Remote sensing target recognition method, related device, equipment and storage medium

Abstract

The invention discloses a remote sensing target recognition method, a related device, equipment and a storage medium, and the method comprises the steps: obtaining a remote sensing image of a target geographic region where a to-be-recognized target object is located; performing target detection based on the remote sensing image to obtain a target image area of the target object; based on the target image region, extracting a first feature representation representing the appearance characteristics of the target object; determining irradiation parameters of a laser on the remote sensing platform based on the first feature representation; analyzing the optical response after the target object is irradiated by the laser according to the irradiation parameters to obtain a second characteristic representation representing the optical characteristics of the target object; and performing prediction based on the first feature representation and the second feature representation to obtain an identification result of the target object about the type to which the target object belongs. According to the scheme, the recognition precision of the remote sensing target can be guaranteed as much as possible, and computing resources are reduced.

Description

Technical Field

[0001] The present application relates to the field of image recognition technology, and in particular to a remote sensing target recognition method and related devices, equipment and storage media. Background Art

[0002] Remote sensing technology uses electromagnetic waves to observe the earth's surface for tasks such as environmental monitoring and target identification. Among them, the use of remote sensing technology to complete target identification tasks has broad prospects in many scenarios such as transportation and coastal defense.

[0003] However, existing remote sensing target recognition technologies are either difficult to guarantee recognition accuracy. For example, target recognition through visible light imaging is easily affected by changes in illumination and background, resulting in insufficient recognition accuracy. Or, they have high requirements for computing resources. For example, although synthetic aperture radar technology has penetration capabilities, it requires high computing resources. In view of this, how to ensure the recognition accuracy of remote sensing targets as much as possible and reduce computing resources has become an urgent problem to be solved. Summary of the invention

[0004] The main technical problem solved by the present application is to provide a remote sensing target recognition method and related devices, equipment and storage media, which can ensure the recognition accuracy of remote sensing targets as much as possible and reduce computing resources.

[0005] In order to solve the above-mentioned technical problems, the first aspect of the present application provides a remote sensing target recognition method, including: obtaining a remote sensing image of a target geographical area where a target object to be identified is located; performing target detection based on the remote sensing image to obtain a target image area of the target object; based on the target image area, extracting a first feature representation characterizing the appearance characteristics of the target object; based on the first feature representation, determining the illumination parameters of a laser on the remote sensing platform; based on the optical response after the laser illuminates the target object according to the illumination parameters, obtaining a second feature representation characterizing the optical characteristics of the target object; and performing prediction based on the first feature representation and the second feature representation to obtain an identification result of the target object regarding its type.

[0006] To solve the above technical problems, the second aspect of the present application provides a remote sensing target recognition device, including: an image acquisition module, a target detection module, a first extraction module, a parameter determination module, a second extraction module, and a type prediction module. The image acquisition module is configured to acquire a remote sensing image of a target geographical area where a target object to be recognized is located; the target detection module is configured to perform target detection based on the remote sensing image to obtain a target image area of the target object; the first extraction module is configured to extract a first feature representation characterizing the appearance characteristics of the target object based on the target image area; the parameter determination module is configured to determine the irradiation parameters of a laser on a remote sensing platform based on the first feature representation; the second extraction module is configured to analyze the optical response after the laser irradiates the target object according to the irradiation parameters to obtain a second feature representation characterizing the optical characteristics of the target object; the type prediction module is configured to perform prediction based on the first feature representation and the second feature representation to obtain an identification result of the target object regarding its belonging type.

[0007] To solve the above technical problems, the third aspect of the present application provides an electronic device, at least including a memory and a processor coupled to each other. The memory stores at least program instructions, and the processor is configured to execute the program instructions to implement the remote sensing target recognition method in the first aspect above.

[0008] To solve the above technical problems, the fourth aspect of the present application provides a computer-readable storage medium storing program instructions that can be run by a processor, and the program instructions are used to implement the remote sensing target recognition method in the first aspect above.

[0009] In the above solution, a remote sensing image of the target geographical area where the target object to be recognized is located is obtained, target detection is performed based on the remote sensing image to obtain the target image area of the target object, and based on the target image area, a first feature representation characterizing the appearance characteristics of the target object is extracted. Based on the first feature representation, the irradiation parameters of the laser on the remote sensing platform are determined, and then based on the optical response after the laser irradiates the target object according to the irradiation parameters, a second feature representation characterizing the optical characteristics of the target object is obtained. Furthermore, based on the first feature representation and the second feature representation, a prediction is made to obtain the recognition result of the target object regarding its type. On the one hand, in the process of remote sensing target recognition, a series of process steps such as target detection, feature extraction, parameter determination, optical analysis, and type prediction can be used to complete the remote sensing target recognition without involving complex data processing, which helps to reduce computing resources. On the other hand, first, a first feature representation characterizing the appearance characteristics is extracted to more specifically determine the irradiation parameters of the laser accordingly, and the laser irradiates the target object according to these irradiation parameters, and the optical response of the target object is analyzed to obtain a second feature representation characterizing the optical characteristics of the target object, which helps to improve the accuracy of the optical characteristics. Finally, the recognition result of the target object regarding its type is jointly predicted by combining the first feature representation and the second feature representation, that is, the recognition prediction is performed by combining the feature representations of the target object's characteristics in different aspects, which further helps to improve the accuracy of remote sensing target recognition. Therefore, it is possible to ensure the recognition accuracy of remote sensing targets as much as possible and reduce computing resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 is a flowchart of an embodiment of the remote sensing target recognition method of the present application; Figure 2 is a framework diagram of an embodiment of the remote sensing target recognition device of the present application; Figure 3 is a framework diagram of an embodiment of the electronic device of the present application; Figure 4 is a framework diagram of an embodiment of the computer-readable storage medium of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0011] The following will combine the accompanying drawings of the specification to elaborate on the solutions of the embodiments of the present application in detail.

[0012] In the following description, specific details such as specific system structures, interfaces, and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the present application.

[0013] The terms "system" and "network" are often used interchangeably herein. The term "and / or" herein merely describes the associated relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the segment " / " herein generally represents an "or" relationship between the associated objects before and after. Furthermore, "plurality" herein means two or more than two.

[0014] Please refer to Figure 1 , Figure 1 is a schematic flowchart of an embodiment of the remote sensing target recognition method of this application. Specifically, it may include the following steps: Step S11: Obtain a remote sensing image of the target geographical area where the target object to be recognized is located.

[0015] In an implementation scenario, the target object to be recognized can be set according to the recognition needs. For example, in scenarios such as ocean monitoring, the target object to be recognized can be a ship; or, in scenarios such as traffic monitoring, the target object to be recognized can be a vehicle; or, in scenarios such as agricultural operations, the target object to be recognized can be a crop; or, in scenarios such as urban planning, the target object to be recognized can be a building. Of course, the above examples are only several possible examples of the target object to be recognized, and the target object is not limited herein. For example, the target object to be recognized can also include but is not limited to: forests, roads, terrains, equipment, animals, etc., and no further examples will be given here.

[0016] In an implementation scenario, detectors on a remote sensing platform in several bands can be used to detect and image the target geographical area respectively to obtain several first image data, and the first image data is attached with geographical positioning information, and preprocessing is performed on the several first image data respectively to obtain several second image data, and the preprocessing at least includes image registration performed based on the geographical positioning information, and then the several second image data are weighted based on the respective fusion weights of the several second image data to obtain a remote sensing image. The above method combines multi-band detection and imaging, and through a series of operations such as preprocessing and weighted fusion, obtains a remote sensing image to perform target recognition based on this, which can help analyze the target from multiple angles simultaneously, thereby overcoming the limitations of a single-band detector in a complex environment, and further improving the robustness of target recognition.

[0017] In a specific implementation scenario, the remote sensing platform can include but is not limited to: satellites, airplanes, drones, etc., and the specific type of the remote sensing platform is not limited herein. In addition, the remote sensing platform can also be selected according to task requirements, such as spatial resolution, temporal resolution, coverage range, etc.

[0018] In a specific implementation scenario, several bands may include but are not limited to: visible light, infrared, radar, etc. It should be noted that different bands can reflect different physical and chemical properties of the target object. For example, the visible light band can usually be used to capture information such as the shape and color of the target object, the infrared band can usually be used to detect information such as the temperature and thermal radiation of the target object, and the radar band can usually be used to detect information such as the geometric structure of the target object.

[0019] In a specific implementation scenario, the detector can detect and image a target geographical area on a remote sensing platform along a predetermined orbit and at a predetermined time to obtain first image data in different bands. It should be noted that during the detection and imaging process, factors such as atmospheric conditions, solar angle, and detector calibration can be considered to ensure that the first image data has relatively high quality and consistency as much as possible. For example, atmospheric scattering and absorption effects will affect the spectral characteristics of the image, and atmospheric correction techniques need to be used for compensation. At the same time, in order to ensure the spatial registration accuracy of images in different bands, high-precision attitude and orbit data can be used for assistance. After the image acquisition is completed, the first image data of all bands will be stored and transmitted to the ground processing system. At this stage, the first image data is usually stored in the original format. For example, the first image data can be attached with geolocation information.Exemplarily, the geolocation information may include longitude and latitude coordinates, that is, each pixel point in the first image data can be located by its longitude and latitude coordinates on the map; in addition, the geolocation information may also include spatial resolution and coordinate system, that is, the actual size of each pixel point (e.g., ground resolution, usually expressed in meters or kilometers) will vary according to the design of the detector and the shooting angle of the image, and this information helps to determine the actual ground area size represented by each pixel point; in addition, the geolocation information may also include projection information (e.g., projection coordinate system, projection area, projection offset, etc.), where the projection coordinate system may include, but is not limited to, UTM (Universal Transverse Mercator Grid System), Mercator projection, etc., the projection area represents the specific area or zone of the projection, which is used to illustrate the geographical range and location covered by the image, and the projection offset represents the deviation between the starting point of the image and the Earth coordinate system; in addition, the geolocation information may also include ground control points (GCPs), which are specific marked points corresponding to the real ground positions of the image, usually the geographical positions suppressed on the ground, and the remote sensing platform can help to correct the image by measuring the actual coordinates of these points to ensure the accuracy of the image; in addition, the geolocation information may also include spatial resolution and projection parameters, where the spatial resolution represents the actual area covered by each pixel point on the ground, which is related to the imaging angle of the remote sensing platform and the resolution of the detector, and the projection parameters may include scanning angle and detector parameters, which are crucial for accurately determining the acquisition position and angle of the image data, especially in satellite remote sensing data for ground reconstruction (especially stereo imaging) and image geometric correction; in addition, the geolocation information may also include time information, that is, the time point when the image data is collected, which is used to help judge the validity and timeliness of the image data, especially in applications such as dynamic monitoring, such as climate change, disaster assessment, etc.; in addition, the geolocation information may also include track and attitude information. In satellite remote sensing or aerial imagery, the geolocation of the image usually also includes track and attitude information when the image is acquired. The track refers to the trajectory of the sensor (such as the orbit of the satellite), and the attitude information describes information such as the angle and direction of the sensor. These data can help to more precisely calibrate the image and ensure its high consistency with the ground position; in addition, the geolocation information may also include internal and external parameters of the detector, where the internal parameters may include the focal length of the detector, the size of the photosensitive element, the aperture, etc., which are used to help calculate the precise position of each pixel point on the ground, and the external parameters may include the position and orientation of the detector during detection and imaging, which is crucial for the spatial imaging of the image data, especially in applications such as stereo image generation and 3D reconstruction; in addition, the geolocation information may also include other metadata, such as image coordinate system, time stamp, etc.

[0020] In a specific implementation scenario, after obtaining the first image data of several bands, the first image data of several bands can be preprocessed separately. As a possible example, the preprocessing may include geometric correction, specifically, based on geolocation information (such as digital elevation model, etc.) and the attitude trajectory data of the remote sensing platform, the first image data can be resampled to the standard map projection coordinate system through geometric transformation, and after that, image registration can be performed to ensure the consistency of subsequent analysis. It should be noted that the purpose of geometric correction is to eliminate geometric distortion. Geometric correction can specifically include ground control point correction, distortion correction, etc. Ground control point correction is used to align the pixel positions of the image data with known ground positions and correct the distortion caused by sensor attitude, orbital deviation, or Earth curvature, while distortion correction is used to correct the effects including but not limited to atmospheric refraction effects, sensor angle errors, etc. In multi-band remote sensing data, the sensor characteristics and imaging angles of each band may be slightly different. Therefore, by performing separate geometric correction on the first image data of each band, the spatial alignment of all bands can be ensured, and the position inconsistency between bands can be avoided. As another possible example, the preprocessing can also include radiometric correction. Radiometric correction is used to convert the radiance values in the first image data into the actual reflectance or radiance of the ground object, and the radiometric correction can include at least one of the following: atmospheric correction, sensor correction, and illumination correction. It should be noted that the main purpose of radiometric correction is to correct the deviation of the image radiance values caused by factors such as atmospheric scattering, absorption, and sensor characteristics. Radiometric information directly reflects the spectral characteristics of the ground object. However, the changes in aerosols, clouds, and solar angles in the atmosphere will all affect the transmission of the radiometric signal. Therefore, through radiometric correction, the interference of factors such as the atmosphere and sensor effects on the spectral characteristics of the image can be effectively eliminated. In addition, atmospheric correction is mainly used to remove the effects of water vapor, aerosols, etc. in the atmosphere on different bands. Sensor correction is mainly used to calibrate the response characteristics of the sensor to ensure the measurement consistency of different bands. Illumination correction is used to eliminate the influence of different illumination conditions on the image radiance values. As yet another possible example, the preprocessing can also include noise removal. Noise removal can include at least one of the following: mean filtering, median filtering, and low-pass filtering, to reduce noise by smoothing the image or suppressing high-frequency components while retaining the edge and detail information of the image. It should be noted that the main purpose of noise removal is to eliminate the noise introduced into the image due to sensor failures, atmospheric interference, or data transmission, etc., which can improve the image quality. Through the above preprocessing, the accuracy of subsequent operations can be further improved. For example, for multi-band images, the accuracy of geometric correction and radiometric correction is crucial for subsequent image fusion. For example, when fusing high-resolution visible light images and low-resolution infrared images, it is necessary to ensure that they are effectively aligned both spatially and radiometrically, which helps to improve the accuracy of a series of subsequent operations such as feature extraction and target recognition.Of course, the above examples are only several possible examples of preprocessing in the actual application process. The possible situations where preprocessing includes other operations are not limited here, and no more examples will be given one by one.

[0021] In a specific implementation scenario, after the first image data in several bands is preprocessed to obtain the second image data in several bands, the respective fusion weights of the second image data in several bands can be obtained, and based on this, the second image data in several bands can be weighted to obtain a remote sensing image. As a possible example, in order to obtain the respective fusion weights of the second image data in several bands, the salience of the target object imaged in different bands can be determined based on the attribute information of the target object, and based on the salience of the target object imaged in different bands and the bands to which the respective second image data belong, the respective fusion weights of the second image data in several bands can be determined, and the salience is positively correlated with the fusion weight, that is to say, the higher the salience, the greater the fusion weight, and vice versa, the lower the salience, the smaller the fusion weight. Exemplarily, taking the attribute information of the target object including the target type (such as, ship, building, vegetation, etc.) as an example, different bands can reveal different features. For example, in a marine scenario, the infrared band is often used to capture the thermal characteristics of the target, while visible light images may be more suitable for capturing features such as the shape of the target. Then, if the key characteristics of the target object are more prominent in a certain band (for example, the infrared band can provide thermal imaging information about the target object, while the visible light band pays more attention to reflection characteristics), then a higher fusion weight can be configured for the corresponding band. As another possible example, in order to obtain the respective fusion weights of the second image data in several bands, the image data of the target object in the second image data can also be analyzed to obtain the statistical characteristics of the target object in the second image data, and dimensionality reduction can be performed based on the statistical characteristics of the target object in each of the second image data respectively to obtain the respective fusion weights of the second image data in several bands. Exemplarily, the statistical characteristics can include but are not limited to entropy value, variance, etc., and based on this, the advantage of an image in a certain band in characterizing the target characteristics can be judged. Then, specific dimensionality reduction methods such as principal component analysis can be used to evaluate the contribution of each band to the target feature information, and further determine its relative weight to obtain the respective fusion weights of the second image data in several bands. As yet another possible example, in order to obtain the respective fusion weights of the second image data in several bands, the second image data in several bands can also be predicted based on the weight prediction model with reference to the attribute information of the target object to obtain the respective fusion weights of the second image data in several bands. Exemplarily, machine learning (such as, convolutional neural network, etc.) can be used for image classification and target detection. By annotating and training the training data, the network can automatically learn the importance of different bands and adjust the weights of different bands through a weighting strategy. In other words, the model can automatically adjust the fusion weight of each band according to the influence of the features of different bands on target classification; or, some deep learning frameworks can also dynamically adjust the fusion weight according to the response and learning process of the input images in different bands. For example, in the process of multi-band fusion, the network can learn the importance of certain bands for recognition accuracy in certain situations (such as, low visibility or night recognition).As another possible example, in order to obtain the respective fusion weights of the second image data in several bands, the respective fusion weights of the second image data in several bands can also be determined based on at least one of the application environment and task requirements of remote sensing target recognition and the bands to which the second image data in several bands belong respectively. For example, in some applications, certain bands may become more important due to special task requirements. For example, in security monitoring, visible light images and infrared images may be equally important, but at night, infrared images become more critical because they can provide thermal information, and during the day, visible light images may be more emphasized; or, for example, different meteorological or environmental conditions (such as haze, rain, snow, night, etc.) will affect the image quality and the effectiveness of different bands. For example, in bad weather, infrared images may have more advantages, while on sunny days, visible light images may be clearer. As another possible example, in order to obtain the respective fusion weights of the second image data in several bands, the respective fusion weights of the second image data in several bands can also be determined based on at least one of the respective noise levels, image resolutions, and spectral responses of the second image data in several bands. For example, if the image quality of a certain band is poor and there is more noise, it may be necessary to adjust its weight to reduce its impact on the final result; or, for example, the spatial resolutions or spectral responses of different bands may be different, which also needs to be considered when determining the weights. A higher spatial resolution means more detailed target information and may therefore require a higher fusion weight. Of course, the above examples are only several possible examples of determining the fusion weights, and other possible situations are not limited here, nor will they be exemplified one by one. For example, a comprehensive strategy can also be used to determine the fusion weights. Exemplarily, based on methods such as mutual information, correlation coefficient, or other similarity metrics, the most informative bands can be selected in the data preprocessing stage to help determine which bands should be given higher fusion weights.

[0022] In a specific implementation scenario, after obtaining the respective fusion weights of the second image data in several bands, the second image data in several bands can be fused accordingly to obtain a remote sensing image. Specifically, multi-scale fusion algorithms such as, but not limited to, wavelet transform and contourlet transform can be used. The basic idea is to decompose the second image data into sub-bands of different scales, directions, and then perform fusion processing on each sub-band. Finally, the fused sub-bands are recombined into a fused image, that is, a remote sensing image. For example, wavelet transform can effectively retain the edge and detail information in the image through multi-scale decomposition of the image. During the fusion process, it can consider both low-frequency information (such as brightness) and high-frequency information (such as texture), thereby generating an image containing rich information, which is used as a remote sensing image. It should be noted that the remote sensing image not only retains the advantages of each band image but also makes up for the deficiencies of a single-band image, providing a richer information basis for subsequent target feature extraction and recognition. Through image fusion, the resolvability and recognition accuracy of the target can be effectively improved. Especially under complex backgrounds or low-contrast conditions, the remote sensing image can more clearly present the key features of the target.

[0023] Step S12: Perform target detection based on the remote sensing image to obtain the target image region of the target object.

[0024] Specifically, the target detection of the remote sensing image can be performed through an image segmentation algorithm to detect the target image region of the target object in the remote sensing image. It should be noted that the image segmentation algorithm can include, but not limited to: threshold-based image segmentation, region growing, edge detection, etc. Of course, it can also be implemented using an image segmentation model such as U-Net. The image segmentation algorithm is not limited here. Taking threshold-based image segmentation as an example, it can separate the target object from the background in the remote sensing image by setting a gray value or color threshold, and thus obtain the target image region of the target object in the remote sensing image; or, taking edge detection as an example, it can detect the boundary of the target object and define the contour of the target object in the remote sensing image based on this, and thus obtain the target image region of the target object. Of course, the above examples are only several possible examples of detecting the target image region of the target object in the remote sensing image. Other possible methods are not limited here and will not be exemplified one by one.

[0025] Step S13: Extract the first feature representation characterizing the appearance characteristics of the target object based on the target image region.

[0026] In an implementation scenario, the first feature representation can specifically be manually designed features, which can specifically include, but not limited to, shape, texture, size, color, etc. The specific content of the first feature representation is not limited here.

[0027] In a specific implementation scenario, taking the first feature as an example of including shape features, as a possible example, the shape features can specifically include contour features. Then, the contour of the target object can be extracted through edge detection (e.g., Canny edge detection). For example, in the case where the target object is a ship, the shape features of the ship (such as the long strip shape of the hull, the flatness of the deck, etc.) can be extracted. Or, as another possible example, the shape features can specifically include geometric features. For example, in the case where the target object is a ship, the geometric structure characteristics of the ship, such as the aspect ratio, area, boundary curvature, etc., can be obtained. Geometric features can be extracted by calculating the geometric center, minimum bounding rectangle, convex hull, etc. of the target image area. Or, as yet another possible example, the shape features can specifically include shape descriptors, such as Hu moments, Zernike moments, etc.

[0028] In a specific implementation scenario, taking the first feature as an example of including texture features, as a possible example, the texture features can be specifically extracted through a gray-level co-occurrence matrix, which is used to describe the texture features in an image. By calculating the spatial relationship of the pixel gray values in the image, texture features such as roughness, contrast, and directionality in the image can be extracted. The texture characteristics of the ship's surface, such as the roughness of the hull coating, the structure of the ship's deck, etc., can all be quantitatively analyzed through this method. Or, as another possible example, the texture features can be specifically extracted through local binary patterns, which extract texture information by performing binary encoding on local regions and are suitable for tasks that are more sensitive to texture changes.

[0029] In a specific implementation scenario, taking the first feature as an example of including size features, as a possible example, the size features can include area and scale. For example, in the case where the target object is a ship, the area of the ship in the image is usually large, especially when captured through a high-resolution image. Then, the size features of the target can be extracted by calculating information such as the area, aspect ratio, and maximum bounding box size of the target region. Or, as another possible example, the size features can include scale-invariant features, which can extract stable local feature points at different scales of the image, and then derive information about the size, shape, and scale change of the target.

[0030] In a specific implementation scenario, when the first feature representation includes color features, as a possible example, the color features can be represented by a color histogram. Specifically, the pixel distribution of different color channels in the image can be statistically analyzed to generate a color histogram. Or, as another possible example, the color features can be obtained through color space conversion. For example, after converting an RGB image to another color space (such as HSV, Lab), features such as the saturation, brightness, and hue of the color can be extracted, which can better describe the color features of the ship, especially more effectively under the influence of light changes or shadows.

[0031] In another implementation scenario, different from the foregoing implementation, as another possible implementation example, the feature extraction model can also be used to extract features from the target image region to obtain the third feature representation of the target object. That is to say, different from the first feature representation which belongs to manually designed features, the third feature representation belongs to machine learning features. Specifically, the feature extraction model can include, but is not limited to, neural network models such as ResNet and VGG. The network structure of the feature extraction model is not limited here.

[0032] It should be noted that after feature extraction, as a possible example, in order to reduce feature redundancy and possible overfitting, dimensionality reduction operations can also be performed on the feature representation. For example, dimensionality reduction can be performed using methods such as principal component analysis, which is not limited here.

[0033] Step S14: Determine the irradiation parameters of the laser on the remote sensing platform based on the first feature representation.

[0034] In an implementation scenario, the lasers on the remote sensing platform can include, but are not limited to: tunable femtosecond lasers, ultrashort pulse laser systems, etc. The types of lasers are not limited here.

[0035] In an implementation scenario, as a possible example, the irradiation parameters may include wavelength. After obtaining the first feature representation, specifically, based on the feature elements representing the material in the first feature representation, the wavelength in the irradiation parameters can be determined. Specifically, according to the feature elements representing the material in the first feature representation, the absorption, reflection, or scattering characteristics of the target object for lasers of different wavelengths can be obtained. For example, some materials may have a higher reflectivity at specific wavelengths, while some materials may have a higher reflectivity at other wavelengths. On this basis, the wavelength with the strongest interaction with the target object can be selected and determined as the wavelength in the irradiation parameters, which can improve the reflectivity or excitation efficiency and help optimize the measurement results. It should be noted that lasers of certain wavelengths may be interfered by environmental or background noise, such as atmospheric scattering or interference signals from other targets. Therefore, as an optimized implementation example, the wavelength with the strongest interaction with the target object can be selected from the wavelengths not interfered by environmental or background noise and determined as the wavelength in the irradiation parameters.

[0036] In another implementation scenario, as another possible example, the irradiation parameters may include the irradiation mode. After obtaining the first feature representation, specifically, based on the feature elements representing the size and shape in the first feature representation, the irradiation mode in the irradiation parameters can be determined. Specifically, in response to the feature element representing the size in the first feature representation being higher than the size threshold and the feature element representing the shape being lower than the shape complexity threshold, the irradiation mode is determined to be one-time irradiation. That is to say, for a target object with a large size and a simple shape, the one-time irradiation mode can be adopted. Conversely, in response to the feature element representing the size in the first feature representation being lower than the size threshold and the feature element representing the shape being higher than the shape complexity threshold, the irradiation mode is determined to be scanning irradiation. That is to say, for a target object with a small size and a complex shape, the scanning irradiation mode can be adopted. It should be noted that in the dimension of size, for a target object with a large size, especially a target object with a relatively simple shape, one-time irradiation (i.e., wide-area irradiation) is usually more effective. By covering a large area with a laser, the target object can be irradiated at one time, and the optical response of the target object can be quickly obtained. Exemplarily, for large ships, buildings, etc., the one-time irradiation mode can be preferably adopted. Conversely, for a target object with a small size, a local scanning or point-by-point scanning irradiation mode is more suitable, which can ensure that each small area can be accurately irradiated as much as possible. Similarly, in the dimension of shape, for a target object with a relatively simple shape (e.g., a relatively smooth surface), the one-time irradiation mode can be adopted. Conversely, for a target object with a relatively complex shape (e.g., a surface with irregularities or concavities and convexities), scanning irradiation is more effective. For example, through point-by-point scanning, the irradiation angle and intensity of each small area can be finely controlled, and the reflection deviation caused by the complex shape can be avoided as much as possible.

[0037] In yet another implementation scenario, as another possible example, the irradiation parameters may include energy. After obtaining the first feature representation, specifically, the energy in the irradiation parameters can be determined based on the feature elements in the first feature representation that characterize the material and surface state. For example, according to the different responses of different materials to laser, especially the absorption characteristics of laser, such as the differences in absorption ability and thermal conductivity of materials like metals, ceramics, plastics, etc., it will affect the energy distribution and energy requirements of laser irradiation. Therefore, for different materials, the laser energy needs to be adjusted to ensure that the materials will not be overheated, melted or damaged during the laser irradiation process. Generally speaking, for materials with higher absorption rates (such as black or metals), lower laser energy may be required, while for materials with lower absorption rates (such as glass), higher laser energy may be required. In addition, as mentioned above, the surface state is also an important basis for adjusting the laser energy. If the surface of the target object is relatively smooth and has no obvious defects, the laser energy can be relatively low, while if the surface of the target object is irregular, rough or has attachments (such as oil stains, rust, etc.), higher laser energy may be required. In addition, the energy in the irradiation parameters can be further determined by combining factors such as the resonance characteristics between the target object and the laser wavelength, the size and irradiation area of the target object, the laser pulse frequency and repetition rate, etc. For example, when considering the resonance characteristics between the target object and the laser wavelength, since the interaction between lasers of different wavelengths and different materials will be different. For example, certain materials are more sensitive to infrared lasers and have weaker absorption of ultraviolet lasers. Therefore, by adjusting the wavelength to select a suitable laser spectrum, the energy transmission and absorption effects can be optimized, avoiding excessive damage and ensuring the laser signal intensity. Or, for example, when considering the size and irradiation area of the target object, if the size of the target is large or the surface area is wide, the irradiation energy of the laser needs to be evenly distributed over the entire surface to avoid overheating and damage in a small area. At this time, the energy of a single pulse can be appropriately reduced and the irradiation coverage area can be increased to avoid energy concentration in a small range. For smaller or finer targets, the focus of the laser irradiation is smaller, and usually a higher pulse energy can be adjusted to ensure the optical analysis effect in the local area. Or, for example, when considering the laser pulse frequency and repetition rate, if the pulse frequency is high, although the energy of each pulse is low, the energy accumulation may still cause overheating of the target surface. Therefore, it is necessary to adjust the energy of each pulse according to the heat resistance characteristics of the target and the repetition rate of the laser to avoid surface damage. In some cases, a lower pulse frequency and a longer time interval can help reduce heat accumulation and avoid damage to the target surface due to overexposure.

[0038] It should be noted that the above examples are only several possible examples for determining the irradiation parameters. Without loss of generality, the irradiation parameters can be specifically set to irradiate the target object as efficiently and comprehensively as possible without damaging the target object and obtain a sufficiently clear reflection signal. Other possible situations are not limited here, and no further examples will be given one by one.

[0039] Step S15: Analyze the optical response after the laser irradiates the target object according to the irradiation parameters to obtain a second feature representation characterizing the optical properties of the target object.

[0040] It should be noted that the optical response can include but is not limited to: reflection spectrum, fluorescence spectrum, etc. The specific content of the optical response is not limited here. Specifically, after the laser irradiates the target object according to the irradiation parameters, devices such as a spectrometer and a photodetector can be used to collect the reflection spectrum and fluorescence spectrum of the target object. Among them, the reflection spectrum is used to record the light intensity distribution reflected from the target object after being irradiated by lasers of different wavelengths. These data can reveal the reflection characteristics of the target object. For example, metals usually exhibit a high reflectivity, while non-metals may have specific absorption peaks. In addition, the fluorescence spectrum is used to record the fluorescence signal emitted by the target object under laser excitation. This signal can provide more microscopic molecular structure information of the material used by the target object. Especially when certain special coatings or pollutants are present, the fluorescence spectrum can help identify these characteristics.

[0041] In one implementation scenario, as a possible example, a second feature representation characterizing the optical properties of the target object can be obtained by comparing the optical response with the spectral diagrams of various materials in the spectral library.

[0042] In another implementation scenario, different from the foregoing implementation manner, as another possible example, a second feature representation characterizing the optical properties of the target object can be obtained by analyzing at least one of the peak value, intensity, and width based on the optical response.

[0043] It should be noted that no matter which of the above methods is used to analyze the optical response, when analyzing the reflection spectrum, the following can be obtained, including but not limited to: reflectivity curve, characteristic wavelength band of reflectivity, peaks and valleys, surface state, etc. Among them, the reflectivity curve is used to describe the intensity of the reflected light on the surface of the target object at different wavelengths. The reflectivity characteristics of different materials are the key factors for distinguishing different materials. For example, metal materials usually show high reflectivity in the visible and near-infrared bands, while non-metal materials may show low reflectivity in some bands. Certain specific wavelength points in the reflectivity spectrum often correspond to specific material characteristics. For example, metal materials may have high reflectivity in the near-infrared band, while some non-metal materials, such as wood, plastic, glass, etc., may show unique absorption or reflection characteristics in the visible and near-infrared bands. The peaks and valleys appearing in the reflection spectrum can reveal physical properties such as the surface roughness and gloss of the material, and are even related to the chemical composition of the material. For example, certain specific materials may have absorption peaks at certain specific wavelengths, and these characteristics can help to judge the material type. In the case of surface oxidation or contamination, the reflection spectrum of the material will change. For example, the oxide layer on the surface of iron may cause a significant decrease in its reflectivity in some bands. By analyzing these changes, the condition of the target surface, such as whether there is rust, contamination or coating, can be inferred. In addition, when analyzing the fluorescence spectrum, the following can be obtained, including but not limited to: fluorescence reflection peak, fluorescence decay characteristics, changes in the fluorescence chromatogram, etc. Among them, different materials will emit fluorescence with specific wavelengths after being irradiated by femtosecond lasers. The position, shape and intensity of the fluorescence emission peak can provide chemical information of the target material. For example, certain materials may have specific fluorescence emission wavelengths, and these characteristics can help to identify the type of the material. The change of fluorescence intensity is closely related to the surface characteristics, chemical composition of the material and the efficiency of the interaction between the laser and the material. A strong fluorescence signal may indicate that the material has high fluorescence activity and may be an indication of certain special chemical elements or compounds. The decay rate and lifetime of fluorescence are another important analysis data. The fluorescence of some materials may show a long decay time, which is related to the change of the molecular structure or electronic state of the material. By analyzing the fluorescence decay characteristics, more in-depth information about the target material can be obtained. The chemical composition, surface state of a specific material and the way of laser irradiation may affect the morphology of the fluorescence chromatogram. For example, the surface of some materials may show obvious fluorescence enhancement or decay under laser irradiation, and these changes can provide a basis for the identification of the material. Combining the above various information or different information can form the second feature representation.

[0044] Step S16: Make a prediction based on the first feature representation and the second feature representation to obtain the recognition result of the target object regarding its type.

[0045] In an implementation scenario, the specific category of the type to which the target object belongs can also be set according to the recognition needs. For example, when the target object is a ship, the specific category of the type to which the target object belongs can include, for example, fishing boats, cruise ships, refrigerated ships, car carriers, liquefied gas carriers, etc.; or, when the target object is a vehicle, the specific category of the type to which the target object belongs can include, for example, cars, vans, oil tankers, etc.; or, when the target object is a building, the specific category of the type to which the target object belongs can include, for example, estimated values representing different degrees of damage; or, when the target object is a crop, the specific category of the type to which the target object belongs can include, for example, estimated values representing the growth state. Of course, the above examples are only possible examples of the recognition results in different scenarios, and the possible contents of the recognition results in other scenarios are not listed one by one here.

[0046] In an implementation scenario, the recognition result can specifically be represented as a type label of the target object. Still taking the target object being a ship as an example, when the recognition result is represented by a category label, the recognition result of the target object can be "fishing boat". Or, the recognition result can specifically also be represented as the probability values (i.e., probability distribution) that the target object belongs to different types. Still taking the target object being a ship as an example, when the recognition result is represented by a probability distribution, the recognition result of the target object can be: "fishing boat" - 75%, "cruise ship" - 15%, "cargo ship" - 5%, "passenger ship" - 5%. Of course, the above examples are only several expression examples of the recognition result, and the other possible expression ways of the recognition result are not limited here, and the other possible expression ways are not listed one by one here either.

[0047] In an implementation scenario, after obtaining the first feature representation and the second feature representation, a machine learning model or a deep learning model can be used to process the first feature representation and the second feature representation to obtain the recognition result of the target object. For example, the machine learning model can include but is not limited to support vector machines, random forests, etc.; or, the deep learning model can include but is not limited to convolutional neural networks, etc., and the model types of the machine learning model or the deep learning model are not limited here. It should be noted that in the model training and validation process, cross - validation techniques can be used to evaluate the performance of the model. Among them, cross - validation divides the data set into multiple subsets for training and testing the model to ensure the generalization ability of the model on different data. This process helps to identify the overfitting situation of the model and optimize its performance. After training is completed, the model can be applied to actual data for target recognition. For example, the first feature representation and the second feature representation can be fused by splicing and other methods to obtain a fused feature representation, and the fused feature representation is input into the trained model. The model will classify the target according to the learned feature distribution to obtain the recognition result of the target object.

[0048] In another implementation scenario, different from the foregoing implementation manner, as another possible example, as described above, a third feature representation of the target object can also be extracted. Then, prediction can also be performed based on the first feature representation, the second feature representation, and the third feature representation to obtain an identification result. For example, the first feature representation, the second feature representation, and the third feature representation can be fused through concatenation or other fusion methods to obtain a fused feature representation, and the fused feature representation is input into a trained model. The model will classify the target according to the learned feature distribution to obtain the identification result of the target object.

[0049] In the above solution, a remote sensing image of the target geographical area where the target object to be identified is located is acquired, target detection is performed based on the remote sensing image to obtain the target image area of the target object, and based on the target image area, a first feature representation characterizing the appearance characteristics of the target object is extracted. Based on the first feature representation, the irradiation parameters of the laser on the remote sensing platform are determined. Then, based on the optical response after the laser irradiates the target object according to the irradiation parameters, a second feature representation characterizing the optical characteristics of the target object is obtained. Furthermore, prediction is performed based on the first feature representation and the second feature representation to obtain the identification result of the target object regarding its type. On the one hand, in the process of remote sensing target identification, a series of process steps such as target detection, feature extraction, parameter determination, optical analysis, and type prediction can be completed, without involving complex data processing, which helps to reduce computing resources. On the other hand, the first feature representation characterizing the appearance characteristics is extracted first to more specifically determine the irradiation parameters of the laser, and the laser irradiates the target object according to these irradiation parameters. By analyzing the optical response of the target object, a second feature representation characterizing the optical characteristics of the target object is obtained, which helps to improve the accuracy of the optical characteristics. Finally, the identification result of the target object regarding its type is jointly predicted by combining the first feature representation and the second feature representation, that is, the identification prediction is performed by combining the feature representations of the target object in different aspects, which further helps to improve the accuracy of remote sensing target identification. Therefore, it is possible to ensure the identification accuracy of remote sensing targets as much as possible and reduce computing resources.

[0050] Please refer to Figure 2 , Figure 2It is a schematic diagram of the framework of an embodiment of the remote sensing target recognition device of the present application. The remote sensing target recognition device 20 includes: an image acquisition module 21, a target detection module 22, a first extraction module 23, a parameter determination module 24, a second extraction module 25 and a type prediction module 26. The image acquisition module 21 is used to acquire a remote sensing image of a target geographical area where a target object to be identified is located; the target detection module 22 is used to perform target detection based on the remote sensing image to obtain a target image area of the target object; the first extraction module 23 is used to extract a first feature representation characterizing the appearance characteristics of the target object based on the target image area; the parameter determination module 24 is used to determine the irradiation parameters of the laser on the remote sensing platform based on the first feature representation; the second extraction module 25 is used to analyze the optical response after the laser irradiates the target object according to the irradiation parameters to obtain a second feature representation characterizing the optical characteristics of the target object; the type prediction module 26 is used to predict based on the first feature representation and the second feature representation to obtain the recognition result of the target object regarding its type.

[0051] In the above scheme, the remote sensing target recognition device 20 obtains a remote sensing image of a target geographical area where a target object to be recognized is located, performs target detection based on the remote sensing image, obtains a target image area of the target object, and extracts a first feature representation characterizing the appearance characteristics of the target object based on the target image area, determines the irradiation parameters of the laser on the remote sensing platform based on the first feature representation, and then analyzes the optical response after the laser irradiates the target object according to the irradiation parameters to obtain a second feature representation characterizing the optical characteristics of the target object, and then predicts based on the first feature representation and the second feature representation to obtain an identification result of the target object regarding its type. On the one hand, in the process of remote sensing target recognition, target detection, feature extraction, parameter determination, optical Remote sensing target recognition can be completed through a series of process steps such as optical analysis and type prediction, without involving complex data processing, which helps to reduce computing resources. On the other hand, the first feature representation that characterizes the appearance characteristics is first extracted to more specifically determine the laser's irradiation parameters, and the laser is used to irradiate the target object according to the irradiation parameters, so as to obtain the second feature representation that characterizes the optical characteristics of the target object through the optical response analysis of the target object, which helps to improve the accuracy of the optical characteristics, and finally the first feature representation and the second feature representation are combined to jointly predict the recognition result of the target object about its type, that is, the feature representations of the characteristics of the target object in different aspects are combined for recognition prediction, which further helps to improve the accuracy of remote sensing target recognition. Therefore, the recognition accuracy of remote sensing targets can be guaranteed as much as possible, and computing resources can be reduced.

[0052] In some disclosed embodiments, the image acquisition module 21 includes a detection sub-module for respectively detecting and imaging a target geographical area by detectors in a plurality of bands on a remote sensing platform to obtain a plurality of first image data; wherein, the first image data is attached with geographical positioning information; the image acquisition module 21 includes a preprocessing sub-module for respectively performing preprocessing on the plurality of first image data to obtain a plurality of second image data; wherein, the preprocessing at least includes image registration performed based on the geographical positioning information; the image acquisition module 21 includes a weighting sub-module for weighting the plurality of second image data based on the respective fusion weights of the plurality of second image data to obtain a remote sensing image.

[0053] In some disclosed embodiments, the preprocessing includes geometric correction, and the preprocessing sub-module includes a geometric correction unit for performing the following geometric correction: based on the geographical positioning information and the attitude trajectory data of the remote sensing platform, resampling the first image data to a standard map projection coordinate system through geometric transformation, and performing image registration after resampling.

[0054] In some disclosed embodiments, the preprocessing includes radiometric correction, and the radiometric correction is used to convert the radiance value in the first image data into the actual reflectance or radiance of the ground object, and the radiometric correction includes at least one of atmospheric correction, sensor correction, and illumination correction.

[0055] In some disclosed embodiments, the preprocessing includes noise removal, and the noise removal includes at least one of mean filtering, median filtering, and low-pass filtering.

[0056] In some disclosed embodiments, the weighting sub-module includes a first determination unit for determining the saliency of the target object in imaging in different bands based on the attribute information of the target object, and determining the respective fusion weights of the plurality of second image data based on the saliency of the target object in imaging in different bands and the bands to which the plurality of second image data respectively belong; wherein, the saliency is positively correlated with the fusion weight; the weighting sub-module includes a second determination unit for analyzing the statistical features of the target object in the second image data based on the image data of the target object in the second image data, and performing dimensionality reduction based on the statistical features of the target object in each of the second image data to obtain the respective fusion weights of the plurality of second image data; the weighting sub-module includes a third determination unit for predicting the plurality of second image data based on the weight prediction model with reference to the attribute information of the target object to obtain the respective fusion weights of the plurality of second image data; the weighting sub-module includes a fourth determination unit for determining the respective fusion weights of the plurality of second image data based on at least one of the application environment and task requirements of remote sensing target recognition and the bands to which the plurality of second image data respectively belong; the weighting sub-module includes a fifth determination unit for determining the respective fusion weights of the plurality of second image data based on at least one of the respective noise levels, image resolutions, and spectral responses of the plurality of second image data.

[0057] In some disclosed embodiments, the first feature is represented as a manually designed feature. The remote sensing target recognition device 20 includes a third extraction module for extracting features from the target image region based on a feature extraction model to obtain a third feature representation of the target object. The type prediction module 26 is specifically configured to perform a prediction based on the first feature representation, the second feature representation, and the third feature representation to obtain an identification result.

[0058] In some disclosed embodiments, the parameter determination module 24 is specifically configured to perform at least one of the following: determine the wavelength in the irradiation parameters based on the feature elements representing the material in the first feature representation; determine the irradiation mode in the irradiation parameters based on the feature elements representing the size and shape in the first feature representation; determine the energy in the irradiation parameters based on the feature elements representing the material and the surface state in the first feature representation.

[0059] In some disclosed embodiments, the parameter determination module 24 includes a first response sub-module for determining that the irradiation mode is a one-time irradiation in response to the feature element representing the size in the first feature representation being higher than the size threshold and the feature element representing the shape being lower than the shape complexity threshold; the parameter determination module 24 includes a second response sub-module for determining that the irradiation mode is a scanning irradiation in response to the feature element representing the size in the first feature representation being lower than the size threshold and the feature element representing the shape being higher than the shape complexity threshold.

[0060] In some disclosed embodiments, the second extraction module 25 is specifically configured to perform at least one of the following: obtain a second feature representation characterizing the optical properties of the target object by comparing the optical response with the spectral diagrams of various materials in the spectral library; obtain a second feature representation characterizing the optical properties of the target object by analyzing at least one of the peak value, intensity, and width of the optical response.

[0061] In some disclosed embodiments, the target object at least includes a ship; and / or, the identification result includes any one of the type label of the target object and the probability values of the target object belonging to different types respectively; and / or, the optical response includes at least one of a reflection spectrum and a fluorescence spectrum.

[0062] Please refer to Figure 3 , Figure 3 is a schematic framework diagram of an embodiment of an electronic device according to the present application. The electronic device 30 at least includes a memory 31 and a processor 32 that are coupled to each other. The memory 31 stores at least program instructions, and the processor 32 is configured to execute the program instructions to implement the steps in any of the above-mentioned embodiments of the remote sensing target recognition method. Specifically, reference can be made to the foregoing disclosed embodiments, which will not be elaborated herein. As a possible example, the electronic device 30 may include, but is not limited to, a server, etc., and the specific type of the electronic device 30 is not limited herein.

[0063] Specifically, the processor 32 is used to control itself and the memory 31 to implement the steps in any of the above-described remote sensing target recognition method embodiments. The processor 32 can also be referred to as a CPU (Central Processing Unit). The processor 32 may be an integrated circuit chip with signal processing capabilities. The processor 32 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. Additionally, the processor 32 can be implemented jointly by integrated circuit chips.

[0064] In the above solution, the electronic device 30 acquires a remote sensing image of the target geographical area where the target object to be recognized is located, performs target detection based on the remote sensing image to obtain the target image area of the target object, and based on the target image area, extracts a first feature representation characterizing the appearance characteristics of the target object. Based on the first feature representation, the irradiation parameters of the laser on the remote sensing platform are determined. Then, based on the optical response after the laser irradiates the target object according to the irradiation parameters, a second feature representation characterizing the optical characteristics of the target object is obtained. Furthermore, based on the first feature representation and the second feature representation, a prediction is made to obtain the recognition result of the target object regarding its type. On the one hand, in the process of remote sensing target recognition, a series of process steps such as target detection, feature extraction, parameter determination, optical analysis, and type prediction can complete the remote sensing target recognition without involving complex data processing, which helps reduce computing resources. On the other hand, first, a first feature representation characterizing the appearance characteristics is extracted to more specifically determine the irradiation parameters of the laser accordingly, and the laser irradiates the target object according to these irradiation parameters, and the optical response of the target object is analyzed to obtain a second feature representation characterizing the optical characteristics of the target object, which helps improve the accuracy of the optical characteristics. Finally, the recognition result of the target object regarding its type is jointly predicted by combining the first feature representation and the second feature representation, that is, the recognition prediction is made by combining the feature representations of the target object in different aspects, which further helps improve the accuracy of remote sensing target recognition. Therefore, it is possible to ensure the recognition accuracy of remote sensing targets as much as possible and reduce computing resources.

[0065] Please refer to Figure 4 , Figure 4The computer-readable storage medium 40 stores program instructions 41 that can be executed by a processor, and the program instructions 41 are used to implement the steps in any of the above-mentioned remote sensing target recognition method embodiments.

[0066] In the above scheme, the computer-readable storage medium 40 obtains a remote sensing image of a target geographical area where a target object to be identified is located, performs target detection based on the remote sensing image, obtains a target image area of the target object, and extracts a first feature representation characterizing the appearance characteristics of the target object based on the target image area, determines the irradiation parameters of the laser on the remote sensing platform based on the first feature representation, and then analyzes the optical response after the laser irradiates the target object according to the irradiation parameters to obtain a second feature representation characterizing the optical characteristics of the target object, and then predicts based on the first feature representation and the second feature representation to obtain an identification result of the target object regarding its type. On the one hand, in the process of remote sensing target recognition, target detection, feature extraction, parameter determination, Remote sensing target recognition can be completed through a series of process steps such as optical analysis and type prediction, without involving complex data processing, which helps to reduce computing resources. On the other hand, the first feature representation that characterizes the appearance characteristics is first extracted to more specifically determine the laser's irradiation parameters, and the laser is used to irradiate the target object according to the irradiation parameters, so as to obtain the second feature representation that characterizes the optical characteristics of the target object through the optical response analysis of the target object, which helps to improve the accuracy of the optical characteristics, and finally the first feature representation and the second feature representation are combined to jointly predict the recognition result of the target object regarding its type, that is, the feature representations of the target object's characteristics in different aspects are combined for recognition prediction, which further helps to improve the accuracy of remote sensing target recognition. Therefore, the recognition accuracy of remote sensing targets can be guaranteed as much as possible, and computing resources can be reduced.

[0067] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0068] The above description of various embodiments tends to emphasize the differences between the various embodiments. The same or similar aspects can be referenced to each other, and for the sake of brevity, they will not be repeated herein.

[0069] In several embodiments provided in the present application, it should be understood that the disclosed methods and apparatuses can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections between each other can be through some interfaces. The indirect couplings or communication connections of the apparatuses or units can be in electrical, mechanical or other forms.

[0070] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0071] In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0072] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods in each embodiment of the present application. The foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs and other various media that can store program codes.

[0073] If the technical solution of this application involves personal information, the product using the technical solution of this application has clearly informed the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using the technical solution of this application has obtained the individual's separate consent before processing the sensitive personal information, and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, clear and prominent signs are set to inform that the personal information collection scope has been entered and personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that he or she agrees to the collection of his or her personal information; or on the device that processes personal information, the personal information processing rules are notified by obvious signs / information, and the individual's authorization is obtained through pop-up information or by asking the individual to upload his or her personal information; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

Claims

1. A remote sensing target recognition method, characterized in that Including: Obtaining a remote sensing image of a target geographical area where a target object to be recognized is located; Performing target detection based on the remote sensing image to obtain a target image area of the target object; Extracting a first feature representation characterizing the appearance characteristics of the target object based on the target image area; Determining the irradiation parameters of a laser on a remote sensing platform based on the first feature representation; Analyzing the optical response after the laser irradiates the target object according to the irradiation parameters to obtain a second feature representation characterizing the optical characteristics of the target object; Performing prediction based on the first feature representation and the second feature representation to obtain an identification result of the target object regarding its type.

2. The method according to claim 1, wherein The obtaining of the remote sensing image of the target geographical area where the target object to be recognized is located includes: Performing detection imaging on the target geographical area respectively by detectors of several bands on the remote sensing platform to obtain several first image data; wherein, the first image data is attached with geographical positioning information; Performing preprocessing on the several first image data respectively to obtain several second image data; wherein, the preprocessing at least includes image registration performed based on the geographical positioning information; Weighting the several second image data based on the respective fusion weights of the several second image data to obtain the remote sensing image.

3. The method according to claim 2, wherein The preprocessing includes geometric correction, and the execution steps of the geometric correction include: resampling the first image data to a standard map projection coordinate system through geometric transformation based on the geographical positioning information and the attitude trajectory data of the remote sensing platform, and the image registration is performed after the resampling; And / or, the preprocessing includes radiometric correction, and the radiometric correction is used to convert the radiance value in the first image data into the actual reflectance or radiance of the ground object, and the radiometric correction includes at least one of atmospheric correction, sensor correction, and illumination correction; And / or, the preprocessing includes noise removal, and the noise removal includes at least one of mean filtering, median filtering, and low-pass filtering.

4. The method according to claim 2, wherein The obtaining steps of the respective fusion weights of the several second image data include at least one of the following: Based on the attribute information of the target object, determining the salience of imaging of the target object in different bands, and based on the salience of imaging of the target object in different bands and the bands to which the several second image data respectively belong, determining the respective fusion weights of the several second image data; wherein, the salience is positively correlated with the fusion weight; Analyzing the statistical features of the target object in the second image data based on the image data of the target object in the second image data, and performing dimensionality reduction based on the statistical features of the target object in each of the second image data to obtain the respective fusion weights of the several second image data; Predicting the several second image data based on a weight prediction model with reference to the attribute information of the target object to obtain the respective fusion weights of the several second image data; Determine the fusion weights of the several second image data respectively based on at least one of the application environment and task requirements of remote sensing target recognition and the bands to which the several second image data respectively belong; Determine the fusion weights of the several second image data respectively based on at least one of the noise level, image resolution, and spectral response of the several second image data.

5. The method according to claim 1, characterized in that Before the first feature representation is an artificially designed feature and predicting the recognition result of the target object regarding its type based on the first feature representation and the second feature representation, the method further includes: Performing feature extraction on the target image region based on a feature extraction model to obtain a third feature representation of the target object; The predicting the recognition result of the target object regarding its type based on the first feature representation and the second feature representation includes: Performing prediction based on the first feature representation, the second feature representation, and the third feature representation to obtain the recognition result.

6. The method according to claim 1, wherein The determining the irradiation parameters of the laser on the remote sensing platform based on the first feature representation includes at least one of the following: Determining the wavelength in the irradiation parameters based on the feature element representing the material in the first feature representation; Determining the irradiation mode in the irradiation parameters based on the feature elements representing the size and shape in the first feature representation; Determining the energy in the irradiation parameters based on the feature elements representing the material and surface state in the first feature representation.

7. The method according to claim 6, wherein The determining the irradiation mode in the irradiation parameters based on the feature elements representing the size and shape in the first feature representation includes at least one of the following: In response to the feature element representing the size in the first feature representation being higher than the size threshold and the feature element representing the shape being lower than the shape complexity threshold, determining the irradiation mode as one-time irradiation; In response to the feature element representing the size in the first feature representation being lower than the size threshold and the feature element representing the shape being higher than the shape complexity threshold, determining the irradiation mode as scanning irradiation.

8. The method according to claim 1, wherein The analyzing the optical response after the laser irradiates the target object according to the irradiation parameters to obtain a second feature representation characterizing the optical properties of the target object includes at least one of the following: Comparing the optical response with the spectral maps of various materials in the spectral library to obtain a second feature representation characterizing the optical properties of the target object; Analyzing at least one of the peak value, intensity, and width of the optical response to obtain a second feature representation characterizing the optical properties of the target object.

9. The method according to any one of claims 1 to 8, characterized in that, The target object includes at least ships; And / or, the recognition result includes any one of the type label of the target object and the probability values of the target object belonging to different types respectively; And / or, the optical response includes at least one of the reflection spectrum and the fluorescence spectrum.

10. A remote sensing target recognition device, characterized in that, Including: An image acquisition module, configured to acquire a remote sensing image of a target geographical area where a target object to be recognized is located; A target detection module, configured to perform target detection based on the remote sensing image to obtain a target image region of the target object; A first extraction module, configured to extract a first feature representation characterizing the appearance characteristics of the target object based on the target image region; A parameter determination module, configured to determine the irradiation parameters of the laser on the remote sensing platform based on the first feature representation; A second extraction module, configured to analyze the optical response after the laser irradiates the target object according to the irradiation parameters, so as to obtain a second feature representation characterizing the optical characteristics of the target object; A type prediction module, configured to perform prediction based on the first feature representation and the second feature representation, so as to obtain an identification result of the target object regarding its belonging type.

11. An electronic device, characterized in that, At least including a memory and a processor coupled to each other, at least program instructions are stored in the memory, and the processor is configured to execute the program instructions to implement the remote sensing target recognition method according to any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, Program instructions that can be run by a processor are stored, and the program instructions are configured to implement the remote sensing target recognition method according to any one of claims 1 to 9.

Images