Forest land disturbance detection method and device, electronic equipment, storage medium and product

Abstract

The application discloses a forest disturbance detection method and device, electronic equipment, storage medium and product, relates to the technical field of land scene detection, and discloses a forest disturbance detection method, which comprises the following steps: inputting an optical image of a target forest area into an optical feature extraction branch of a preset disturbance detection model to obtain optical features, wherein the optical features comprise spatial texture features and spectral features; inputting a radar image of the target forest area into a radar feature extraction branch of the disturbance detection model to obtain radar features, wherein the radar features comprise scattering features and terrain features; performing cross-modal feature fusion on the optical features and the radar features to obtain fusion features; and performing classification and identification based on the fusion features to determine the disturbance state of the target forest area. The application can optimize the forest disturbance change detection effect.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of land scene detection, in particular to a forest land disturbance detection method and device, an electronic device, a storage medium and a product. BACKGROUND

[0002] Under the background of global forest land resources facing continuous reduction and ecological environment worsening, timely and accurate disturbance detection of forest land has become a key link to protect forest land resources and maintain ecological balance.

[0003] However, the current forest land disturbance detection technology mainly relies on a single data source for monitoring and analysis, which is difficult to comprehensively and accurately extract feature information that can fully represent the forest land disturbance situation, resulting in a high misjudgment rate and omission rate of forest land disturbance detection results.

[0004] In summary, how to optimize the forest land disturbance change detection effect has become a technical problem urgently to be solved in the field. SUMMARY

[0005] The main purpose of the present application is to provide a forest land disturbance detection method and device, an electronic device, a storage medium and a product, which aims to optimize the forest land disturbance change detection effect.

[0006] To achieve the above purpose, the present application provides a forest land disturbance detection method, which comprises:

[0007] inputting an optical image of a target forest land area into an optical feature extraction branch of a preset disturbance detection model to obtain optical features, wherein the optical features include spatial texture features and spectral features;

[0008] inputting a radar image of the target forest land area into a radar feature extraction branch of the disturbance detection model to obtain radar features, wherein the radar features include scattering features and terrain features;

[0009] performing cross-modal feature fusion on the optical features and the radar features to obtain fusion features;

[0010] performing classification and recognition based on the fusion features to determine the disturbance state of the target forest land area. In an embodiment, before the step of inputting an optical image of a target forest land area into an optical feature extraction branch of a preset disturbance detection model to obtain optical features, the method further comprises:

[0011] obtaining training sample data containing forest land areas with different disturbance states, wherein the training sample data includes optical image samples and radar image samples, and each sample is labeled with a disturbance state category label;

[0012] The network training of the deep learning network to be trained is performed according to the training sample data to obtain a disturbance detection model, wherein a loss function used in the network training includes a binary cross-entropy loss function and a Dice loss function.

[0013] In an embodiment, before the step of inputting the optical image of the target forest region into the preset optical feature extraction branch of the disturbance detection model to obtain optical features, the method further comprises:

[0014] Collecting an initial optical image and an initial radar image of the target forest region;

[0015] Performing a first preprocessing operation on the initial optical image to obtain an optical image, wherein the first preprocessing operation includes atmospheric correction, radiation correction, geometric correction, processing of missing values and abnormal values, data normalization and standardization processing;

[0016] Performing a second preprocessing operation on the initial radar image to obtain a radar image, wherein the second preprocessing operation includes applying an orbit file, removing boundary noise and thermal noise, radiation correction, terrain correction and fine processing of ground distance image data.

[0017] In an embodiment, the step of performing cross-modal feature fusion on the optical features and the radar features to obtain fused features comprises:

[0018] Through a feature concatenation layer of the disturbance detection model, dynamically adjusting the weight contribution of the optical features and the radar features according to the current weather condition, and performing cross-modal feature fusion on the optical features and the radar features after adjusting the weight contribution to obtain preliminary fused features;

[0019] Using a self-attention mechanism to perform deep cross-modal feature fusion on the preliminary fused features to obtain fused features.

[0020] In an embodiment, the step of performing classification and recognition based on the fused features to determine the disturbance state of the target forest region comprises:

[0021] Performing dimension reduction processing on the fused features through a fully connected layer of the disturbance detection model to obtain a feature vector after dimension reduction;

[0022] Performing classification calculation on the feature vector based on a preset classification threshold through a classifier of the disturbance detection model to output a classification result;

[0023] According to the classification result, combining a preset disturbance state category label to determine the disturbance state of the target forest region, wherein the disturbance state category label includes undisturbed, deforestation disturbance, insect pest disturbance and fire disturbance.

[0024] In an embodiment, the optical feature extraction branch adopts a residual network, including multiple convolution layers, pooling layers and activation function layers, the step of inputting the optical image of the target forest region into the preset disturbance detection model through the optical feature extraction branch of the disturbance detection model, to obtain optical features, includes:

[0025] The optical image of the target forest region is input into the optical feature extraction branch, and forward propagation is performed layer by layer according to the network structure of the optical feature extraction branch, to obtain optical features;

[0026] The radar feature extraction branch adopts a convolution network, including multiple convolution layers, batch normalization layers and activation function layers, the step of inputting the radar image of the target forest region into the radar feature extraction branch of the disturbance detection model, to obtain radar features, includes:

[0027] The radar image of the target forest region is input into the radar feature extraction branch, and forward propagation is performed layer by layer according to the network structure of the radar feature extraction branch, to obtain radar features.

[0028] In addition, to achieve the above-mentioned purpose, the present application also proposes a forest disturbance detection device, which comprises:

[0029] An optical feature extraction module is configured to input an optical image of a target forest region into an optical feature extraction branch of a preset disturbance detection model, to obtain optical features, wherein the optical features include spatial texture features and spectral features;

[0030] A radar feature extraction module is configured to input a radar image of the target forest region into a radar feature extraction branch of the disturbance detection model, to obtain radar features, wherein the radar features include scattering features and terrain features;

[0031] A feature fusion module is configured to perform cross-modal feature fusion on the optical features and the radar features, to obtain fusion features;

[0032] A state recognition module is configured to perform classification recognition based on the fusion features, to determine a disturbance state of the target forest region.

[0033] In addition, to achieve the above-mentioned purpose, the present application also proposes an electronic device, which comprises a memory, a processor and a computer program stored on the memory and executable on the processor, the computer program being configured to implement the steps of the forest disturbance detection method as described above.

[0034] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and the storage medium stores a computer program, the computer program being executed by a processor to implement the steps of the forest disturbance detection method as described above.

[0035] In addition, to achieve the above object, the application further provides a computer program product, which comprises a computer program, and the computer program realizes the steps of the forest disturbance detection method when executed by a processor.

[0036] The application provides a forest disturbance detection method. An optical image of a target forest area is input into an optical feature extraction branch of a preset disturbance detection model to obtain optical features, wherein the optical features include spatial texture features and spectral features. A radar image of the target forest area is input into a radar feature extraction branch of the disturbance detection model to obtain radar features, wherein the radar features include scattering features and terrain features. Cross-modal feature fusion is performed on the optical features and the radar features to obtain fused features. Classification and recognition are performed based on the fused features to determine a disturbance state of the target forest area.

[0037] In summary, in the application, the optical image of the target forest area is input into the optical feature extraction branch of the disturbance detection model to extract the optical features including the spatial texture features and the spectral features. The optical features can reflect the visual information of the vegetation distribution and growth trend of the target forest area. Meanwhile, the radar image is input into the radar feature extraction branch to extract the radar features including the scattering features and the terrain features. The radar features can reflect the physical structure changes and terrain fluctuation information of the target forest area. On this basis, the cross-modal feature fusion is performed on the optical features and the radar features to break the limitation of a single data source, which can more comprehensively and accurately represent the disturbance of the target forest area. The classification and recognition based on the fused features can improve the reliability of the judgment of the forest disturbance state, thereby optimizing the forest disturbance detection result. BRIEF DESCRIPTION OF DRAWINGS

[0038] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and serve to explain the principles of the application together with the specification.

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, for those skilled in the art, the other drawings can also be obtained without creative labor based on these drawings.

[0040] Figure 1 A flowchart is provided for the forest disturbance detection method embodiment one of the application;

[0041] Figure 2 A disturbance detection model structure diagram is provided for the forest disturbance detection method embodiment one of the application;

[0042] Figure 3A forest disturbance detection flowchart provided for the second embodiment of the forest disturbance detection method of the present application is shown in FIG. 1.

[0043] Figure 4 A structural diagram of the forest disturbance detection device of the embodiment of the present application is shown in FIG. 2.

[0044] Figure 5 A device structure diagram of the hardware running environment involved in the forest disturbance detection method in the embodiment of the present application is shown in FIG. 3.

[0045] The object implementation, functional features and advantages of the present application will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0046] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and do not limit the present application.

[0047] In order to better understand the technical solutions of the present application, the specific embodiments will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0048] Under the background of continuous reduction of global forest resources and continuous deterioration of ecological environment, timely and accurate disturbance detection of forest land has become a key link for protecting forest resources and maintaining ecological balance.

[0049] However, the current forest disturbance detection technology mainly relies on a single data source for monitoring and analysis, which is difficult to fully and accurately extract feature information that can fully represent the forest disturbance situation, resulting in a high misjudgment rate and omission rate of forest disturbance detection results.

[0050] In summary, how to optimize the forest disturbance change detection effect has become a technical problem urgently to be solved in the field.

[0051] The main solution of the embodiment of the present application is: inputting the optical image of the target forest area into the optical feature extraction branch of the preset disturbance detection model to obtain optical features, wherein the optical features include spatial texture features and spectral features; inputting the radar image of the target forest area into the radar feature extraction branch of the disturbance detection model to obtain radar features, wherein the radar features include scattering features and terrain features; performing cross-modal feature fusion on the optical features and the radar features to obtain fusion features; and performing classification and recognition based on the fusion features to determine the disturbance state of the target forest area.

[0052] Thus, in the present embodiment, the optical image of the target forest region is input to the optical feature extraction branch of the disturbance detection model, and the optical features containing spatial texture features and spectral features are extracted, which can reflect the visual level information of the vegetation distribution and growth trend of the target forest region; at the same time, the radar image is input to the radar feature extraction branch, and the radar features including scattering features and terrain features are extracted, which can reflect the physical structure changes and terrain undulation information of the target forest region; on this basis, the optical features and the radar features are fused, breaking the limitation of a single data source, which can more comprehensively and accurately represent the disturbance situation of the target forest region, and based on the fused features, the classification and recognition can be performed, which can improve the reliability of the judgment of the forest disturbance state, thereby optimizing the forest disturbance detection result.

[0053] It should be noted that the execution subject of the present embodiment can be a computer service device with data processing, network communication and program running functions, such as a server, an upper computer, a personal computer, etc., or an electronic device capable of realizing the above functions, etc.

[0054] Based on this, the present embodiment provides a forest disturbance detection method, which is described with reference to Figure 1 , Figure 1 The flowchart of the first embodiment of the forest disturbance detection method of the present application is shown in FIG. 1.

[0055] In the present embodiment, the forest disturbance detection method comprises steps S10-S40:

[0056] In step S10, the optical image of the target forest region is input to the optical feature extraction branch of the preset disturbance detection model, and the optical features are obtained, wherein the optical features include spatial texture features and spectral features.

[0057] In the process of detecting the disturbance of the target forest region, the optical image of the target forest region is periodically acquired, and then the optical image is input to the optical feature extraction branch in the preset disturbance detection model, and the optical features are extracted.

[0058] It should be noted that the optical feature extraction branch can extract the optical features from the optical image by analyzing the pixel distribution, color difference, spatial structure and other elements in the optical image, which can comprehensively represent the optical level characteristics of the target forest region. The optical features include but are not limited to spatial texture features and spectral features, wherein the spatial texture features reflect the roughness of the surface of the target forest region, the arrangement of patterns and other spatial information, which is helpful to identify different types of vegetation cover and landform; the spectral features reflect the reflection characteristics of different ground objects to different wavelengths of light, which can distinguish different vegetation species, soil types, etc.

[0059] Step S20, inputting the radar image of the target forest region to the radar feature extraction branch of the disturbance detection model to obtain radar features, wherein the radar features include scattering features and terrain features;

[0060] In the process of disturbance detection for the target forest region, the radar image of the target forest region is periodically acquired, and then the radar image is input to the radar feature extraction branch in the disturbance detection model to extract radar features.

[0061] It should be noted that the radar feature extraction branch extracts radar features by analyzing the interaction between the radar signal and the target forest region. The radar features include but are not limited to scattering features and terrain features. The scattering features reflect the scattering of different ground objects in the target forest region. Different ground objects have different scattering characteristics due to differences in shape, material and structure. By analyzing the scattering features, information such as changes in forest structure and vegetation density can be identified. The terrain features reflect the topography and slope of the target forest region.

[0062] It is worth mentioning that the optical image of the target forest region can be obtained from Planet optical remote sensing data (a source of optical remote sensing image data) or other high-resolution data sources. The data source of the optical image is not limited in this embodiment. However, the acquisition of optical images is severely affected by weather conditions such as clouds, fog and light changes. In severe weather conditions, the quality of the optical image will decrease significantly, and even effective image data cannot be obtained, which will affect the normal development of the detection work and further affect the accurate judgment of the disturbance state of the forest. The radar image can be obtained from Sentinel-1 radar remote sensing data (a source of radar remote sensing image data). However, the radar image does not reflect the spectral information of the forest in a direct and rich manner, making it difficult to accurately distinguish some subtle forest disturbance types, especially in forests with different vegetation types and growth stages. It is difficult to achieve high-precision disturbance detection relying solely on radar images. Due to the limitations of optical images and radar images, using only one data source for forest disturbance detection cannot fully and accurately extract feature information that fully represents the forest disturbance situation, resulting in a high misjudgment rate and a high omission rate of the detection results.

[0063] Step S30, cross-modal feature fusion of the optical features and the radar features to obtain fusion features;

[0064] After the optical features and the radar features are extracted respectively, cross-modal feature fusion operations are performed on the features of the two different modalities to integrate the information carried by the optical features and the radar features, that is, to deeply fuse the spatial texture features and the spectral features in the optical features with the scattering features and the terrain features in the radar features, so that the fused features can more comprehensively and accurately reflect the actual situation of the target forest region.

[0065] In a feasible embodiment, step S30 can include steps S301-S302:

[0066] In step S301, the feature concatenation layer of the disturbance detection model is disturbed to dynamically adjust the weight contribution degrees of the optical features and the radar features according to the current meteorological condition, and cross-modal feature fusion is performed on the optical features and the radar features after the weight contribution degrees are adjusted to obtain preliminary fused features.

[0067] In actual application scenarios, the meteorological condition of the current target forest region can be obtained based on sensor detection or a third-party interface. The meteorological condition is a meteorological and weather factor that affects the imaging quality of the optical image, such as cloud cover, fog, haze, rain, snow, light intensity, visibility, etc. The disturbance detection model adjusts the weight contribution degrees of the optical features and the radar features according to the obtained meteorological condition. For example, in sunny conditions, the optical image quality is high, and the optical features contain rich effective information. At this time, the weight contribution degree of the optical features is appropriately increased. In rainy or cloudy weather, the optical image is greatly disturbed, and the advantages of the radar image are more prominent, so the weight contribution degree of the radar features is correspondingly increased.

[0068] In a feasible implementation scenario, the step of dynamically adjusting the weight contribution degrees of the optical features and the radar features according to the current meteorological condition includes: obtaining the current meteorological condition, and generating an imaging quality score according to the current meteorological condition. The imaging quality score is a score reflecting the imaging quality condition of the current optical image. The weight contribution degree of the optical features is proportional to the imaging quality score, and the weight contribution degree of the radar features is inversely proportional to the imaging quality score. Then, after the weight contribution degrees of the optical features and the radar features are dynamically adjusted, the two adjusted features are cross-modal feature fused to obtain preliminary fused features.

[0069] In step S302, a self-attention mechanism is used to perform deep cross-modal feature fusion on the preliminary fused features to obtain fused features.

[0070] To further tap the deep correlation between different modal features in the preliminary fusion features, a self-attention mechanism is used for deep cross-modal feature fusion of the preliminary fusion features. Specifically, the self-attention mechanism can automatically learn the mutual relationship between the features, highlight the feature part more important to the classification and recognition task, and suppress the unimportant features. When processing the preliminary fusion features, the self-attention mechanism will focus on the internal relationship between the features in different positions and different modalities, reweight and combine the features, so as to obtain the fusion features that can more comprehensively and accurately reflect the feature information of the target forest region.

[0071] Thus, in the present embodiment, the weight contribution of the optical features and the radar features can be dynamically adjusted according to the meteorological conditions of the target forest region, making the fusion process more flexible and adaptive. Under different meteorological and weather conditions, effective data can be fully utilized, and the self-attention mechanism is further used for deep cross-modal feature fusion of the preliminary fusion features, tapping the deep correlation between different modal features, so as to obtain more discriminative fusion features, making the classification and recognition based on the fusion features more accurately determine the disturbance state of the target forest region, and improving the accuracy and robustness of the disturbance detection.

[0072] Step S40, classification and recognition based on the fusion features to determine the disturbance state of the target forest region.

[0073] The classification and recognition based on the fusion features by the disturbance detection model determines the disturbance state of the target forest region, wherein the disturbance state can include different levels such as no disturbance, light disturbance, moderate disturbance and severe disturbance, and can be further subdivided into specific disturbance types such as deforestation, fire, pest and disease, etc. Through classification and recognition, the disturbance situation of the target forest region can be accurately and timely grasped.

[0074] In a feasible embodiment, step S40 can include steps S401-S403:

[0075] Step S401, dimensionality reduction processing of the fusion features by the fully connected layer of the disturbance detection model to obtain a feature vector after dimensionality reduction;

[0076] Since the fusion features usually contain rich information but have high dimensions, directly using them for classification calculation may cause problems such as high computational complexity and overfitting. In this embodiment, the obtained fusion features are input into the full connection layer of the disturbance detection model, the full connection layer integrates and compresses the fusion features through a series of linear transformations and nonlinear activation functions, removes redundant information, extracts the most representative features, and thus obtains the dimension-reduced feature vectors. The dimension-reduced feature vectors not only retain the key information in the fusion features but also reduce the dimensions, providing more efficient and stable feature representations for subsequent classification calculation.

[0077] In step S402, the feature vectors are classified and calculated by the classifier of the disturbance detection model based on a preset classification threshold, and a classification result is output.

[0078] The obtained dimension-reduced feature vectors are input into the classifier of the disturbance detection model, and the classifier classifies and calculates the input feature vectors based on a preset classification threshold.

[0079] It should be noted that in the training phase of the disturbance detection model, the classifier learns the mapping relationship between different disturbance state categories and feature vectors through a large amount of labeled data, and determines a suitable classification threshold. In the classification calculation process, the classifier will determine the category to which the feature vector belongs according to the similarity between the dimension-reduced feature vector and the feature distribution of each category, in combination with the preset classification threshold, and output the classification result.

[0080] In step S403, the disturbance state of the target forest region is determined according to the classification result and in combination with preset disturbance state category labels. The disturbance state category labels include undisturbed, deforestation disturbance, insect pest disturbance, and fire disturbance.

[0081] The preset disturbance state category labels are specific classification definitions of the forest disturbance state, and can specifically include undisturbed, deforestation disturbance, insect pest disturbance, and fire disturbance categories. According to the determined classification result, in combination with the preset disturbance state category labels, the disturbance state of the target forest region is finally determined.

[0082] For example, if the classification result shows that in terms of optical features, most of the vegetation in the target forest region grows normally, and there is no obvious spectral abnormal change caused by deforestation, insect pest, or fire; at the same time, the radar features show that the vegetation structure of the target forest region is complete, and the backscattering coefficient is stable, which is consistent with the features of the undisturbed healthy forest region, it is determined that the region is in an undisturbed state.

[0083] If the classification result shows that, in terms of optical features, there is a large area of tree felling in the target forest area, the vegetation coverage is significantly reduced, and obvious empty land or forest road is formed; at the same time, the radar features show that the vegetation height is greatly reduced, and the backscattering coefficient changes obviously, which presents the radar signal features consistent with felling disturbance, it is determined that the target forest area is in felling disturbance state.

[0084] If the classification result shows that, in terms of optical features, part of the trees in the target forest area appear yellowing leaves and damaged branches, forming a pest patch, and the spectral features are consistent with the vegetation invaded by pests; at the same time, the radar features show that the tree canopy structure changes, resulting in abnormal fluctuations of the backscattering coefficient in the corresponding area, which is consistent with the radar features of pest disturbance, it is determined that the target forest area is in pest disturbance state.

[0085] If the classification result shows that, in terms of optical features, the vegetation in the target forest area is burned, the ground is black, and there are branches and leaves; at the same time, in terms of radar features, the vegetation structure is completely destroyed, the ground roughness changes greatly, and the backscattering coefficient presents a radar signal mode highly consistent with fire disturbance, it is determined that the target forest area is in fire disturbance state.

[0086] Exemplarily, in a feasible implementation scenario, the trained disturbance detection model structure is as shown in Figure 2 The optical image is input to the optical feature extraction branch of the disturbance detection model for optical feature extraction, the radar image is input to the radar feature extraction branch for radar feature extraction, then the extracted features are input to the feature splicing layer for feature fusion, the obtained fusion features are input to the full connection layer for dimension reduction processing, and then the obtained dimension-reduced feature vector is input to the classifier for classification calculation to obtain the classification result, so as to fully utilize the feature representation of the two data sources of the optical image and the radar image, and improve the accuracy of the forest disturbance detection result.

[0087] In this way, in the embodiment of the present application, by inputting the optical image of the target forest area to the optical feature extraction branch of the disturbance detection model, the optical features containing spatial texture features and spectral features are extracted, which can reflect the visual information such as vegetation distribution and growth trend of the target forest area; at the same time, by inputting the radar image to the radar feature extraction branch, the radar features including scattering features and terrain features are extracted, which can reflect the physical structure change and terrain information of the target forest area; on this basis, the optical features and the radar features are cross-modal feature fused, breaking the limitation of single data source, which can more comprehensively and accurately represent the disturbance situation of the target forest area, and based on the fusion features, the classification recognition is carried out, which can improve the judgment reliability of the forest disturbance state, thereby optimizing the forest disturbance detection result.

[0088] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as the above embodiment one can refer to the above introduction, and the subsequent will not be described. On this basis, step S10 can also include steps A10-A20 before it:

[0089] Step A10, obtaining training sample data containing forest regions in different disturbance states, wherein the training sample data includes optical image samples and radar image samples, and each sample is labeled with a disturbance state category label;

[0090] In the training phase of the forest disturbance model, first, training sample data containing forest regions in different disturbance states is obtained, which contains optical image samples and radar image samples. The optical image can provide visible light information of the forest region in different disturbance states, reflecting the color, texture and other characteristics of the vegetation; the radar image can penetrate through bad weather conditions such as clouds to obtain the structure and height information of the forest. In order to enable the model to be trained to accurately learn the characteristics of different disturbance states, each sample needs to be labeled with a clear disturbance state category label, which is a classification definition of the disturbance of the forest region.

[0091] Step A20, training the deep learning network to be trained according to the training sample data to obtain a disturbance detection model, wherein the loss function used in the network training includes a binary cross entropy loss function and a Dice loss function.

[0092] Based on the obtained training sample data, the deep learning network to be trained is trained, and the loss function used in the network training includes a binary cross entropy loss function and a Dice loss function, and then a disturbance detection model is obtained.

[0093] It is worth mentioning that the binary cross entropy loss function can measure the difference between the probability distribution of the model prediction and the true label, guide the model to optimize the parameters, and make the prediction result closer to the true situation; the Dice loss function (i.e. the Dice loss function) focuses on measuring the overlap between the model prediction result and the true label, and has good effect on dealing with the class imbalance problem (such as fewer disturbance state samples in forest disturbance). By combining the two loss functions, the binary cross entropy loss function reduces false positives, such as misjudging shadows as disturbances, and the Dice loss function reduces missed detection, such as small broken disturbance areas, so as to enhance the region integrity of the Dice loss in the small target disturbance (such as local illegal logging) scene, and avoid missing scattered pixels; in the weak signal disturbance (such as early fire) scene, the binary cross entropy loss function enhances the classification confidence through pixel-level supervision, so as to more comprehensively guide the model training, improve the generalization ability and classification accuracy of the model.

[0094] In addition, in one feasible implementation, the model training process also employs early stopping and L2 regularization. Early stopping is applied during training by monitoring the performance of the validation set loss under a set "patience value". When the validation loss no longer improves significantly, training is automatically stopped to prevent overfitting. L2 regularization further reduces the risk of overfitting by limiting the weight size.

[0095] In one feasible embodiment, steps B10 to B30 may be included before step S10:

[0096] Step B10: Acquire initial optical and radar images of the target forest area;

[0097] Initial optical images of the target forest area can be obtained from Planet optical remote sensing data, while initial radar images can be obtained from Sentinel-1 radar remote sensing data. Alternatively, specialized remote sensing data acquisition equipment or related technologies can be used to acquire image data of the target forest area, obtaining initial optical and radar images. Initial optical images reflect the spectral information of the forest area in the visible and near-infrared bands, which are closely related to the growth status and coverage of vegetation. Initial radar images are not limited by lighting and weather conditions and can provide information on the structure, height, and scattering characteristics of the forest. These two types of image data can describe the characteristics of the target forest area from different perspectives.

[0098] Step B20: Perform a first preprocessing operation on the initial optical image to obtain an optical image. The first preprocessing operation includes atmospheric correction, radiometric correction, geometric correction, handling of missing and outlier values, data normalization and standardization.

[0099] A first preprocessing operation is performed on the acquired initial optical image to obtain a high-quality optical image. This first preprocessing operation includes several key steps:

[0100] Atmospheric correction: Gas molecules and aerosols in the atmosphere can scatter and absorb light from optical images, causing distortion in image radiance values. Atmospheric correction aims to eliminate the influence of the atmosphere on images, restore the true reflectance of ground features, and thus improve image quality and classification accuracy.

[0101] Radiometric correction: Due to the inherent characteristics of the sensor and various factors during the imaging process, the radiometric values ​​of optical images may contain errors. Radiometric correction is used to correct these errors, ensuring that the radiometric values ​​of the image accurately reflect the radiometric characteristics of the ground object.

[0102] Geometric Correction: During image acquisition, due to factors such as sensor attitude, Earth curvature, and terrain undulations, the positions of ground objects in the image may be geometrically distorted. Geometric correction establishes a mapping relationship between image coordinates and geographic coordinates, and corrects the geometric distortion of the image, so that the positions of ground objects in the image are consistent with the actual geographic positions.

[0103] Handling Missing Values and Outliers: During image acquisition and transmission, missing values or outliers may occur in the image due to sensor failure, signal interference, etc. These abnormal data will affect the accuracy of subsequent image analysis and processing, so they need to be identified and processed, such as using interpolation method to fill in missing values, and smoothing processing for outliers, etc.

[0104] Data Normalization and Standardization: Different bands of optical image data may have different dimensions and value ranges, which will affect the stability and convergence speed of subsequent model training. Data normalization and standardization scale the image data to a specific range or make it have a specific statistical distribution, such as normalizing the data to the [0, 1] interval, or Z-score standardization to make the data have zero mean and unit variance, thereby improving the training effect of the model.

[0105] It is worth mentioning that before the initial optical image and the initial radar image are preprocessed, the initial optical image and the initial radar image are first spatio-temporally configured, specifically, in terms of spatial alignment, first, obvious feature points are extracted on the initial optical image by means of a corner point detection algorithm, such as a Harris corner point detection algorithm (an algorithm for finding corner points in an image), and corresponding feature points are extracted according to the gray level variation and texture features of the initial radar image, then a feature matching algorithm, such as SIFT (Scale-Invariant Feature Transform), SURF (Speeded Up Robust Features), etc., is used to match the feature points of the two images to find corresponding feature point pairs; then, according to the matched feature point pairs, a spatial transformation model of the optical image to the radar image is estimated, such as an affine transformation model or a polynomial transformation model, the former can handle linear transformations such as translation, rotation, scaling, etc., and the latter is suitable for complex nonlinear transformations; finally, the initial optical image is resampled using the estimated model, and methods such as bilinear interpolation and cubic convolution interpolation are used to improve the quality of the resampled image, so that it is spatially aligned with the initial radar image. In terms of time registration, the imaging time information is obtained from the metadata of the two images, and the time is uniformly converted to Coordinated Universal Time (UTC) as the reference. If the imaging time interval of the two images is large and the disturbance state has continuous changes, linear interpolation or spline interpolation can also be used for time interpolation method, according to the image information of adjacent time points to estimate the image data close to the imaging time of the other image, so as to complete the spatio-temporal registration of the initial optical image and the initial radar image.

[0106] Step B30, performing a second preprocessing operation on the initial radar image to obtain a radar image, wherein the second preprocessing operation includes applying an orbit file, removing boundary noise and thermal noise, radiation correction, terrain correction, and fine processing of ground distance image data.

[0107] The collected initial radar image is subjected to a second preprocessing operation to obtain a high-quality radar image, and the second preprocessing operation includes multiple important steps:

[0108] Apply the orbit file: The orbit file records the orbit parameters of the radar satellite during imaging. Applying the orbit file can accurately geolocate the radar image, ensuring that each pixel in the image can accurately correspond to the actual position on the Earth's surface.

[0109] Remove boundary noise and thermal noise: During the radar image acquisition process, some noise signals unrelated to actual ground objects may be generated, such as boundary noise and thermal noise. These noises will interfere with the extraction and analysis of ground object information, so appropriate methods need to be used to remove them to improve the signal-to-noise ratio of the image.

[0110] Radiometric correction: Similar to optical images, the radiometric values of radar images may also be affected by factors such as sensor characteristics, imaging conditions, etc. and have errors. Radiometric correction is used to correct these errors so that the radiometric values of the radar image can accurately reflect the scattering characteristics of the ground object.

[0111] Terrain correction: Due to the influence of terrain undulations, the ground object scattering echoes in the radar image may be distorted. Terrain correction corrects the radar image by considering terrain factors to eliminate the scattering echo distortion caused by terrain, so that the image can more accurately reflect the true scattering characteristics of the ground object.

[0112] Fine processing of ground range image data: GRD (Ground Range Detected) data is a common format of radar images, and fine processing of it can further improve the quality and usability of the image. For example, filtering the image to suppress noise, multi-view processing to improve the signal-to-noise ratio and resolution of the image, etc.

[0113] Therefore, by performing the first preprocessing operation on the initial optical image, the influence of factors such as atmosphere and sensor on the image can be effectively eliminated, the quality and accuracy of the image can be improved, and the image can more truly reflect the spectral characteristics of the target forest region; by performing the second preprocessing operation on the initial radar image, interference factors in the radar image can be removed, geometric and radiometric distortion of the image can be corrected, and the signal-to-noise ratio and resolution of the image can be improved, so that the radar image can more accurately reflect the structure and scattering characteristics of the target forest region, thereby helping to improve the accuracy and reliability of forest disturbance detection.

[0114] In a feasible embodiment, the optical feature extraction branch adopts a residual network, including multiple convolution layers, pooling layers and activation function layers, and step S10 can include step S101:

[0115] Step S101, inputting the optical image of the target forest region into the optical feature extraction branch, performing forward propagation layer by layer according to the network structure of the optical feature extraction branch, and obtaining the optical feature.

[0116] For feature extraction of optical images, an optical feature extraction branch based on a residual network (ResNet) is adopted, which includes multiple convolution layers, pooling layers and activation function layers. Specifically, the convolution layer extracts local features of the input optical image through a series of learnable convolution kernels, capturing edge, texture and other detailed information in the image; the pooling layer is used to reduce the dimension of the feature map, reduce the amount of calculation, and enhance the translation invariance of the feature; the activation function layer introduces nonlinearity, enabling the network to learn more complex feature representations. Finally, the optical features that can fully represent the optical characteristics of the target forest area are obtained from the output layer of the optical feature extraction branch.

[0117] It is worth mentioning that, in this embodiment, since the optical image usually contains multi-band spectral information, a deep network is needed to extract complex features. The residual skip connection of the residual network can alleviate the gradient vanishing, support the construction of a deeper network, and thus learn more discriminative optical features.

[0118] In a feasible embodiment, the radar feature extraction branch adopts a convolutional network, including multiple convolution layers, batch normalization layers and activation function layers. Step S20 can include step S201:

[0119] In step S201, the radar image of the target forest area is input into the radar feature extraction branch, and is forward propagated layer by layer according to the network structure of the radar feature extraction branch to obtain radar features.

[0120] For feature extraction of radar images, a radar feature extraction branch based on a convolutional neural network (CNN) is adopted, which includes multiple convolution layers, batch normalization layers and activation function layers. Specifically, the convolution layer is used to extract information related to ground object scattering characteristics and structural features from the radar image; the batch normalization layer normalizes each batch of data to accelerate the convergence speed of the network and improve the stability of the training; the activation function layer gives the network non-linear fitting ability, enabling it to better learn complex features in the radar image. Finally, the radar features that can reflect the radar characteristics of the target forest area are obtained from the output layer of the radar feature extraction branch.

[0121] It is worth mentioning that, in this embodiment, the radar image is usually single-channel or low-channel, and the feature complexity is lower than that of the optical image. The convolutional network is sufficient to extract the local texture and structural features of the radar, which can reduce the computational overhead, accelerate the network convergence, and improve the generalization ability of the network.

[0122] In addition, in a feasible implementation, a moving time window algorithm is used to continuously monitor the time series data to identify the monthly newly added forest land disturbance area. Specifically, the size of the moving time window can be determined according to the actual monitoring requirements and the time resolution of the data, for example, the window size can be set to one month, that is, the forest land data of two adjacent months are compared each time, and the forest land data of the two periods (for example, the nth month and the n+1 month) in the window are compared using image difference method, change vector analysis method and the like, and the area appearing in the latter month but not appearing in the former month is selected as the newly added forest land disturbance area of the window (i.e., the latter month), so as to further optimize the detection result and reduce false detection and missed detection.

[0123] By way of example, in order to facilitate understanding of the implementation process of the forest land disturbance detection method obtained by combining the above-mentioned embodiments, please refer to Figure 3 , a schematic diagram of a brief process of a forest land disturbance detection method is provided, specifically:

[0124] When the target forest land area is detected for disturbance, first, the initial optical image and the initial radar image of the target forest land area are obtained, and then the initial optical image and the initial radar image are preprocessed to obtain the optical image and the radar image. Then, the optical image is input into the optical feature extraction branch of the disturbance detection model to obtain the optical image, the radar image is input into the radar feature extraction branch of the disturbance detection model to obtain the radar feature, and then the cross-modal feature fusion is performed through the feature splicing layer and the self-attention mechanism to obtain the fusion feature. After dimension reduction through the full connection layer, the classification result is output by the model classifier to determine the disturbance state of the target forest land area.

[0125] It should be noted that the above examples are only used for understanding the present application and do not constitute a limitation on the forest land disturbance detection method of the present application. More forms of simple transformation based on this technical concept are within the protection scope of the present application.

[0126] The present application also provides a forest land disturbance detection device, please refer to Figure 4 , the forest land disturbance detection device comprises:

[0127] The optical feature extraction module 10 is configured to input the optical image of the target forest land area into the optical feature extraction branch of the preset disturbance detection model to obtain the optical feature, wherein the optical feature includes spatial texture feature and spectral feature.

[0128] The radar feature extraction module 20 is configured to input the radar image of the target forest land area into the radar feature extraction branch of the disturbance detection model to obtain the radar feature, wherein the radar feature includes scattering feature and terrain feature.

[0129] The feature fusion module 30 is configured to perform cross-modal feature fusion on the optical features and the radar features to obtain fused features.

[0130] The state recognition module 40 is configured to perform classification recognition based on the fused features to determine the disturbance state of the target forest region.

[0131] Optionally, the forest disturbance detection device further comprises a model training module (not shown in the figure), which is configured to:

[0132] acquire training sample data containing forest regions in different disturbance states, wherein the training sample data comprises optical image samples and radar image samples, and each sample is labeled with a disturbance state category label;

[0133] perform network training on the deep learning network to be trained based on the training sample data to obtain a disturbance detection model, wherein the loss function used in the network training comprises a binary cross-entropy loss function and a Dice loss function.

[0134] Optionally, the forest disturbance detection device further comprises a preprocessing module (not shown in the figure), which is configured to:

[0135] acquire initial optical images and initial radar images of the target forest region;

[0136] perform a first preprocessing operation on the initial optical images to obtain optical images, wherein the first preprocessing operation comprises atmospheric correction, radiation correction, geometric correction, processing of missing values and abnormal values, data normalization and standardization processing;

[0137] perform a second preprocessing operation on the initial radar images to obtain radar images, wherein the second preprocessing operation comprises applying an orbit file, removing boundary noise and thermal noise, radiation correction, terrain correction, and fine processing of ground distance image data.

[0138] Optionally, the feature fusion module 30 is further configured to:

[0139] adjust the weight contribution of the optical features and the radar features according to the current weather condition through a feature concatenation layer of the disturbance detection model, perform cross-modal feature fusion on the optical features and the radar features after adjusting the weight contribution to obtain preliminary fused features;

[0140] perform deep cross-modal feature fusion on the preliminary fused features using a self-attention mechanism to obtain fused features.

[0141] Optionally, the state recognition module 40 is further configured to:

[0142] perform dimension reduction processing on the fused features through a fully connected layer of the disturbance detection model to obtain a feature vector after dimension reduction;

[0143] The classifier of the disturbance detection model classifies the feature vector based on a preset classification threshold, and outputs a classification result;

[0144] According to the classification result, a disturbance state of the target forest region is determined in combination with preset disturbance state category labels, where the disturbance state category labels include undisturbed, felling disturbance, pest disturbance, and fire disturbance.

[0145] Optionally, the optical feature extraction branch adopts a residual network, including multiple convolution layers, pooling layers, and activation function layers, and the optical feature extraction module 10 is further configured to:

[0146] The optical feature extraction module 10 is further configured to: input the optical image of the target forest region into the optical feature extraction branch, and perform forward propagation layer by layer according to the network structure of the optical feature extraction branch, to obtain the optical feature.

[0147] Optionally, the radar feature extraction branch adopts a convolution network, including multiple convolution layers, batch normalization layers, and activation function layers, and the radar feature extraction module 20 is further configured to:

[0148] The radar feature extraction module 20 is further configured to: input the radar image of the target forest region into the radar feature extraction branch, and perform forward propagation layer by layer according to the network structure of the radar feature extraction branch, to obtain the radar feature.

[0149] The forest disturbance detection device provided in the embodiments of the present application adopts the forest disturbance detection method in the above embodiments, and can optimize the forest disturbance change detection effect. Compared with the prior art, the beneficial effects of the forest disturbance detection device provided in the embodiments of the present application are the same as those of the forest disturbance detection method provided in the above embodiments, and other technical features of the forest disturbance detection device are the same as those disclosed in the forest disturbance detection method in the above embodiments, which will not be repeated here.

[0150] The embodiments of the present application provide an electronic device, which includes at least one processor, and a memory in communication connection with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the forest disturbance detection method in the above embodiment one.

[0151] Reference will be made to the following Figure 5 which shows a structural schematic diagram of an electronic device suitable for implementing the embodiments of the present application. The electronic device in the embodiments of the present application can include but is not limited to a whole machine device such as a multimedia interactive all-in-one machine, a touch all-in-one machine device, etc. Figure 5 The electronic device shown is only an example, and should not impose any limitation on the functions and use range of the embodiments of the present application.

[0152] As Figure 5As shown, the electronic device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the electronic device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. The communication device 1009 allows the electronic device to communicate wirelessly or wiredly with other devices to exchange data. Although the diagrams show electronic devices with various systems, it should be understood that it is not required to implement or have all of the systems shown. More or fewer systems may be implemented alternatively.

[0153] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0154] The electronic device provided in this application, employing the forest disturbance detection method described in the above embodiments, can optimize the detection effect of forest disturbance changes. Compared with the prior art, the beneficial effects of the electronic device provided in this application are the same as those of the forest disturbance detection method provided in the above embodiments, and other technical features of the electronic device are the same as those disclosed in the forest disturbance detection method of the previous embodiment, and will not be repeated here.

[0155] It should be understood that the various parts disclosed in the embodiments of this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0156] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0157] The embodiment of the present application provides a computer readable storage medium having computer readable program instructions (i.e. computer programs) stored thereon, the computer readable program instructions being used for executing the forest disturbance detection method in the above embodiment.

[0158] The computer readable storage medium provided by the embodiment of the present application may, for example, be a U disk, but is not limited to an electric, magnetic, optical, electromagnetic, infrared, or semiconductor system or device, or any combination of the above. More specific examples of the computer readable storage medium can include, but are not limited to, an electric connection having one or more conductive lines, a portable computer disk, a hard disk, a random access memory (RAM: Random Access Memory), a read only memory (ROM: Read Only Memory), an erasable programmable read only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), an optical fiber, a portable compact disk read only memory (CD-ROM: CD-Read Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the embodiment, the computer readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system or device. The program code contained on the computer readable storage medium can be transmitted by any suitable medium, including but not limited to an electric wire, an optical cable, an RF (Radio Frequency: radio frequency), etc., or any suitable combination of the above.

[0159] The above computer readable storage medium can be contained in an electronic device; or can exist separately without being assembled into an electronic device.

[0160] The computer readable storage medium described above carries one or more programs, when the one or more programs are executed by the electronic device, cause the electronic device to: input an optical image of a target forest area into an optical feature extraction branch of a preset disturbance detection model to obtain optical features, wherein the optical features include spatial texture features and spectral features; input a radar image of the target forest area into a radar feature extraction branch of the disturbance detection model to obtain radar features, wherein the radar features include scattering features and terrain features; perform cross-modal feature fusion on the optical features and the radar features to obtain fused features; and perform classification recognition based on the fused features to determine a disturbance state of the target forest area.

[0161] Computer program code for carrying out operations of embodiments of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0162] The flow diagrams and the block diagrams in the drawings are illustrations of architectures, functionalities, and operations of possible implementations of systems, methods, and computer program products according to various embodiments of present application. In this regard, each block in the flow diagrams or block diagrams can represent a module, a procedure, or a part of code, which comprises one or more executable instructions for implementing the specified functions. It should also be noted that, in some alternative implementations, the functions noted in the blocks can occur in a different order than that noted in the figures. For example, two blocks noted in succession can in fact be executed substantially concurrently or in the opposite order, depending on the functionality involved. It should also be noted that each block in the block diagrams and / or flow diagrams, and combinations of blocks in the block diagrams and / or flow diagrams, can be implemented by dedicated hardware-based systems that perform the specified functions or operations, or can be implemented by a combination of dedicated hardware-based systems and computer instructions.

[0163] The modules described in the embodiments of the present application can be implemented in the form of software or in the form of hardware. In some cases, the names of the modules do not constitute a limitation on the modules themselves.

[0164] The readable storage medium provided by the embodiments of the present application is a computer readable storage medium, and the computer readable storage medium stores computer readable program instructions (i.e., a computer program) for executing the forest disturbance detection method. The computer readable program instructions can optimize the forest disturbance change detection effect. Compared with the prior art, the computer readable storage medium provided by the embodiments of the present application has the same beneficial effects as the forest disturbance detection method provided by the above embodiments, and will not be described here.

[0165] The embodiments of the present application also provide a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the forest disturbance detection method are implemented.

[0166] The computer program product provided by the embodiments of the present application can mine effective information from the data generated by the information technology system. Compared with the prior art, the computer program product provided by the embodiments of the present application has the same beneficial effects as the forest disturbance detection method provided by the above embodiments, and will not be described here.

[0167] The above is only some embodiments of the present application, and does not limit the patent scope of the present application. Any equivalent structural transformation made by using the content of the present application specification and drawings, or direct / indirect application in other related technical fields is included in the patent protection scope of the present application.

Claims

1. A method for detecting forest disturbance, characterized in that, The forest disturbance detection method includes: The optical image of the target forest area is input into the optical feature extraction branch of the preset disturbance detection model to obtain optical features, wherein the optical features include spatial texture features and spectral features; The radar image of the target forest area is input into the radar feature extraction branch of the disturbance detection model to obtain radar features, wherein the radar features include scattering features and terrain features; Cross-modal feature fusion is performed on the optical features and the radar features to obtain fused features; Classification and identification are performed based on the fused features to determine the disturbance status of the target forest area; The step of performing cross-modal feature fusion on the optical features and the radar features to obtain fused features includes: Through the feature stitching layer of the disturbance detection model, the weight contribution of the optical features and the radar features is dynamically adjusted according to the current meteorological conditions. Cross-modal feature fusion is then performed on the adjusted optical features and radar features to obtain preliminary fused features. The step of dynamically adjusting the weight contribution of the optical features and the radar features according to the current meteorological conditions includes: acquiring the current meteorological conditions and generating an imaging quality score based on them, where the imaging quality score reflects the current imaging quality of the optical image; the weight contribution of the optical features is directly proportional to the imaging quality score, and the weight contribution of the radar features is inversely proportional to the imaging quality score. A self-attention mechanism is used to perform deep cross-modal feature fusion on the preliminary fused features to obtain fused features; The optical feature extraction branch employs a residual network, including multiple convolutional layers, pooling layers, and activation function layers. The step of inputting the optical image of the target forest area into the optical feature extraction branch of the preset perturbation detection model to obtain optical features includes: The optical image of the target forest area is input into the optical feature extraction branch, and the optical features are obtained by propagating forward layer by layer according to the network structure of the optical feature extraction branch. The radar feature extraction branch employs a convolutional network, including multiple convolutional layers, batch normalization layers, and activation function layers. The step of inputting the radar image of the target forest area into the radar feature extraction branch of the disturbance detection model to obtain radar features includes: The radar image of the target forest area is input into the radar feature extraction branch, and the radar features are obtained by propagating forward layer by layer according to the network structure of the radar feature extraction branch.

2. The forest disturbance detection method as described in claim 1, characterized in that, Before the step of inputting the optical image of the target forest area into the optical feature extraction branch of the preset disturbance detection model to obtain the optical features, the method further includes: Acquire training sample data containing forest areas under different disturbance states, wherein the training sample data includes optical image samples and radar image samples, and each sample is labeled with a disturbance state category label; The perturbation detection model is obtained by training the deep learning network to be trained based on the training sample data. The loss function used in the network training includes the binary cross-entropy loss function and the Dess loss function.

3. The forest disturbance detection method as described in claim 1, characterized in that, Before the step of inputting the optical image of the target forest area into the optical feature extraction branch of the preset disturbance detection model to obtain the optical features, the method further includes: Acquire initial optical and radar images of the target forest area; The initial optical image is subjected to a first preprocessing operation to obtain an optical image, wherein the first preprocessing operation includes atmospheric correction, radiometric correction, geometric correction, handling of missing values ​​and outliers, data normalization and standardization. The initial radar image is subjected to a second preprocessing operation to obtain a radar image. The second preprocessing operation includes applying orbit files, removing boundary noise and thermal noise, radiometric correction, terrain correction, and fine processing of ground distance image data.

4. The forest disturbance detection method as described in claim 1, characterized in that, The step of classifying and identifying based on the fused features to determine the disturbance state of the target forest area includes: The fused features are reduced in dimension by using the fully connected layer of the perturbation detection model to obtain the reduced feature vector. The perturbation detection model uses a classifier to classify the feature vector based on a preset classification threshold and outputs the classification result. Based on the classification results and combined with preset disturbance state category labels, the disturbance state of the target forest area is determined, wherein the disturbance state category labels include undisturbed, logging disturbance, insect pest disturbance, and fire disturbance.

5. A forest disturbance detection device, characterized in that, The forest disturbance detection device includes: The optical feature extraction module is used to input the optical image of the target forest area into the optical feature extraction branch of the preset disturbance detection model to obtain optical features, which include spatial texture features and spectral features. The radar feature extraction module is used to input radar images of the target forest area into the radar feature extraction branch of the disturbance detection model to obtain radar features, which include scattering features and terrain features. The feature fusion module is used to perform cross-modal feature fusion of optical and radar features to obtain fused features; The state recognition module is used to perform classification and recognition based on fused features to determine the disturbance state of the target forest area. The feature fusion module is also used for: Through the feature stitching layer of the disturbance detection model, the weight contribution of the optical features and the radar features is dynamically adjusted according to the current meteorological conditions. Cross-modal feature fusion is then performed on the adjusted optical and radar features to obtain preliminary fused features. The current meteorological conditions are acquired, and an imaging quality score is generated based on these conditions. This imaging quality score reflects the current imaging quality of the optical image. The weight contribution of the optical features is directly proportional to the imaging quality score, and the weight contribution of the radar features is inversely proportional to the imaging quality score. A self-attention mechanism is then used to perform deep cross-modal feature fusion on the preliminary fused features to obtain the fused features. The optical feature extraction branch employs a residual network, including multiple convolutional layers, pooling layers, and activation function layers. The optical feature extraction module is also used for: The optical image of the target forest area is input into the optical feature extraction branch, and the optical features are obtained by propagating forward layer by layer according to the network structure of the optical feature extraction branch. The radar feature extraction branch employs a convolutional network, including multiple convolutional layers, batch normalization layers, and activation function layers. The radar feature extraction module is also used for: The radar image of the target forest area is input into the radar feature extraction branch, and the radar features are obtained by propagating forward layer by layer according to the network structure of the radar feature extraction branch.

6. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the forest disturbance detection method as described in any one of claims 1 to 4.

7. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the forest disturbance detection method as described in any one of claims 1 to 4.

8. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the forest disturbance detection method as described in any one of claims 1 to 4.

Images