Forest land reduction detection method and device

Abstract

The present application discloses a forest reduction detection method and device, and relates to the technical field of forest change detection. The method of the present application includes: obtaining a front-phase remote sensing image and a back-phase remote sensing image corresponding to a target area; inputting the front-phase remote sensing image and the back-phase remote sensing image into a forest reduction detection model to obtain a first forest reduction spot set corresponding to the target area, wherein the forest reduction detection model is obtained by iteratively training a preset deep learning model based on a training sample set in advance; determining a second forest reduction spot set corresponding to the target area according to a preset extraction rule set, the front-phase remote sensing image and the back-phase remote sensing image; determining a third forest reduction spot set corresponding to the target area according to the first forest reduction spot set and the second forest reduction spot set; and determining a forest reduction detection result corresponding to the target area according to the third forest reduction spot set.

Description

Technical Field

[0001] The present application relates to the technical field of forest land change detection, and in particular to a forest land reduction detection method and device. Background Art

[0002] Forest resources are a general term for forests, trees, woodlands, and the wild animals, plants, and microorganisms that depend on them for survival. Among them, woodlands, as an important component of forest resources, are important habitats for a large number of animals and plants, and play an important role in maintaining ecological balance, protecting land and water sources, etc. Therefore, in order to protect forest resources, it is necessary to regularly monitor changes in woodlands.

[0003] Forest land changes include two aspects: forest land increase and forest land decrease. Among them, forest land increase refers to the change from non-forest land to forest land, and forest land decrease refers to the change from forest land to non-forest land. Since forest land decrease may be accompanied by illegal deforestation and illegal construction in regional management business, forest land decrease has received more attention.

[0004] At present, it is usually necessary to iteratively train the deep learning model based on the training samples to obtain the forest change detection model, and then perform forest reduction detection on the target area based on the forest change detection model; however, when the training samples do not cover the scene comprehensively, the omission rate of forest reduction detection on the target area based on the forest change detection model is high, making it impossible to effectively perform forest reduction detection on the target area. Summary of the invention

[0005] The embodiments of the present application provide a method and device for detecting forest land reduction, the main purpose of which is to effectively detect forest land reduction in a target area and reduce the detection omission rate.

[0006] In order to solve the above technical problems, the embodiments of the present application provide the following technical solutions:

[0007] In a first aspect, the present application provides a method for detecting forest reduction, the method comprising:

[0008] Obtaining the front-phase remote sensing image and the back-phase remote sensing image corresponding to the target area;

[0009] Inputting the front-phase remote sensing image and the back-phase remote sensing image into a forest reduction detection model to obtain a first forest reduction patch set corresponding to the target area, wherein the forest reduction detection model is obtained by iteratively training a preset deep learning model based on a training sample set in advance;

[0010] Determine a second forest land reduction patch set corresponding to the target area according to a preset extraction rule set, the front-phase remote sensing image, and the back-phase remote sensing image;

[0011] Determining a third forest land reduction spot set corresponding to the target area according to the first forest land reduction spot set and the second forest land reduction spot set;

[0012] The forest land reduction detection result corresponding to the target area is determined according to the third forest land reduction patch set.

[0013] In a second aspect, the present application further provides a forest reduction detection device, the device comprising:

[0014] An acquisition unit, used for acquiring a front-phase remote sensing image and a back-phase remote sensing image corresponding to a target area;

[0015] An input unit, used for inputting the front-phase remote sensing image and the back-phase remote sensing image into a forest reduction detection model to obtain a first forest reduction patch set corresponding to the target area, wherein the forest reduction detection model is obtained by iteratively training a preset deep learning model based on a training sample set in advance;

[0016] A first determination unit is used to determine a second forest land reduction patch set corresponding to the target area according to a preset extraction rule set, the front-phase remote sensing image and the back-phase remote sensing image;

[0017] A second determining unit is used to determine a third forest land reduction spot set corresponding to the target area according to the first forest land reduction spot set and the second forest land reduction spot set;

[0018] The third determining unit is used to determine the forest land reduction detection result corresponding to the target area according to the third forest land reduction spot set.

[0019] In a third aspect, an embodiment of the present application provides a storage medium, wherein the storage medium includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the forest reduction detection method described in the first aspect.

[0020] In a fourth aspect, an embodiment of the present application provides a forest land reduction detection device, the device comprising a storage medium; and one or more processors, the storage medium is coupled to the processor, the processor is configured to execute program instructions stored in the storage medium; when the program instructions are executed, the forest land reduction detection method described in the first aspect is executed.

[0021] By means of the above technical solution, the technical solution provided by this application has at least the following advantages:

[0022] The present application provides a forest land reduction detection method and device. The present application can, after a forest land reduction detection application obtains a front-phase remote sensing image and a back-phase remote sensing image corresponding to a target area, input the front-phase remote sensing image and the back-phase remote sensing image corresponding to the target area into a pre-trained forest land reduction detection model to obtain a first forest land reduction spot set corresponding to the target area; then, according to a preset extraction rule set, the front-phase remote sensing image and the back-phase remote sensing image, determine a second forest land reduction spot set corresponding to the target area; then, according to the first forest land reduction spot set and the second forest land reduction spot set, determine a third forest land reduction spot set corresponding to the target area; finally, determine the forest land reduction detection result corresponding to the target area according to the third forest land reduction spot set. Since, in the present application, forest land reduction detection is performed on the target area based on the forest land reduction detection model, and then forest land reduction detection is performed on the target area based on a preset extraction rule set, it is possible to describe the objects in the target area in terms of size, shape, color, brightness, texture and the spatial environment in which they are located, and detect forest land reduction patches that cannot be confirmed by the forest land reduction detection model, thereby effectively reducing the omission rate of forest land reduction detection in the target area.

[0023] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present application will become easy to understand. In the accompanying drawings, several embodiments of the present application are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0025] Figure 1 A flow chart of a forest reduction detection method provided in an embodiment of the present application is shown;

[0026] Figure 2 A flow chart of another forest reduction detection method provided in an embodiment of the present application is shown;

[0027] Figure 3 A block diagram of a forest reduction detection device provided in an embodiment of the present application is shown;

[0028] Figure 4 A block diagram of another device for detecting forest reduction provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0029] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0030] In addition, the words “first”, “second” and the like used in the present application do not indicate any order, quantity or importance, but are only used to distinguish different parts.

[0031] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application should have the common meanings understood by technicians in the field to which this application belongs.

[0032] At present, it is usually necessary to iteratively train the deep learning model based on the training samples to obtain the forest change detection model, and then perform forest reduction detection on the target area based on the forest change detection model; however, when the training samples do not cover the scene comprehensively, the omission rate of forest reduction detection on the target area based on the forest change detection model is high.

[0033] Therefore, in order to ensure that the forest reduction detection in the target area can be effectively performed, the embodiment of the present application provides a forest reduction detection method, such as Figure 1 As shown, the method includes:

[0034] 101. Obtain the front-phase remote sensing image and the back-phase remote sensing image corresponding to the target area.

[0035] In the embodiment of the present application, the execution subject in each step is a forest reduction detection application running in a target terminal device, wherein the target terminal device may be, but is not limited to, a computer, a server, a tablet computer, and the like.

[0036] The target area is the area where forest loss detection is required. It should be noted that when a complete remote sensing image corresponding to the target area cannot be obtained from any data source, partial remote sensing images corresponding to the target area can be obtained from multiple data sources, and then the multiple partial remote sensing images corresponding to the target area are spliced ​​to obtain a complete remote sensing image corresponding to the target area, but the present invention is not limited thereto.

[0037] When the staff needs to conduct forest loss detection on the target area, the staff will send a detection request including the first target date interval and the second target date interval to the forest loss detection application. After receiving the detection request, the forest loss detection application will obtain the previous phase remote sensing image corresponding to the target area according to the first target date interval, and obtain the later phase remote sensing image corresponding to the target area according to the second target date interval, wherein the first target date interval is before the second target date interval.

[0038] 102. Input the previous phase remote sensing image and the next phase remote sensing image into the forest land reduction detection model to obtain the first forest land reduction patch set corresponding to the target area.

[0039] Among them, the first forest land reduction patch set corresponding to the target area includes multiple forest land reduction patches; wherein the forest land reduction detection model is obtained by iteratively training a preset deep learning model based on a training sample set in advance.

[0040] After obtaining the previous phase remote sensing images and the next phase remote sensing images corresponding to the target area, the forest land reduction detection application can input the previous phase remote sensing images and the next phase remote sensing images corresponding to the target area into a pre-trained forest land reduction detection model, so that the forest land reduction detection model outputs multiple forest land reduction spots corresponding to the target area to obtain a first forest land reduction spot set corresponding to the target area.

[0041] It should be noted that the specific process of iteratively training the preset deep learning model based on the training sample set in advance can refer to the process of iteratively training the deep learning model established based on any deep learning algorithm in the prior art, and the embodiments of the present application will not go into details about this; when the training stop condition is reached, the trained preset deep learning model is determined as the forest land reduction detection model, wherein the training stop condition may be but is not limited to: the accuracy of the preset deep learning model reaches a preset accuracy threshold, the recall rate of the preset deep learning model reaches a preset recall rate threshold, etc., and the embodiments of the present application do not specifically limit this.

[0042] 103. Determine a second forest land reduction patch set corresponding to the target area based on a preset extraction rule set, a previous phase remote sensing image, and a later phase remote sensing image.

[0043] Among them, the preset extraction rule set is used to extract forest land reduction spots in the previous phase remote sensing images and the next phase remote sensing images corresponding to the target area; forest land reduction detection is performed on the target area based on the preset extraction rule set, and the objects in the target area can be described in terms of size, shape, color, brightness, texture and the spatial environment in which they are located, so that forest land reduction spots that cannot be confirmed by the forest land reduction detection model can be detected, thereby effectively reducing the omission rate of forest land reduction detection in the target area; wherein, the second forest land reduction spot set corresponding to the target area contains multiple forest land reduction spots.

[0044] After obtaining the first forest land reduction patch set corresponding to the target area based on the forest land reduction detection model, the forest land reduction detection application can determine multiple forest land reduction patches corresponding to the target area according to a preset extraction rule set, previous phase remote sensing images and later phase remote sensing images to obtain a second forest land reduction patch set corresponding to the target area.

[0045] It should be noted that, in actual application, step 103 may be performed first, and then step 102; or step 102 and step 103 may be performed in parallel, which is not specifically limited in the embodiment of the present application.

[0046] 104. Determine a third forest land reduction spot set corresponding to the target area based on the first forest land reduction spot set and the second forest land reduction spot set.

[0047] After obtaining the second forest land reduction spot set corresponding to the target area, the forest land reduction detection application can determine the third forest land reduction spot set corresponding to the target area based on the first forest land reduction spot set and the second forest land reduction spot set corresponding to the target area.

[0048] Specifically, in this step, the multiple forest land reduction spots included in the first forest land reduction spot set and the multiple forest land reduction spots included in the second forest land reduction spot set can be summarized to obtain a third forest land reduction spot set corresponding to the target area; or the first forest land reduction spot set and the second forest land reduction spot set can be spatially intersected, and the multiple forest land reduction spots included in the first forest land reduction spot set and the multiple forest land reduction spots included in the second forest land reduction spot set can be divided into multiple forest land reduction spots that intersect spatially and multiple forest land reduction spots that do not intersect spatially, and then the multiple forest land reduction spots that do not intersect spatially can be de-aliased, and finally the multiple forest land reduction spots that intersect spatially and the multiple forest land reduction spots that do not intersect spatially after de-aliasing can be determined as the third forest land reduction spot set corresponding to the target area. The embodiment of the present application does not specifically limit this.

[0049] 105. Determine the forest land reduction detection result corresponding to the target area according to the third forest land reduction patch set.

[0050] After determining the third forest land reduction patch set corresponding to the target area, the forest land reduction detection application can determine the forest land reduction detection result corresponding to the target area based on the third forest land reduction patch set.

[0051] Specifically, in this step, the third forest land reduction spot set corresponding to the target area can be directly determined as the forest land reduction detection result corresponding to the target area; or the multiple forest land reduction spots included in the third forest land reduction spot set corresponding to the target area can be first de-aliased, and the remaining forest land reduction spots in the third forest land reduction spot set can be determined as the forest land reduction detection result corresponding to the target area. The embodiments of the present application do not specifically limit this.

[0052] An embodiment of the present application provides a forest land reduction detection method. According to the embodiment of the present application, after the forest land reduction detection application obtains the previous phase remote sensing image and the next phase remote sensing image corresponding to the target area, the forest land reduction detection application first inputs the previous phase remote sensing image and the next phase remote sensing image corresponding to the target area into a pre-trained forest land reduction detection model to obtain a first forest land reduction spot set corresponding to the target area; then, according to a preset extraction rule set, the previous phase remote sensing image and the next phase remote sensing image, a second forest land reduction spot set corresponding to the target area is determined; then, according to the first forest land reduction spot set and the second forest land reduction spot set, a third forest land reduction spot set corresponding to the target area is determined; finally, according to the third forest land reduction spot set, a forest land reduction detection result corresponding to the target area is determined. Because, in the embodiment of the present application, forest land reduction detection is performed on the target area based on the forest land reduction detection model, and then forest land reduction detection is performed on the target area based on a preset extraction rule set, so that the ground objects in the target area can be described in terms of size, shape, color, brightness, texture and the spatial environment in which they are located, and forest land reduction patches that cannot be confirmed by the forest land reduction detection model can be detected, thereby effectively reducing the omission rate of forest land reduction detection in the target area.

[0053] To explain in more detail below, the present application embodiment provides another method for detecting forest reduction, specifically: Figure 2 As shown, the method includes:

[0054] 201. Obtain a front-phase remote sensing image and a back-phase remote sensing image corresponding to the target area.

[0055] Regarding step 201, obtaining the remote sensing image of the previous phase and the remote sensing image of the subsequent phase corresponding to the target area, reference can be made to Figure 1 The description of the corresponding parts will not be repeated here in the embodiments of the present application.

[0056] 202. Preprocess the front-phase remote sensing images and the back-phase remote sensing images.

[0057] After obtaining the front-phase remote sensing images and the back-phase remote sensing images corresponding to the target area, the forest land reduction detection application also needs to preprocess the front-phase remote sensing images and the back-phase remote sensing images corresponding to the target area. The preprocessing of the front-phase remote sensing images and the back-phase remote sensing images corresponding to the target area may include but is not limited to: image orthorectification and registration processing, image fusion processing, image bit depth reduction processing, image mosaicking and color uniformity processing, cropping according to the boundary of the monitoring area, etc.

[0058] 203. Input the preprocessed front-phase remote sensing image and the preprocessed back-phase remote sensing image into a forest loss detection model to obtain a first forest loss patch set corresponding to the target area.

[0059] After preprocessing the pre-phase remote sensing images and post-phase remote sensing images corresponding to the target area, the forest land loss detection application can input the preprocessed pre-phase remote sensing images and the preprocessed post-phase remote sensing images into the forest land loss detection model, so that the forest land loss detection model outputs multiple forest land loss patches corresponding to the target area to obtain a first forest land loss patch set corresponding to the target area.

[0060] 204. Determine a second forest land reduction patch set corresponding to the target area according to a preset extraction rule set, a preprocessed front-phase remote sensing image, and a preprocessed back-phase remote sensing image.

[0061] After obtaining the first forest land reduction patch set corresponding to the target area based on the forest land reduction detection model, the forest land reduction detection application can determine multiple forest land reduction patches corresponding to the target area according to a preset extraction rule set, preprocessed front-phase remote sensing images, and preprocessed back-phase remote sensing images to obtain a second forest land reduction patch set corresponding to the target area.

[0062] Specifically, in this step, the specific process of determining the second forest land reduction patch set corresponding to the target area according to the preset extraction rule set, the preprocessed front-phase remote sensing image and the preprocessed rear-phase remote sensing image is as follows:

[0063] The preset extraction rule set includes a plurality of feature extraction rules and a plurality of judgment rules.

[0064] (1) performing multi-scale segmentation processing on the front-phase remote sensing image and the back-phase remote sensing image according to a preset large-scale parameter, a preset middle-scale parameter, and a preset small-scale parameter to obtain a large-scale image object layer, a middle-scale image object layer, and a small-scale image object layer, wherein the large-scale image object layer includes a plurality of first objects, the middle-scale image object layer includes a plurality of second objects, and the small-scale image object layer includes a plurality of third objects;

[0065] (2) determining, according to a plurality of feature extraction rules and the previous phase remote sensing image, a plurality of previous phase object features corresponding to each first object, a plurality of previous phase object features corresponding to each second object, and a plurality of previous phase object features corresponding to each third object;

[0066] Among them, for any first object, the multiple previous phase object features corresponding to the first object include: the previous phase improved Lab color space features corresponding to the first object (qsx_L321, qsx_a321, qsx_b321, qsx_L432, qsx_a432, qsx_b432); the previous phase normalized vegetation index (qsx_NDVI) and the previous phase normalized water index (qsx_NDWI) corresponding to the first object; the previous phase edge layer average value (qsx_NDWI) corresponding to the first object x_lee_sigma); the average value of the standard deviation of the three bands of the true color combination of the previous phase corresponding to the first object (qsx_std321), the average value of the standard deviation of the three bands of the false color combination of the previous phase (qsx_std432) and the average value of the standard deviation of the four bands of the previous phase (qsx_std_mean); the narrowness index (p_a_rate) corresponding to the first object; the span in the X direction (x_interval) and the span in the Y direction (y_interval) corresponding to the first object, etc.

[0067] Among them, for any second object, the multiple previous phase object features corresponding to the second object include: the previous phase improved Lab color space features corresponding to the second object (qsx_L321, qsx_a321, qsx_b321, qsx_L432, qsx_a432, qsx_b432); the previous phase normalized vegetation index (qsx_NDVI) and the previous phase normalized water index (qsx_NDWI) corresponding to the second object; the previous phase edge layer average value (qsx_NDWI) corresponding to the second object x_lee_sigma); the average value of the standard deviation of the three bands of the true color combination of the previous phase corresponding to the second object (qsx_std321), the average value of the standard deviation of the three bands of the false color combination of the previous phase (qsx_std432) and the average value of the standard deviation of the four bands of the previous phase (qsx_std_mean); the narrowness index (p_a_rate) corresponding to the second object; the span in the X direction (x_interval) and the span in the Y direction (y_interval) corresponding to the second object, etc.

[0068] Among them, for any third object, the multiple previous phase object features corresponding to the third object include: the previous phase improved Lab color space features corresponding to the third object (qsx_L321, qsx_a321, qsx_b321, qsx_L432, qsx_a432, qsx_b432); the previous phase normalized vegetation index (qsx_NDVI) and the previous phase normalized water index (qsx_NDWI) corresponding to the third object; the previous phase edge layer average value (qsx_NDWI) corresponding to the third object x_lee_sigma); the average value of the standard deviation of the three bands of the true color combination of the previous phase corresponding to the third object (qsx_std321), the average value of the standard deviation of the three bands of the false color combination of the previous phase (qsx_std432) and the average value of the standard deviation of the four bands of the previous phase (qsx_std_mean); the narrowness index (p_a_rate) corresponding to the third object; the span in the X direction (x_interval) and the span in the Y direction (y_interval) corresponding to the third object, etc.

[0069] The specific process of determining multiple previous phase object features corresponding to each first object according to multiple feature extraction rules and previous phase remote sensing images is as follows:

[0070] 1. For any first object, obtain the pixel value corresponding to each pixel contained in the first object in the previous phase remote sensing image, and determine the average value of the pixel values ​​of multiple pixels contained in the first object as the target pixel value corresponding to the first object; determine the NIR (near infrared band) pixel value, R (red light band) pixel value, G (green light band) pixel value, and B (blue light band) pixel value corresponding to the first object according to the target pixel value corresponding to the first object; calculate the L value, a value, and b value of the first object in the Lab color space according to the R pixel value, G pixel value, and B pixel value corresponding to the first object, and add the preset threshold C to the L value, a value, and b value corresponding to the first object, respectively, to obtain the improved L value, improved a value, and improved color value corresponding to the first object. b value, the improved L value corresponding to the first object is determined as qsx_L321, the improved a value corresponding to the first object is determined as qsx_a321, and the improved b value corresponding to the first object is determined as qsx_b321; the L value, a value, and b value of the first object in the Lab color space are calculated according to the NIR pixel value, R pixel value, and G pixel value corresponding to the first object, and the L value, a value, and b value corresponding to the first object are respectively added to the preset threshold C to obtain the improved L value, improved a value, and improved b value corresponding to the first object, the improved L value corresponding to the first object is determined as qsx_L432, the improved a value corresponding to the first object is determined as qsx_L432, and the improved b value corresponding to the first object is determined as qsx_L432.

[0071] 2. For any first object, obtain the pixel value corresponding to each pixel contained in the first object in the previous phase remote sensing image, determine the average pixel value NIR1 of all pixels in the near infrared band, the average pixel value R1 of all pixels in the red band, and the average pixel value G1 of all pixels in the green band according to the pixel value corresponding to each pixel, and use NIR1, R1 and G1 in the preset formula to obtain the previous phase normalized vegetation index (qsx_NDVI) and the previous phase normalized water index (qsx_NDWI) corresponding to the first object, wherein the preset formula is specifically:

[0072] qsx_NDVI=(NIR1-R1) / (NIR1+R1+0.000001)

[0073] qsx_NDWI=(G1-NIR1) / (G1+NIR1+0.000001)

[0074] 3. First, perform edge detection based on the Lee Sigma operator on the near-infrared band (Layer 4) of the previous phase remote sensing image to obtain the previous phase edge layer qsx_lee_sigma_layer; for any first object, calculate the average value of the first object on the previous phase edge layer qsx_lee_sigma_layer to obtain the previous phase edge layer average value (qsx_lee_sigma) corresponding to the first object.

[0075] 4. For any first object, obtain the pixel value corresponding to each pixel contained in the first object in the previous phase remote sensing image, and calculate the standard deviation std_NIR1 of all pixels in the first object in the near infrared band, the standard deviation std_R1 of all pixels in the red band, the standard deviation std_G1 of all pixels in the green band, and the standard deviation std_B1 of all pixels in the first object in the blue band according to the pixel value corresponding to each pixel contained in the first object. R1, the standard deviation std_R1 in the red light band, the standard deviation std_G1 in the green light band, and the standard deviation std_B1 in the blue light band are substituted into the preset formula to obtain the average value (qsx_std321) of the standard deviation of the three bands of the front-phase true color combination, the average value (qsx_std432) of the standard deviation of the three bands of the front-phase false color combination, and the average value (qsx_std_mean) of the standard deviation of the four bands of the front-phase, wherein the preset formula is specifically:

[0076] qsx_std321=(std_R1+std_G1+std_B1) / 3

[0077] qsx_std432=(std_NIR1+std_R1+std_G1) / 3

[0078] qsx_std_mean=(std_NIR1+std_R1+std_G1+std_B1) / 4

[0079] 5. For any first object, calculate the narrowness index (p_a_rate) corresponding to the first object according to the number of pixels contained in the first object, where p_a_rate = P / Area-1, P is the object perimeter expressed by the number of pixels contained in the first object, and Area is the object area expressed by the number of pixels contained in the first object.

[0080] 6. For any object, the span in the X direction (x_interval) and the span in the Y direction (y_interval) corresponding to the first object are calculated according to the maximum value [X Max], the minimum value [X Min], the maximum value [Y MaY] and the minimum value [Y Min] of the first object in the X direction, where x_interval = [X Max] - [X Min], y_interval = [Y Max] - [Y Min].

[0081] It should be noted that the specific process of determining the multiple previous phase object features corresponding to each second object and the multiple previous phase object features corresponding to each third object based on multiple feature extraction rules and previous phase remote sensing images can refer to the above-mentioned specific process of determining the multiple previous phase object features corresponding to each first object based on multiple feature extraction rules and previous phase remote sensing images, and the embodiments of the present application will not elaborate on this.

[0082] (3) Determine a plurality of post-phase object features corresponding to each third object according to a plurality of feature extraction rules and the post-phase remote sensing image.

[0083] Among them, for any third object, the multiple post-phase object features corresponding to the third object include: the post-phase improved Lab color space features corresponding to the third object (hsx_L321, hsx_a321, hsx_b321, hsx_L432, hsx_a432, hsx_b432); the post-phase normalized vegetation index (hsx_NDVI) and the post-phase normalized water index (hsx_NDWI) corresponding to the third object; the post-phase edge layer average value (hsx_lee_sigma) corresponding to the third object, the post-phase normalized water index (hsx_NDWI) corresponding to the third object, The difference between the average value of the edge layer of the phase and the average value of the edge layer of the previous phase (Lee_sigmaD); the average value of the standard deviation of the three bands of the true color combination of the later phase corresponding to the third object (hsx_std321), the average value of the standard deviation of the three bands of the false color combination of the later phase (hsx_std432) and the average value of the standard deviation of the four bands of the later phase (hsx_std_mean); the narrowness index (p_a_rate) corresponding to the third object; the span in the X direction (x_interval) and the span in the Y direction (y_interval) corresponding to the third object, etc.

[0084] It should be noted that the specific process of determining the multiple post-phase object features corresponding to each third object based on multiple feature extraction rules and post-phase remote sensing images can refer to the above-mentioned specific process of determining the multiple pre-phase object features corresponding to each first object based on multiple feature extraction rules and pre-phase remote sensing images, and the embodiments of the present application will not elaborate on this.

[0085] (4) Selecting multiple previous-phase forestland objects from multiple third objects based on multiple judgment rules, multiple previous-phase object features corresponding to each first object, multiple previous-phase object features corresponding to each second object, and multiple previous-phase object features corresponding to each third object.

[0086] Specifically, in this step, according to multiple judgment rules, multiple previous phase object features corresponding to each first object, multiple previous phase object features corresponding to each second object, and multiple previous phase object features corresponding to each third object, the specific process of selecting multiple previous phase forestland objects from multiple third objects is as follows:

[0087] First, according to multiple judgment rules and multiple previous phase object features corresponding to each first object, multiple first forest objects are selected from multiple first objects. For example, the first object that satisfies any one of the conditions (1) qsx_ndvi>0.3and qsx_lee_sigma>1(2) qsx_ndvi>0.3and qsx_lee_sigma>0.4and qsx_L321<180 in the multiple first objects is determined as the first forest object; secondly, according to multiple judgment rules, multiple first forest objects and multiple previous phase object features corresponding to each second object, multiple second forest objects are selected from multiple second objects, that is, multiple second objects contained in each first forest object are first determined as second forest objects, and then according to multiple judgment rules and multiple previous phase object features corresponding to each remaining second object, multiple second forest objects are selected from the remaining multiple second objects. For example, the first object that satisfies any one of the conditions (1) qsx_ndvi>0.3and qsx_lee_sigma>0.4and qsx_std_mean<19and qsx_L432<190(2)qsx_ndvi>0.5and qsx_lee_sigma>0.4andqsx_std_mean<22and qsx_L432<190(3)qsx_ndvi>0.4and qsx_lee_sigma>0.3and qsx_std_mean<22and The second object with qsx_L432<180 is determined as the second forestland object; finally, according to multiple judgment rules, multiple second forestland objects and multiple previous phase object features corresponding to each third object, multiple previous phase forestland objects are selected from multiple third objects, that is, multiple third objects contained in each second forestland object are first determined as third forestland objects, and then according to multiple judgment rules and multiple previous phase object features corresponding to each remaining third object, multiple third forestland objects are selected from the remaining multiple third objects. For example, the third object that meets the conditions (1) qsx_ndvi>0.45and qsx_lee_sigma>1 in the remaining multiple third objects is determined as the third forestland object. Finally, the third forestland objects that meet any one of the conditions (1) qsx_L432>190and qsx_ndvi<0.45(2) are selected.

[0088] qsx_L432>187and qsx_ndvi<0.38(3)qsx_L432>180and qsx_ndvi<0.3(4)

[0089] qsx_L432>180and qsx_lee_sigma<0.9(5)qsx_L432>189and qsx_lee_sigma<1.2(6)qsx_L321>180and qsx_lee_sigma<1(7)qsx_L432<178and qsx_ndvi<0.4(8)qsx_L432<175and qsx_a432<175(9)qsx_L432<175and qsx_lee_sigma<0.35(10)qsx_b432<145(11)qsx_b321<146(12)Length / Width>8(13)qsx_lee_sigma<0.1(14)Common_border_rate(Unclassified)>0.8and qsx_L432>180(15)Area>200Pxl and p_a_rate>-0.3and qsx_a432<180(16)qsx_L432<185and qsx_ndvi<0.4and Exist_super_forest_1=0(17)qsx_L432<185andqsx_ndvi<0.45and The third forestland object with Exist_super_forestland_2=0 is discarded, and the remaining third forestland objects are determined as multiple previous phase forestland objects, where Length / Width is the aspect ratio of the object; Exist_super_forestland_1 indicates that the corresponding object category in the medium-scale image object layer is "forestland"; Exist_super_forestland_2 indicates that the corresponding object category in the large-scale image object layer is "forestland"; Common_border_rate indicates the boundary overlap, which is used to indicate the boundary overlap of two objects, and the calculation formula is Common_border_rate=e i&j / pi, where e i&j represents the length of the overlapping borders of objects i and j, pi represents the perimeter of object i, and common_border_rate(unclassified) represents the degree of overlap between the borders of the current object and the adjacent object whose category is "unclassified".

[0090] (5) According to multiple judgment rules and multiple post-phase object features corresponding to each third object, multiple forest land objects that are changed to non-forest land are selected from multiple front-phase forest land objects, that is, according to multiple post-phase object features corresponding to each front-phase forest land object, the front-phase forest land objects that meet any one of the following conditions among the multiple front-phase forest land objects: (1) qsx_ndvi>0.35and hsx_ndvi<0.12and qsx_L432>170; (2) qsx_ndvi>0.5and hsx_ndvi<0.24and qsx_L432>170 are determined as forest land objects that are changed to non-vegetation, and the adjacent forest land objects that are changed to non-vegetation are merged, and the forest land objects that meet any one of the following conditions among the multiple forest land objects that are changed to non-vegetation are selected. The forest land with hsx_L432<180(4)hsx_L432<195and hsx_std321>19 becomes a non-vegetation object and is discarded; according to the characteristics of multiple later-phase objects corresponding to each previous-phase forest land object, the forest land objects in the previous-phase that meet the conditions (1)qsx_lee_sigma>1and hsx_lee_sigma<0.8and hsx_L432>195and hsx_a432>170andhsx_ndvi>0.3and The forest land objects in the previous phase with hsx_L321<195 are determined as forest land converted to cultivated land or grassland with vegetation objects. The adjacent forest land converted to cultivated land or grassland with vegetation objects are merged, and the forest land converted to cultivated land or grassland with vegetation objects that meet any of the following conditions (1) p_a_rate>-0.4 (2) Area<400 (3) y_interval<8 (4) x_interval<8 are discarded; the remaining multiple forest lands converted to non-vegetated objects and the remaining multiple forest lands converted to cultivated land or grassland with vegetation objects are determined as multiple forest lands converted to non-forest land objects.

[0091] (6) generating a second forest land reduction patch set corresponding to the target area according to multiple forest land-to-non-forest land objects, that is, firstly merging adjacent forest land-to-non-forest land objects among the multiple forest land-to-non-forest land objects; secondly, post-processing the multiple forest land-to-non-forest land objects after merging, wherein the post-processing includes hole filling processing, burr removal processing, etc.; thirdly, outputting all the forest land-to-non-forest land objects after post-processing into vectors to obtain multiple forest land reduction patches; finally, determining the multiple forest land reduction patches as the second forest land reduction patch set corresponding to the target area.

[0092] 205. Determine a third forest land reduction spot set corresponding to the target area based on the first forest land reduction spot set and the second forest land reduction spot set.

[0093] After obtaining the second forest land reduction spot set corresponding to the target area, the forest land reduction detection application can determine the third forest land reduction spot set corresponding to the target area based on the first forest land reduction spot set and the second forest land reduction spot set corresponding to the target area.

[0094] Specifically, in this step, the specific process of determining the third forest land reduction spot set corresponding to the target area according to the first forest land reduction spot set and the second forest land reduction spot set is as follows:

[0095] (1) First, spatial intersection processing is performed on the first forest land reduction spot set and the second forest land reduction spot set, and multiple forest land reduction spots that spatially intersect with the second forest land reduction spot set are selected from the first forest land reduction spot set to obtain multiple first forest land reduction spots, and multiple forest land reduction spots that do not spatially intersect with the first forest land reduction spot set are selected from the second forest land reduction spot set to obtain multiple second forest land reduction spots.

[0096] (2) Secondly, according to multiple anti-counterfeiting rules and multiple attribute values ​​corresponding to each second forest land reduction patch, the multiple second forest land reduction patches are subjected to anti-counterfeiting processing to obtain multiple third forest land reduction patches;

[0097] (3) Finally, a third forest land reduction spot set is generated according to the plurality of first forest land reduction spots and the plurality of third forest land reduction spots, that is, the plurality of first forest land reduction spots and the plurality of third forest land reduction spots are aggregated to obtain the third forest land reduction spot set.

[0098] Furthermore, in the embodiment of the present application, it is necessary to generate multiple anti-counterfeiting rules in advance, wherein the specific process of generating multiple anti-counterfeiting rules is as follows:

[0099] First, a sample patch set and multiple historical forest land reduction patches are obtained, wherein the sample patch set includes multiple real forest land reduction sample patches, and the historical forest land reduction patches are forest land reduction patches to be de-falsed, which are obtained by performing forest land reduction detection on the target area based on a preset extraction rule set.

[0100] Secondly, spatial intersection processing is performed on multiple true forest land reduction sample patches and multiple historical forest land reduction patches, and historical forest land reduction patches that spatially intersect with multiple true forest land reduction sample patches are selected from multiple historical forest land reduction patches to obtain multiple forest land reduction patches that intersect with the true values, and historical forest land reduction patches that do not spatially intersect with multiple true forest land reduction sample patches are selected from multiple historical forest land reduction patches to obtain multiple forest land reduction patches that do not intersect with the true values.

[0101] Finally, multiple anti-false rules are generated based on the preset rules, multiple attribute values ​​corresponding to each forest land reduction patch that intersects the true value, and multiple attribute values ​​corresponding to each forest land reduction patch that does not intersect the true value. That is, for any attribute, multiple forest land reduction patches that intersect with the true value and multiple forest land reduction patches that do not intersect with the true value are sorted according to the attribute value of each forest land reduction patch that intersects with the true value under the attribute and the attribute value of each forest land reduction patch that does not intersect with the true value under the attribute. The sorting results are divided into two cases: (1) The sorting results contain three areas. The first area is all forest land reduction patches that intersect with the true value. The second area contains some forest land reduction patches that intersect with the true value and some forest land reduction patches that do not intersect with the true value. The third area contains (2) The sorting result contains two areas. The first area contains forest land reduction spots that intersect with the true value, and the second area contains forest land reduction spots that do not intersect with the true value. For the first case, the de-false rule corresponding to the attribute is generated based on the forest land reduction spots that do not intersect with the true value at the boundary of the third area. For example, the attribute value of the forest land reduction spots that do not intersect with the true value at the boundary of the third area is a, and the attribute values ​​of the remaining forest land reduction spots that do not intersect with the true value are all greater than a. Then the de-false rule corresponding to the attribute is to remove the forest land reduction spots whose attribute values ​​are greater than or equal to a under this attribute. For the second case, the de-false rule corresponding to the attribute is generated based on the forest land reduction spots that do not intersect with the true value at the boundary of the second area. For example, The attribute value of the forest land reduction patch that does not intersect with the true value at the boundary of the two regions is b, and the attribute values ​​of the other forest land reduction patches that do not intersect with the true value are all less than b, then the de-false rule corresponding to the attribute is to remove the forest land reduction patches whose attribute values ​​are less than or equal to b under the attribute; wherein, for any forest land reduction patch that intersects with the true value, the multiple attribute values ​​corresponding to the forest land reduction patch that intersects with the true value are determined according to the multiple pre-phase object features and multiple post-phase object features of the third object corresponding to the forest land reduction patch that intersects with the true value, and may include but are not limited to: qsx_L321, qsx_L432, qsx_a321, qsx_a432, qsx_b321, qsx_b432, hsx_ L321, hsx_L432, hsx_a321, hsx_a432, hsx_b321, hsx_b432, Area, LengthWidt, x_interval, y_interval, p_a_rate, Compactnes, qsx_leesigma, hsx _leesigma, Lee_sigmaD, hsx_std321, hsx_std432, hsx_std_me, qsx_std321, qsx_std432, qsx_std_me, hsx_ndvi, qsx_ndvi, hsx_ndwi, qsx_ndwi, etc.;

[0102] For any forest land reduction patch that does not intersect with the true value, the multiple attribute values ​​corresponding to the forest land reduction patch that does not intersect with the true value are determined based on the multiple previous phase object features and the multiple subsequent phase object features of the third object corresponding to the forest land reduction patch that does not intersect with the true value, and may include but are not limited to: qsx_L321, qsx_L432, qsx_a321, qsx_a432, qsx_b321, qsx_b432, hsx_L321, hsx_L432, hsx_a321, hsx_a432, hsx_b321, hsx_L432, sx_b432, Area, LengthWidt, x_interval, y_interval, p_a_rate, Compactnes, qsx_leesigma, hsx_leesigma, Lee_sigmaD, hsx_std321, hsx_std432, hsx_std_me, qsx_std321, qsx_std432, qsx_std_me, hsx_ndvi, qsx_ndvi, hsx_ndwi, qsx_ndwi and so on.

[0103] 206. Determine the forest land reduction detection result corresponding to the target area according to the third forest land reduction patch set.

[0104] After determining the third forest land reduction patch set corresponding to the target area, the forest land reduction detection application can determine the forest land reduction detection result corresponding to the target area based on the third forest land reduction patch set.

[0105] Specifically, in this step, the specific process of determining the forest land reduction detection result corresponding to the target area according to the third forest land reduction patch set is as follows:

[0106] (1) First, the auxiliary data corresponding to the target area is obtained, wherein the auxiliary data includes the digital surface model raster data corresponding to the target area, the base vector data of the individual buildings, the mosaic block vector data of the post-phase remote sensing image, and the metadata file of the post-phase remote sensing image, wherein the base vector data of the individual buildings includes the base patch vector of the individual buildings corresponding to each individual building in the target area;

[0107] (2) Secondly, determine the height value corresponding to each individual building based on the digital surface model raster data;

[0108] (3) Again, according to the mosaic block vector data of the post-phase remote sensing image, the satellite azimuth and incidence angle corresponding to the post-phase remote sensing image are searched in the metadata file of the post-phase remote sensing image;

[0109] (4) Then, according to the base spot vector, height value and satellite azimuth and incident angle of each building, the displaced base spot vector of each building is determined. That is, for any building, the height value and satellite azimuth and incident angle of the building are substituted into a preset formula to calculate the base displacement distance and base displacement direction of the building. Then, according to the base displacement distance and base displacement direction of the building, the base spot vector of the building is displaced to obtain the displaced base spot vector of the building. For any building, the displaced base spot vector of the building is aligned with the roof of the building in the post-phase remote sensing image. The preset formula is:

[0110] L = tan(θ) × H,

[0111] κ=ω+180°

[0112] Among them, L is the roof displacement distance corresponding to the single building, θ is the incident angle corresponding to the post-phase remote sensing image, H is the height value corresponding to the single building, κ is the roof displacement direction corresponding to the single building, ω is the satellite azimuth corresponding to the post-phase remote sensing image; among them, when the value of κ is greater than 360°, it is subtracted by 360°.

[0113] It should be noted that if a single building spans two or more mosaic blocks, the principle of area dominance will be used to determine which set of parameters to use for displacement, that is, the mosaic block in which the single building falls has the largest area will be used for displacement using the set of parameters corresponding to that mosaic block; secondly, if the satellite azimuth and incident angle corresponding to the post-phase remote sensing image involve multiple points, the average of the multiple points will be substituted into the preset formula for calculation.

[0114] (5) Finally, the third forest land reduction spot set is de-aliased according to the roof spot vector corresponding to each single building to obtain the forest land reduction detection result corresponding to the target area. That is, for the roof spot vector corresponding to any single building, if the roof spot vector corresponding to the single building falls into a forest land reduction spot included in a third forest land reduction spot set, the ratio of the roof spot vector corresponding to the single building to the forest land reduction spot is calculated. If the ratio is greater than a preset threshold, the forest land reduction spot is removed from the third forest land reduction spot set, thereby avoiding the situation where the roof of the building in the target area blocks the surrounding trees due to the existence of projection difference and is mistakenly detected as forest land reduction, thereby improving the accuracy of forest land reduction detection in the target area.

[0115] Furthermore, as a response to the above Figure 1 and Figure 2 In order to realize the method shown in the figure, another embodiment of the present application also provides a device for detecting forest loss. This device embodiment corresponds to the aforementioned method embodiment. For ease of reading, this device embodiment will not repeat the details of the aforementioned method embodiment one by one, but it should be clear that the device in this embodiment can correspond to all the contents of the aforementioned method embodiment. This device is used to effectively detect forest loss in the target area and reduce the detection omission rate. Specifically, Figure 3 As shown, the device comprises:

[0116] An acquisition unit 31 is used to acquire a front-phase remote sensing image and a back-phase remote sensing image corresponding to a target area;

[0117] An input unit 32 is used to input the front-phase remote sensing image and the back-phase remote sensing image acquired by the acquisition unit 31 into a forest reduction detection model to obtain a first forest reduction patch set corresponding to the target area, wherein the forest reduction detection model is obtained by iteratively training a preset deep learning model based on a training sample set in advance;

[0118] A first determining unit 33 is used to determine a second forest land reduction patch set corresponding to the target area according to a preset extraction rule set, the front-phase remote sensing image and the back-phase remote sensing image acquired by the acquiring unit 31;

[0119] A second determining unit 34 is configured to determine a third forest land reduction spot set corresponding to the target area according to the first forest land reduction spot set obtained by the input unit 32 and the second forest land reduction spot set determined by the first determining unit 33;

[0120] The third determining unit 35 is used to determine the forest land reduction detection result corresponding to the target area according to the third forest land reduction patch set determined by the second determining unit 34 .

[0121] Further, such as Figure 4 As shown, the preset extraction rule set includes multiple feature extraction rules and multiple judgment rules; the first determination unit 33 includes:

[0122] A first processing module 331 is used to perform multi-scale segmentation processing on the front-phase remote sensing image and the back-phase remote sensing image according to a preset large-scale parameter, a preset middle-scale parameter and a preset small-scale parameter to obtain a large-scale image object layer, a middle-scale image object layer and a small-scale image object layer, wherein the large-scale image object layer includes a plurality of first objects, the middle-scale image object layer includes a plurality of second objects, and the small-scale image object layer includes a plurality of third objects;

[0123] A first determination module 332 is used to determine a plurality of previous phase object features corresponding to each of the first objects, a plurality of previous phase object features corresponding to each of the second objects, and a plurality of previous phase object features corresponding to each of the third objects according to a plurality of the feature extraction rules and the previous phase remote sensing image;

[0124] A second determination module 333 is used to determine a plurality of post-phase object features corresponding to each of the third objects according to the plurality of feature extraction rules and the post-phase remote sensing image;

[0125] A first selection module 334 is used to select a plurality of previous phase forestland objects from a plurality of the third objects according to a plurality of the judgment rules, a plurality of previous phase object features corresponding to each of the first objects determined by the first determination module 332, a plurality of previous phase object features corresponding to each of the second objects, and a plurality of previous phase object features corresponding to each of the third objects;

[0126] A second selection module 335 is used to select a plurality of forest land-to-non-forest land objects from the plurality of forest land objects selected by the first selection module 334 according to the plurality of judgment rules and the plurality of post-phase object features corresponding to each of the third objects determined by the second determination module 333;

[0127] The first generating module 336 is used to generate a second forest land reduction patch set corresponding to the target area according to the multiple forest land-to-non-forest land objects selected by the second selecting module 335 .

[0128] Further, such as Figure 4 As shown, the first selection module 334 is specifically used to select multiple first forest objects from multiple first objects according to multiple judgment rules and multiple previous phase object features corresponding to each first object; select multiple second forest objects from multiple second objects according to multiple judgment rules, multiple first forest objects and multiple previous phase object features corresponding to each second object; select multiple previous phase forest objects from multiple third objects according to multiple judgment rules, multiple second forest objects and multiple previous phase object features corresponding to each third object.

[0129] Further, such as Figure 4 As shown, the first generating module 336 is specifically used to merge adjacent forest land to non-forest land objects among the multiple forest land to non-forest land objects; post-process the multiple forest land to non-forest land objects after the merge process; output the multiple forest land to non-forest land objects after the post-processing into vectors to obtain multiple forest land reduction spots; determine the multiple forest land reduction spots as the second forest land reduction spot set corresponding to the target area.

[0130] Further, such as Figure 4 As shown, the second determining unit 34 includes:

[0131] The second processing module 341 is used to perform spatial intersection processing on the first forest land reduction spot set and the second forest land reduction spot set, and select a plurality of forest land reduction spots that spatially intersect with the second forest land reduction spot set from the first forest land reduction spot set to obtain a plurality of first forest land reduction spots, and select a plurality of forest land reduction spots that spatially do not intersect with the first forest land reduction spot set from the second forest land reduction spot set to obtain a plurality of second forest land reduction spots;

[0132] A de-aliasing module 342 is configured to perform de-aliasing processing on a plurality of second forest land reduction spots according to a plurality of de-aliasing rules and a plurality of attribute values ​​corresponding to each of the second forest land reduction spots obtained by the second processing module 341, so as to obtain a plurality of third forest land reduction spots;

[0133] The second generating module 343 is used to generate the third forest land reduction spot set according to the multiple first forest land reduction spots obtained by the second processing module 341 and the multiple third forest land reduction spots obtained by the anti-aliasing module 342 .

[0134] Further, such as Figure 4 As shown, the device also includes:

[0135] The generating unit 36 ​​is used to obtain a sample patch set and a plurality of historical forest land reduction patches, wherein the sample patch set includes a plurality of true forest land reduction sample patches, and the historical forest land reduction patches are forest land reduction patches to be de-falsed, which are obtained by performing forest land reduction detection on the target area based on the preset extraction rule set; spatially intersect the plurality of true forest land reduction sample patches and the plurality of historical forest land reduction patches, and select historical forest land reduction patches that spatially intersect with the plurality of true forest land reduction sample patches from the plurality of historical forest land reduction patches to obtain a plurality of forest land reduction patches that intersect with the true values, and select historical forest land reduction patches that spatially intersect with the plurality of true forest land reduction sample patches from the plurality of historical forest land reduction patches. Historical forest land reduction patches that do not intersect with the true value are obtained to obtain multiple forest land reduction patches that do not intersect with the true value; multiple anti-false rules are generated according to preset rules, multiple attribute values ​​corresponding to each of the forest land reduction patches that intersect with the true value, and multiple attribute values ​​corresponding to each of the forest land reduction patches that do not intersect with the true value, wherein the multiple attribute values ​​corresponding to the forest land reduction patches that intersect with the true value are determined based on multiple previous phase object features and multiple later phase object features of the third object corresponding to the forest land reduction patches that intersect with the true value, and the multiple attribute values ​​corresponding to the forest land reduction patches that do not intersect with the true value are determined based on multiple previous phase object features and multiple later phase object features of the third object corresponding to the forest land reduction patches that do not intersect with the true value.

[0136] Further, such as Figure 4 As shown, the third determining unit 35 is specifically used to obtain auxiliary data corresponding to the target area, wherein the auxiliary data includes digital surface model raster data corresponding to the target area, single building base vector data, post-phase remote sensing image mosaic block vector data and post-phase remote sensing image metadata file, wherein the single building base vector data includes single building base spot vectors corresponding to each single building in the target area; determine the height value corresponding to each single building according to the digital surface model raster data; search the post-phase remote sensing image mosaic block vector data in the metadata file of the post-phase remote sensing image. The displaced single building base spot vector corresponding to each single building is determined according to the single building base spot vector, height value and the satellite azimuth and angle of incidence corresponding to the post-phase remote sensing image, wherein, for any single building, the displaced single building base spot vector corresponding to the single building is aligned with the roof corresponding to the single building in the post-phase remote sensing image; the third forest land reduction spot set is de-aliased according to the displaced single building base spot vector corresponding to each single building to obtain the forest land reduction detection result corresponding to the target area.

[0137] Further, such as Figure 4 As shown, the device also includes:

[0138] The processing unit 37 is used to pre-process the front-phase remote sensing image and the back-phase remote sensing image after the acquisition unit 31 acquires the front-phase remote sensing image and the back-phase remote sensing image corresponding to the target area;

[0139] The input unit 32 is specifically used to input the pre-processed remote sensing image of the previous phase and the pre-processed remote sensing image of the later phase into the forest loss detection model to obtain a first forest loss patch set corresponding to the target area;

[0140] The first determination unit 33 is specifically configured to determine a second forest land reduction patch set corresponding to the target area according to a preset extraction rule set, the pre-processed remote sensing image of the previous phase, and the pre-processed remote sensing image of the next phase.

[0141] The embodiments of the present application provide a forest land reduction detection method and device. According to the embodiments of the present application, after the forest land reduction detection application obtains the previous phase remote sensing image and the next phase remote sensing image corresponding to the target area, the forest land reduction detection application first inputs the previous phase remote sensing image and the next phase remote sensing image corresponding to the target area into a pre-trained forest land reduction detection model to obtain a first forest land reduction spot set corresponding to the target area; then, according to a preset extraction rule set, the previous phase remote sensing image and the next phase remote sensing image, a second forest land reduction spot set corresponding to the target area is determined; then, according to the first forest land reduction spot set and the second forest land reduction spot set, a third forest land reduction spot set corresponding to the target area is determined; finally, according to the third forest land reduction spot set, a forest land reduction detection result corresponding to the target area is determined. Because, in the embodiment of the present application, forest land reduction detection is performed on the target area based on the forest land reduction detection model, and then forest land reduction detection is performed on the target area based on a preset extraction rule set, so that the ground objects in the target area can be described in terms of size, shape, color, brightness, texture and the spatial environment in which they are located, and forest land reduction patches that cannot be confirmed by the forest land reduction detection model can be detected, thereby effectively reducing the omission rate of forest land reduction detection in the target area.

[0142] An embodiment of the present application provides a storage medium, which includes a stored program, wherein when the program is executed, the device where the storage medium is located is controlled to execute the above-mentioned forest reduction detection method.

[0143] The storage medium may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM, and the memory includes at least one storage chip.

[0144] An embodiment of the present application also provides a forest reduction detection device, which includes a storage medium; and one or more processors, wherein the storage medium is coupled to the processor, and the processor is configured to execute program instructions stored in the storage medium; when the program instructions are executed, the forest reduction detection method described above is executed.

[0145] The embodiment of the present application provides a device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, the following steps are implemented:

[0146] Obtaining the front-phase remote sensing image and the back-phase remote sensing image corresponding to the target area;

[0147] Inputting the front-phase remote sensing image and the back-phase remote sensing image into a forest reduction detection model to obtain a first forest reduction patch set corresponding to the target area, wherein the forest reduction detection model is obtained by iteratively training a preset deep learning model based on a training sample set in advance;

[0148] Determine a second forest land reduction patch set corresponding to the target area according to a preset extraction rule set, the front-phase remote sensing image, and the back-phase remote sensing image;

[0149] Determining a third forest land reduction spot set corresponding to the target area according to the first forest land reduction spot set and the second forest land reduction spot set;

[0150] The forest land reduction detection result corresponding to the target area is determined according to the third forest land reduction patch set.

[0151] Furthermore, the preset extraction rule set includes a plurality of feature extraction rules and a plurality of judgment rules; and determining the second forest land reduction patch set corresponding to the target area according to the preset extraction rule set, the previous phase remote sensing image and the next phase remote sensing image includes:

[0152] Perform multi-scale segmentation processing on the front-phase remote sensing image and the back-phase remote sensing image according to a preset large-scale parameter, a preset middle-scale parameter and a preset small-scale parameter to obtain a large-scale image object layer, a middle-scale image object layer and a small-scale image object layer, wherein the large-scale image object layer includes a plurality of first objects, the middle-scale image object layer includes a plurality of second objects, and the small-scale image object layer includes a plurality of third objects;

[0153] Determine, according to the plurality of feature extraction rules and the front-phase remote sensing image, a plurality of front-phase object features corresponding to each of the first objects, a plurality of front-phase object features corresponding to each of the second objects, and a plurality of front-phase object features corresponding to each of the third objects;

[0154] Determine a plurality of post-phase object features corresponding to each of the third objects according to the plurality of feature extraction rules and the post-phase remote sensing image;

[0155] Selecting a plurality of previous phase forestland objects from a plurality of the third objects according to a plurality of the judgment rules, a plurality of previous phase object features corresponding to each of the first objects, a plurality of previous phase object features corresponding to each of the second objects, and a plurality of previous phase object features corresponding to each of the third objects;

[0156] According to the plurality of judgment rules and the plurality of post-phase object features corresponding to each of the third objects, a plurality of forest land objects are selected from the plurality of pre-phase forest land objects to be converted into non-forest land objects;

[0157] A second forest land reduction patch set corresponding to the target area is generated according to the plurality of forest land-to-non-forest land objects.

[0158] Further, selecting a plurality of previous phase forestland objects from a plurality of the third objects according to a plurality of the judgment rules, a plurality of previous phase object features corresponding to each of the first objects, a plurality of previous phase object features corresponding to each of the second objects, and a plurality of previous phase object features corresponding to each of the third objects, comprises:

[0159] Selecting a plurality of first forestland objects from a plurality of first objects according to a plurality of the judgment rules and a plurality of previous phase object features corresponding to each of the first objects;

[0160] Selecting a plurality of second forestland objects from the plurality of second objects according to the plurality of judgment rules, the plurality of the first forestland objects and the plurality of previous phase object features corresponding to each of the second objects;

[0161] According to the plurality of judgment rules, the plurality of the second forestland objects and the plurality of previous phase object features corresponding to each of the third objects, a plurality of the previous phase forestland objects are selected from the plurality of the third objects.

[0162] Furthermore, generating a second forest land reduction patch set corresponding to the target area according to the plurality of forest land-to-non-forest land objects includes:

[0163] Merging adjacent forest land-to-non-forest land objects among the plurality of forest land-to-non-forest land objects;

[0164] Post-processing the multiple forest lands that have been merged and converted into non-forest land objects;

[0165] Converting the plurality of forest lands after post-processing into non-forest land objects and outputting them into vectors to obtain a plurality of forest land reduction patches;

[0166] The plurality of forest land reduction spots are determined as a second forest land reduction spot set corresponding to the target area.

[0167] Furthermore, determining a third forest land reduction spot set corresponding to the target area according to the first forest land reduction spot set and the second forest land reduction spot set includes:

[0168] Performing spatial intersection processing on the first forest land reduction spot set and the second forest land reduction spot set, and selecting a plurality of forest land reduction spots that spatially intersect with the second forest land reduction spot set from the first forest land reduction spot set to obtain a plurality of first forest land reduction spots, and selecting a plurality of forest land reduction spots that spatially do not intersect with the first forest land reduction spot set from the second forest land reduction spot set to obtain a plurality of second forest land reduction spots;

[0169] According to multiple anti-counterfeiting rules and multiple attribute values ​​corresponding to each of the second forest land reduction spots, a plurality of the second forest land reduction spots are subjected to anti-counterfeiting processing to obtain a plurality of third forest land reduction spots;

[0170] The third forest land reduction spot set is generated according to a plurality of the first forest land reduction spots and a plurality of the third forest land reduction spots.

[0171] Furthermore, the method further comprises:

[0172] Acquire a sample patch set and a plurality of historical forestland reduction patches, wherein the sample patch set includes a plurality of real forestland reduction sample patches, and the historical forestland reduction patches are forestland reduction patches to be de-falsified, which are obtained by previously performing forestland reduction detection on the target area based on the preset extraction rule set;

[0173] Performing spatial intersection processing on the multiple true forestland reduction sample spots and the multiple historical forestland reduction spots, and selecting historical forestland reduction spots that spatially intersect with the multiple true forestland reduction sample spots from the multiple historical forestland reduction spots to obtain multiple forestland reduction spots that intersect with the true values, and selecting historical forestland reduction spots that spatially do not intersect with the multiple true forestland reduction sample spots from the multiple historical forestland reduction spots to obtain multiple forestland reduction spots that do not intersect with the true values;

[0174] According to preset rules, multiple attribute values ​​corresponding to each forest land reduction patch that intersects with the true value and multiple attribute values ​​corresponding to each forest land reduction patch that does not intersect with the true value, multiple de-false rules are generated, wherein the multiple attribute values ​​corresponding to the forest land reduction patch that intersects with the true value are determined based on multiple previous phase object features and multiple later phase object features of the third object corresponding to the forest land reduction patch that intersects with the true value, and the multiple attribute values ​​corresponding to the forest land reduction patch that does not intersect with the true value are determined based on multiple previous phase object features and multiple later phase object features of the third object corresponding to the forest land reduction patch that does not intersect with the true value.

[0175] Furthermore, determining the forest land reduction detection result corresponding to the target area according to the third forest land reduction patch set includes:

[0176] Acquire auxiliary data corresponding to the target area, wherein the auxiliary data includes digital surface model raster data corresponding to the target area, single building base vector data, post-phase remote sensing image mosaic block vector data and post-phase remote sensing image metadata file, wherein the single building base vector data includes single building base spot vectors corresponding to each single building in the target area;

[0177] Determine the height value corresponding to each of the individual buildings according to the digital surface model raster data;

[0178] Searching for the satellite azimuth and the incident angle corresponding to the post-phase remote sensing image in the metadata file of the post-phase remote sensing image according to the post-phase remote sensing image mosaic block vector data;

[0179] Determine the displaced single building base spot vector corresponding to each single building according to the single building base spot vector, height value and the satellite azimuth and incident angle corresponding to the post-phase remote sensing image corresponding to each single building, wherein, for any single building, the displaced single building base spot vector corresponding to the single building is aligned with the roof corresponding to the single building in the post-phase remote sensing image;

[0180] The third forest land reduction patch set is subjected to de-aliasing processing according to the displaced single building base patch vector corresponding to each of the single buildings, so as to obtain the forest land reduction detection result corresponding to the target area.

[0181] Furthermore, after obtaining the front-phase remote sensing image and the back-phase remote sensing image corresponding to the target area, the method further includes:

[0182] Preprocessing the front-phase remote sensing image and the back-phase remote sensing image;

[0183] The step of inputting the front-phase remote sensing image and the back-phase remote sensing image into a forest reduction detection model to obtain a first forest reduction patch set corresponding to the target area includes:

[0184] Inputting the pre-processed remote sensing image of the previous phase and the pre-processed remote sensing image of the later phase into a forest land reduction detection model to obtain a first forest land reduction patch set corresponding to the target area;

[0185] The step of determining the second forest land reduction patch set corresponding to the target area according to the preset extraction rule set, the previous phase remote sensing image, and the next phase remote sensing image comprises:

[0186] A second forest land reduction patch set corresponding to the target area is determined according to a preset extraction rule set, the pre-processed remote sensing image of the previous phase, and the pre-processed remote sensing image of the subsequent phase.

[0187] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program code that initializes the following method steps: obtaining a previous phase remote sensing image and a later phase remote sensing image corresponding to a target area; inputting the previous phase remote sensing image and the later phase remote sensing image into a forest reduction detection model to obtain a first forest reduction spot set corresponding to the target area, wherein the forest reduction detection model is obtained by iteratively training a preset deep learning model based on a training sample set in advance; determining a second forest reduction spot set corresponding to the target area according to a preset extraction rule set, the previous phase remote sensing image and the later phase remote sensing image; determining a third forest reduction spot set corresponding to the target area according to the first forest reduction spot set and the second forest reduction spot set; and determining a forest reduction detection result corresponding to the target area according to the third forest reduction spot set.

[0188] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0189] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0190] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0191] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0192] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0193] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0194] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0195] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0196] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0197] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for detecting forest land reduction, characterized in that: The method comprises: Obtaining the front-phase remote sensing image and the back-phase remote sensing image corresponding to the target area; Inputting the front-phase remote sensing image and the back-phase remote sensing image into a forest reduction detection model to obtain a first forest reduction patch set corresponding to the target area, wherein the forest reduction detection model is obtained by iteratively training a preset deep learning model based on a training sample set in advance; Determine a second forest land reduction patch set corresponding to the target area according to a preset extraction rule set, the front-phase remote sensing image, and the back-phase remote sensing image; Determining a third forest land reduction spot set corresponding to the target area according to the first forest land reduction spot set and the second forest land reduction spot set; Determine the forest land reduction detection result corresponding to the target area according to the third forest land reduction patch set; The preset extraction rule set includes a plurality of feature extraction rules and a plurality of judgment rules; determining the second forest land reduction patch set corresponding to the target area according to the preset extraction rule set, the previous phase remote sensing image and the next phase remote sensing image, includes: Perform multi-scale segmentation processing on the front-phase remote sensing image and the back-phase remote sensing image according to a preset large-scale parameter, a preset middle-scale parameter and a preset small-scale parameter to obtain a large-scale image object layer, a middle-scale image object layer and a small-scale image object layer, wherein the large-scale image object layer includes a plurality of first objects, the middle-scale image object layer includes a plurality of second objects, and the small-scale image object layer includes a plurality of third objects; Determine a plurality of previous phase object features corresponding to each of the first objects, a plurality of previous phase object features corresponding to each of the second objects, and a plurality of previous phase object features corresponding to each of the third objects according to a plurality of feature extraction rules and the previous phase remote sensing images; determine a plurality of later phase object features corresponding to each of the third objects according to a plurality of feature extraction rules and the later phase remote sensing images; wherein the plurality of feature extraction rules are rules for extracting improved Lab color space features, normalized vegetation index, normalized water index, edge layer average, average of three band standard deviations of true color combination, average of three band standard deviations of false color combination, average of four band standard deviations, narrowness index, span in the X direction, and span in the Y direction from remote sensing images; According to a plurality of the judgment rules, a plurality of previous-phase object features corresponding to each of the first objects, a plurality of previous-phase object features corresponding to each of the second objects, and a plurality of previous-phase object features corresponding to each of the third objects, a plurality of previous-phase forestland objects are selected from the plurality of the third objects; according to a plurality of the judgment rules and a plurality of later-phase object features corresponding to each of the third objects, a plurality of forestland-to-non-forestland objects are selected from the plurality of previous-phase forestland objects; wherein the plurality of the judgment rules are rules for selecting a plurality of previous-phase forestland objects from the plurality of the third objects according to a plurality of previous-phase object features corresponding to each of the first objects, a plurality of previous-phase object features corresponding to each of the second objects, and a plurality of previous-phase object features corresponding to each of the third objects, and for selecting a plurality of forestland-to-non-forestland objects from the plurality of previous-phase forestland objects according to a plurality of later-phase object features corresponding to each of the third objects; A second forest land reduction patch set corresponding to the target area is generated according to the plurality of forest land-to-non-forest land objects.

2. The method according to claim 1, characterized in that The selecting a plurality of previous phase forestland objects from the plurality of third objects according to the plurality of judgment rules, the plurality of previous phase object features corresponding to each of the first objects, the plurality of previous phase object features corresponding to each of the second objects, and the plurality of previous phase object features corresponding to each of the third objects comprises: Selecting a plurality of first forestland objects from a plurality of first objects according to a plurality of the judgment rules and a plurality of previous phase object features corresponding to each of the first objects; Selecting a plurality of second forestland objects from the plurality of second objects according to the plurality of judgment rules, the plurality of the first forestland objects and the plurality of previous phase object features corresponding to each of the second objects; According to the plurality of judgment rules, the plurality of the second forestland objects and the plurality of previous phase object features corresponding to each of the third objects, a plurality of the previous phase forestland objects are selected from the plurality of the third objects.

3. The method according to claim 1, characterized in that: The step of generating a second forest land reduction patch set corresponding to the target area according to the plurality of forest land-to-non-forest land objects includes: Merging adjacent forest land-to-non-forest land objects among the plurality of forest land-to-non-forest land objects; Post-processing the multiple forest lands that have been merged and converted into non-forest land objects; Converting the plurality of forest lands after post-processing into non-forest land objects and outputting them into vectors to obtain a plurality of forest land reduction patches; The plurality of forest land reduction spots are determined as a second forest land reduction spot set corresponding to the target area.

4. The method according to claim 1, characterized in that The determining, according to the first forest land reduction spot set and the second forest land reduction spot set, a third forest land reduction spot set corresponding to the target area includes: Performing spatial intersection processing on the first forest land reduction spot set and the second forest land reduction spot set, and selecting a plurality of forest land reduction spots that spatially intersect with the second forest land reduction spot set from the first forest land reduction spot set to obtain a plurality of first forest land reduction spots, and selecting a plurality of forest land reduction spots that spatially do not intersect with the first forest land reduction spot set from the second forest land reduction spot set to obtain a plurality of second forest land reduction spots; According to multiple anti-counterfeiting rules and multiple attribute values ​​corresponding to each of the second forest land reduction spots, a plurality of the second forest land reduction spots are subjected to anti-counterfeiting processing to obtain a plurality of third forest land reduction spots; The third forest land reduction spot set is generated according to a plurality of the first forest land reduction spots and a plurality of the third forest land reduction spots.

5. The method according to claim 4, characterized in that The method further comprises: Acquire a sample patch set and a plurality of historical forestland reduction patches, wherein the sample patch set includes a plurality of real forestland reduction sample patches, and the historical forestland reduction patches are forestland reduction patches to be de-falsified, which are obtained by previously performing forestland reduction detection on the target area based on the preset extraction rule set; Performing spatial intersection processing on the multiple true forestland reduction sample spots and the multiple historical forestland reduction spots, and selecting historical forestland reduction spots that spatially intersect with the multiple true forestland reduction sample spots from the multiple historical forestland reduction spots to obtain multiple forestland reduction spots that intersect with the true values, and selecting historical forestland reduction spots that spatially do not intersect with the multiple true forestland reduction sample spots from the multiple historical forestland reduction spots to obtain multiple forestland reduction spots that do not intersect with the true values; According to preset rules, multiple attribute values ​​corresponding to each forest land reduction patch that intersects with the true value and multiple attribute values ​​corresponding to each forest land reduction patch that does not intersect with the true value, multiple de-false rules are generated, wherein the multiple attribute values ​​corresponding to the forest land reduction patch that intersects with the true value are determined based on multiple previous phase object features and multiple later phase object features of the third object corresponding to the forest land reduction patch that intersects with the true value, and the multiple attribute values ​​corresponding to the forest land reduction patch that does not intersect with the true value are determined based on multiple previous phase object features and multiple later phase object features of the third object corresponding to the forest land reduction patch that does not intersect with the true value.

6. The method according to claim 1, characterized in that The determining the forest land reduction detection result corresponding to the target area according to the third forest land reduction spot set includes: Acquire auxiliary data corresponding to the target area, wherein the auxiliary data includes digital surface model raster data corresponding to the target area, single building base vector data, post-phase remote sensing image mosaic block vector data and post-phase remote sensing image metadata file, wherein the single building base vector data includes single building base spot vectors corresponding to each single building in the target area; Determine the height value corresponding to each of the individual buildings according to the digital surface model raster data; According to the mosaic block vector data of the post-phase remote sensing image, searching the metadata file of the post-phase remote sensing image for the satellite azimuth and the incident angle corresponding to the post-phase remote sensing image; Determine the displaced single building base spot vector corresponding to each single building according to the single building base spot vector, height value and the satellite azimuth and incident angle corresponding to the post-phase remote sensing image corresponding to each single building, wherein, for any single building, the displaced single building base spot vector corresponding to the single building is aligned with the roof corresponding to the single building in the post-phase remote sensing image; The third forest land reduction patch set is subjected to de-aliasing processing according to the displaced single building base patch vector corresponding to each of the single buildings, so as to obtain the forest land reduction detection result corresponding to the target area.

7. The method according to any one of claims 1 to 6, characterized in that: After acquiring the front-phase remote sensing image and the back-phase remote sensing image corresponding to the target area, the method further includes: Preprocessing the front-phase remote sensing image and the back-phase remote sensing image; The step of inputting the front-phase remote sensing image and the back-phase remote sensing image into a forest reduction detection model to obtain a first forest reduction patch set corresponding to the target area includes: Inputting the pre-processed remote sensing image of the previous phase and the pre-processed remote sensing image of the later phase into a forest reduction detection model to obtain a first forest reduction patch set corresponding to the target area; The step of determining the second forest land reduction patch set corresponding to the target area according to the preset extraction rule set, the previous phase remote sensing image, and the next phase remote sensing image comprises: A second forest land reduction patch set corresponding to the target area is determined according to a preset extraction rule set, the pre-processed remote sensing image of the previous phase, and the pre-processed remote sensing image of the subsequent phase.

8. A device for detecting forest land reduction, characterized in that: The device comprises: An acquisition unit, used for acquiring a front-phase remote sensing image and a back-phase remote sensing image corresponding to a target area; An input unit, used for inputting the front-phase remote sensing image and the back-phase remote sensing image into a forest reduction detection model to obtain a first forest reduction patch set corresponding to the target area, wherein the forest reduction detection model is obtained by iteratively training a preset deep learning model based on a training sample set in advance; A first determination unit is used to determine a second forest land reduction patch set corresponding to the target area according to a preset extraction rule set, the front-phase remote sensing image and the back-phase remote sensing image; A second determining unit is used to determine a third forest land reduction spot set corresponding to the target area according to the first forest land reduction spot set and the second forest land reduction spot set; A third determining unit, configured to determine a forest land reduction detection result corresponding to the target area according to the third forest land reduction patch set; The preset extraction rule set includes a plurality of feature extraction rules and a plurality of judgment rules; the first determination unit includes: A first processing module is used to perform multi-scale segmentation processing on the front-phase remote sensing image and the back-phase remote sensing image according to a preset large-scale parameter, a preset middle-scale parameter and a preset small-scale parameter to obtain a large-scale image object layer, a middle-scale image object layer and a small-scale image object layer, wherein the large-scale image object layer includes a plurality of first objects, the middle-scale image object layer includes a plurality of second objects, and the small-scale image object layer includes a plurality of third objects; A first determination module is used to determine a plurality of previous phase object features corresponding to each of the first objects, a plurality of previous phase object features corresponding to each of the second objects, and a plurality of previous phase object features corresponding to each of the third objects according to a plurality of the feature extraction rules and the previous phase remote sensing image; A second determination module is used to determine a plurality of post-phase object features corresponding to each of the third objects according to the plurality of feature extraction rules and the post-phase remote sensing image; wherein the plurality of feature extraction rules are rules for extracting improved Lab color space features, normalized vegetation index, normalized water index, edge layer average, average of three band standard deviations of true color combination, average of three band standard deviations of false color combination, average of four band standard deviations, narrowness index, span in the X direction, and span in the Y direction in the remote sensing image; A first selection module is used to select a plurality of previous phase forestland objects from a plurality of the third objects according to a plurality of the judgment rules, a plurality of previous phase object features corresponding to each of the first objects, a plurality of previous phase object features corresponding to each of the second objects, and a plurality of previous phase object features corresponding to each of the third objects; A second selection module is used to select a plurality of forest land objects that are changed into non-forest land objects from the plurality of forest land objects in the previous phase according to the plurality of judgment rules and the plurality of post-phase object features corresponding to each of the third objects; wherein the plurality of judgment rules are rules for selecting a plurality of forest land objects in the previous phase according to the plurality of pre-phase object features corresponding to each of the first objects, the plurality of pre-phase object features corresponding to each of the second objects, and the plurality of pre-phase object features corresponding to each of the third objects, and for selecting a plurality of forest land objects that are changed into non-forest land objects from the plurality of forest land objects in the previous phase according to the plurality of post-phase object features corresponding to each of the third objects; The first generating module is used to generate a second forest land reduction patch set corresponding to the target area according to a plurality of forest land-to-non-forest land objects.

9. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is executed, the device where the storage medium is located is controlled to execute the forest reduction detection method according to any one of claims 1 to 7.

10. A device for detecting forest land reduction, characterized in that: The device includes a storage medium; and one or more processors, wherein the storage medium is coupled to the processor, and the processor is configured to execute program instructions stored in the storage medium; when the program instructions are executed, the forest reduction detection method described in any one of claims 1 to 7 is executed.

Images