortion caused by the movement of the acquisition equipment. Geometric correction can also combine GNSS coordinates and DEM data to correct pixel shifts caused by terrain undulations using a rational function model (RFM). In this embodiment, the data output through the above corrections eliminates interference from atmospheric and terrain factors, enabling accurate identification of geochemical remote sensing images.

[0080] In some embodiments of this application, the hyperspectral image samples are downsampled multiple times to extract image features and reduce spatial size, including: using the Otsu algorithm to binarize the hyperspectral image and extract the rough outline of the target region; and performing multiple downsampling operations on the hyperspectral image samples containing the rough outline to extract image features and reduce spatial size.

[0081] In this embodiment, the input image is first binarized using the Otsu algorithm to extract the rough outline of the target region, which serves as an auxiliary input for the remote sensing image recognition model. This can also act as a guiding signal for the attention mechanism and reduce background interference.

[0082] In some embodiments of this application, inputting a hyperspectral image into a remote sensing image recognition model to obtain a remote sensing image recognition result includes: outputting a remote sensing image recognition result with geographic coordinates based on the geographic coordinates and relative attitude parameters at the time of hyperspectral image acquisition.

[0083] The specific steps involved in implementing this embodiment are as follows.

[0084] 1. The hyperspectral sensor and positioning system are physically fixed together using a rigid bracket to form a spatiotemporally aligned data acquisition unit. The sensor acquisition time and positioning data timing are calibrated so that each frame of the spectral image corresponds to a set of positioning parameters. The positioning system receives positioning signals and outputs relevant positioning coordinates. Further attitude parameter measurements, such as pitch angle, roll angle, and yaw angle, are used for subsequent image geometric correction.

[0085] Each pixel of each frame of hyperspectral image has its geographic coordinates calculated using formula (8).

[0086] (8)

[0087] Where (X0,Y0,Z0) are the sensor center coordinates, R is the distance from the sensor to the ground, θ is the attitude angle, (Δx,Δy) is the pixel offset relative to the center, and h is the terrain elevation.

[0088] 2. Rotate the image using pitch and roll angles to eliminate distortion caused by the movement of the acquisition device. Correct for other influencing factors.

[0089] 3. Input the hyperspectral image containing geographic coordinate metadata into the remote sensing image recognition model. The feature maps output by each layer of the downsampling operation retain the spatial location information of the pixels. Through cross-scale feature fusion operations, combined with feature entropy to dynamically allocate spectral and spatial weights, when identifying targets such as alteration halos, each recognition result corresponds to the geographic coordinates of the original image.

[0090] 4. The recognition results generate an ESRIShapefile vector file, where each patch contains: land cover category, confidence level, 10-dimensional spectral feature vector, and geographic coordinate metadata. This embodiment can further overlay satellite imagery on the recognition results to generate a recognized map with geographic coordinates.

[0091] This embodiment achieves higher visualization of remote sensing image recognition through deep coupling of geographic coordinates and remote sensing data, and improves the efficiency of geochemical remote sensing image recognition through automation technology.

[0092] Based on the aforementioned remote sensing image recognition method, this application also provides a geochemical remote sensing image recognition device. The following will be combined with... Figure 7 The device is described in detail.

[0093] Figure 7 A schematic block diagram of a geochemical remote sensing image recognition device according to an embodiment of this application is shown.

[0094] like Figure 7 As shown, the geochemical remote sensing image recognition device 700 of this embodiment includes an image acquisition module 710 and an image recognition module 720.

[0095] The image acquisition module 710 is used to input the hyperspectral image into the image recognition module in response to acquiring the hyperspectral image.

[0096] The image recognition module 720 is used to obtain remote sensing image recognition results based on hyperspectral images and remote sensing image recognition models.

[0097] The remote sensing image recognition model is trained using hyperspectral image samples and labels using the following method: multiple downsampling operations are performed on the hyperspectral image samples to extract image features and reduce the spatial size, thus obtaining image coding features; the downsampling operation includes sequential convolution, activation function operation, and max pooling operation; multiple upsampling operations are performed on the image coding features to restore the spatial resolution and obtain image decoding features; the upsampling operation includes: outputting a downsampled feature map based on the activation function operation, performing a cross-scale feature fusion operation on the downsampled feature map, concatenating it with the feature map of the same level of upsampling operation layer, and performing a convolution operation to fuse them to obtain image decoding features.

[0098] The image decoding features output from the last upsampling operation are subjected to convolution and classification operations to output the image classification result.

[0099] The cross-scale feature fusion operation includes fusing features from different levels of downsampled feature maps based on the feature complexity of the downsampled feature maps.

[0100] According to embodiments of this application, any plurality of modules in the image acquisition module 710 and the image recognition module 720 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least a portion of the functionality of one or more of these modules can be combined with at least a portion of the functionality of other modules and implemented in one module. According to embodiments of this application, at least one of the image acquisition module 710 and the image recognition module 720 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any appropriate combination of any of these three implementation methods. Alternatively, at least one of the image acquisition module 710 and the image recognition module 720 can be at least partially implemented as a computer program module, which, when run, can perform corresponding functions.

[0101] Figure 8 A block diagram schematically illustrates an electronic device suitable for implementing a geochemical remote sensing image recognition method according to an embodiment of this application.

[0102] like Figure 8 As shown, an electronic device 800 according to an embodiment of this application includes a processor 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage portion 808 into a random access memory (RAM) 803. The processor 801 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 801 may also include onboard memory for caching purposes. The processor 801 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0103] RAM 803 stores various programs and data required for the operation of electronic device 800. Processor 801, ROM 802, and RAM 803 are interconnected via bus 804. Processor 801 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 802 and / or RAM 803. It should be noted that the programs may also be stored in one or more memories other than ROM 802 and RAM 803. Processor 801 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0104] According to embodiments of this application, the electronic device 800 may further include an input / output (I / O) interface 805, which is also connected to a bus 804. The electronic device 800 may also include one or more of the following components connected to the input / output (I / O) interface 805: an input section 806 including a keyboard, mouse, etc.; an output section 807 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN card, modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the input / output (I / O) interface 805 as needed. A removable medium 811, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 810 as needed so that computer programs read from it can be installed into the storage section 808 as needed.

[0105] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0106] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 802 and / or RAM 803 and / or one or more memories other than ROM 802 and RAM 803 described above.

[0107] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the geochemical remote sensing image recognition method provided in the embodiments of this application.

[0108] When the computer program is executed by the processor 801, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0109] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 809, and / or installed from a removable medium 811. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0110] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 809, and / or installed from the removable medium 811. When the computer program is executed by the processor 801, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0111] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0112] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0113] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A geochemical remote sensing image recognition method, characterized in that, include: In response to acquiring a hyperspectral image, the hyperspectral image is input into a remote sensing image recognition model to obtain a remote sensing image recognition result; The remote sensing image recognition model was trained using hyperspectral image samples and labels in the following manner: The hyperspectral image samples are downsampled multiple times to extract image features and reduce their spatial size, thereby obtaining image coding features; the downsampling operation includes sequentially performing convolution, activation function, and max pooling operations. The image encoding features are upsampled multiple times to restore spatial resolution and obtain image decoding features. The upsampling operation includes: outputting a downsampled feature map based on the activation function operation, performing a cross-scale feature fusion operation on the downsampled feature map and concatenating it with the feature map of the same level upsampling operation layer, and performing a convolution operation to fuse them to obtain image decoding features. Perform convolution and classification operations on the image decoding features output from the last upsampling operation to output the image classification result; The cross-scale feature fusion operation includes fusing features from different levels of the downsampled feature maps based on the feature complexity of the downsampled feature maps. The step of fusing features of different levels of downsampled feature maps based on feature complexity includes: constructing a multi-scale feature pyramid using the downsampled feature maps; calculating the fusion weights of the multi-scale features based on the spectral and spatial entropy values ​​of the downsampled feature maps; and fusing the multi-scale features based on the fusion weights.

2. The method according to claim 1, characterized in that, The downsampling operation includes: performing convolution, activation function, convolutional block attention module operation and max pooling operation in sequence; The step of performing cross-scale feature fusion on the downsampled feature map and then concatenating it with the feature map of the same level upsampled operation layer includes: performing cross-scale feature fusion on the feature map output by activation function operation and convolutional block attention module operation in the downsampled operation, and then concatenating it with the feature map of the same level upsampled operation layer.

3. The method according to claim 1, characterized in that, Before performing multiple downsampling operations on the hyperspectral image samples to extract image features and reduce spatial size, the method further includes: The hyperspectral image is preprocessed, and the preprocessing operation includes at least one of atmospheric correction, topographic correction, and geometric correction.

4. The method according to claim 1, characterized in that, The step of performing multiple downsampling operations on the hyperspectral image samples to extract image features and reduce spatial size includes: The Otsu algorithm is used to binarize the hyperspectral image and extract the rough outline of the target region. The hyperspectral image samples containing the rough outline are downsampled multiple times to extract image features and reduce spatial size.

5. The method according to claim 1, characterized in that, The step of inputting the hyperspectral image into the remote sensing image recognition model to obtain the remote sensing image recognition result includes: Based on the geographic coordinates and relative attitude parameters at the time of hyperspectral image acquisition, the remote sensing image recognition result with geographic coordinates is output.

6. A geochemical remote sensing image recognition device, characterized in that, The device includes: The image acquisition module is used to input the hyperspectral image into the image recognition module in response to acquiring the hyperspectral image; The image recognition module is used to obtain remote sensing image recognition results based on the hyperspectral image and the remote sensing image recognition model. The remote sensing image recognition model is trained using hyperspectral image samples and labels using the following method: multiple downsampling operations are performed on the hyperspectral image samples to extract image features and reduce their spatial size, resulting in image encoding features; the downsampling operation includes sequentially performing convolution, activation function, and max pooling operations; multiple upsampling operations are performed on the image encoding features to restore spatial resolution and obtain image decoding features; the upsampling operation includes: outputting a downsampled feature map based on the activation function operation, performing cross-scale feature fusion on the downsampled feature map, concatenating it with feature maps from the same level of upsampling operation layer, and then performing convolution to fuse them and obtain image decoding features; Perform convolution and classification operations on the image decoding features output from the last upsampling operation to output the image classification result; The cross-scale feature fusion operation includes fusing features from different levels of the downsampled feature maps based on the feature complexity of the downsampled feature maps. The step of fusing features of different levels of downsampled feature maps based on feature complexity includes: constructing a multi-scale feature pyramid using the downsampled feature maps; calculating the fusion weights of the multi-scale features based on the spectral and spatial entropy values ​​of the downsampled feature maps; and fusing the multi-scale features based on the fusion weights.

7. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 5.

9. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 5.

Images