An automatic selection method for tree canopy samples in high spatial resolution remote sensing images

Abstract

This invention discloses an automatic canopy sample selection method in high spatial resolution remote sensing imagery. The method involves calculating the Normalized Difference Vegetation Index (NDVI) based on digitally quantized data; traversing each pixel of the NDVI and constructing an inner and outer window for each traversed pixel; calculating the NDVI variation coefficients of the pixels contained in the inner and outer windows; determining whether the currently traversed pixel is a canopy pixel; and determining whether a canopy pixel is a tree canopy pixel. This invention improves the efficiency of canopy information recognition, overcomes the problem of poor canopy sample quality caused by terrain shadows and canopy overlap in high spatial resolution remote sensing imagery, and avoids the influence of terrain shadows and canopy overlap on sample selection. This invention has good generalization ability and can accurately identify tree canopies at the single-tree scale for different scenarios, thus improving the quality of canopy samples.

Description

Technical Field

[0001] This invention belongs to the field of satellite remote sensing, specifically relating to a method for automatically selecting tree canopy samples in high spatial resolution remote sensing images. It is applicable to the automatic construction of single-tree-scale tree canopy samples in deep learning of high spatial resolution remote sensing images. Background Technology

[0002] The canopy is a crucial component of trees, serving as the carrier for photosynthesis. Studying the canopy allows for timely monitoring of tree growth and biomass. The advent of high spatial resolution satellite data has provided a new data source for acquiring canopy structure information over large areas. Furthermore, the application of deep learning technology in satellite remote sensing has made canopy information acquisition faster and more accurate. Therefore, using deep learning technology to identify canopy information in high spatial resolution remote sensing imagery is an important application area of ​​satellite remote sensing.

[0003] The accuracy of the canopy distribution information output by deep learning models depends heavily on the quantity and quality of canopy samples. Therefore, deep learning-based canopy identification requires a large number of high-quality samples as input. Traditional canopy sample databases are often constructed through manual annotation. However, manual annotation of canopy samples is often inefficient, resulting in insufficient input samples for the model. Furthermore, the inconsistencies in manual annotation—specifically, the lack of standardized visual interpretation of canopies—lead to inconsistent sample quality, introducing uncertainty into the model's output.

[0004] High spatial resolution remote sensing images possess rich texture information. Based on fully utilizing this information, a series of image processing methods have been developed to quickly obtain a large number of effective tree canopy samples. These methods are based on image segmentation principles and mainly include watershed segmentation, global grayscale thresholding, and gradient segmentation. The fundamental premise of these methods is that the tree canopy has good distinguishability from the background. After obtaining the segmentation threshold, ground features within the scene are segmented to obtain tree canopy information.

[0005] However, due to the complexity of the imaging process in remote sensing images, the undulating terrain within the scene causes tree canopies to overlap with shadows, making it difficult to distinguish tree canopies from the background. Furthermore, in densely wooded areas, the overlap of multiple tree canopies can cause multiple canopies to be fragmented into a single canopy, reducing the accuracy of canopy samples. Therefore, there is an urgent need for a method to automatically select tree canopy samples from high spatial resolution remote sensing images without relying on manual annotation. This method needs to address the impact of canopy overlap and terrain shadows on canopy recognition in remote sensing images, automatically obtain a large number of high-precision canopy samples, better construct a sample library for deep learning, and improve the accuracy of satellite remote sensing in identifying tree canopies. Summary of the Invention

[0006] This invention addresses the aforementioned technical problems in constructing deep learning sample libraries in existing technologies by proposing an automatic tree canopy sample selection method from high spatial resolution remote sensing images. This method can automatically select tree canopy samples from high spatial resolution remote sensing images, thereby better constructing a sample library for deep learning and improving the accuracy of satellite remote sensing in identifying tree canopies.

[0007] The above-mentioned technical problems of the present invention are mainly solved by the following technical solutions:

[0008] An automatic selection method for tree canopy samples in high spatial resolution remote sensing images includes the following steps:

[0009] Step 1: Calculate the normalized vegetation index based on the numerical quantified data;

[0010] Step 2: Traverse each cell of the normalized vegetation index and construct an inner window and an outer window for the traversed cell. The outer window is larger than the inner window, and the traversed cell is located at the center of the inner and outer windows.

[0011] Step 3: Calculate the normalized vegetation index variation coefficient for the pixels contained in the inner and outer windows;

[0012] Step 4: Determine whether the currently traversed pixel is a vegetation canopy pixel based on the normalized vegetation index variation coefficient of the pixels contained in the inner and outer windows.

[0013] Step 5: For the identified vegetation canopy pixels, if the normalized vegetation index corresponding to the vegetation canopy pixel is greater than or equal to the set threshold, then the vegetation canopy pixel is a tree canopy pixel.

[0014] As mentioned above, the normalized vegetation index Based on the following formula:

[0015]

[0016] in, and These are digital quantization values ​​for the near-infrared and red bands, respectively.

[0017] The normalized vegetation index variation coefficients of the pixels contained in the inner and outer windows, as described above, are based on the following formula:

[0018]

[0019] in, and These are the normalized vegetation index variation coefficients for the pixels contained in the inner and outer windows, respectively. and These are the normalized standard deviations of the vegetation index for the pixels contained in the inner and outer windows, respectively. and These are the average normalized vegetation indices of the pixels contained in the inner and outer windows, respectively.

[0020] As described above, determining whether the currently traversed cell is a vegetation canopy cell includes the following steps:

[0021] like If so, then the currently traversed pixels are vegetation canopy pixels;

[0022] like If so, then the currently traversed pixels are non-vegetation canopy pixels.

[0023] As mentioned above, the threshold value set in step 5 is 0.15.

[0024] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the above-described method for automatically selecting tree canopy samples in high spatial resolution remote sensing images.

[0025] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method for automatically selecting tree canopy samples from high spatial resolution remote sensing images.

[0026] A computer program product includes a computer program that, when executed by a processor, implements the steps of the above-described method for automatically selecting tree canopy samples from high spatial resolution remote sensing images.

[0027] Compared with the prior art, the present invention has the following advantages:

[0028] 1. Automatically identify tree canopy information in high spatial resolution remote sensing images. Compared with the method of manually annotating samples, it can meet the sample size requirements of deep learning models and improve the efficiency of tree canopy information identification.

[0029] 2. This invention overcomes the problem of poor tree canopy sample quality caused by terrain shadows and canopy overlap in high spatial resolution remote sensing images, avoiding the impact of terrain shadows and canopy overlap on sample selection. This invention has good generalization ability and can accurately identify tree canopies at the single-tree scale for different scenarios, improving the quality of tree canopy samples. Attached Figure Description

[0030] Figure 1 This is a schematic diagram showing the setting of inner and outer windows per pixel;

[0031] Figure 2(a) Digital quantization data of GF-2 image;

[0032] Figure 2(b) Calculation results of normalized vegetation index;

[0033] Figure 2(c) Automatic recognition results of the present invention (the red circle indicates the recognized tree canopy pixels);

[0034] Figure 2(d) Tree canopy pixels interpreted by human visual interpretation (the red circle indicates the identified tree canopy pixels).

[0035] Figure 3 This is a flowchart of the present invention. Detailed Implementation

[0036] To facilitate understanding and implementation of the present invention by those skilled in the art, the present invention will be further described in detail below with reference to embodiments. It should be understood that the embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0037] Example 1:

[0038] To make the objectives, technical solutions and advantages of the present invention clearer, the following detailed description is provided in conjunction with the accompanying drawings and examples.

[0039] An automatic selection method for tree canopy samples in high spatial resolution remote sensing images includes the following steps:

[0040] Step 1: Calculate the Normalized Differential Vegetation Index (NDVI) using digital number (DN) data from domestic GF-2 imagery (spatial resolution 0.8 meters). Normalized Differential Vegetation Index The calculation formula is:

[0041] (1)

[0042] in, , These are digital quantization values ​​of the near-infrared and red bands obtained from observations using domestically produced GF-2 imagery.

[0043] Step 2: Traverse each pixel of the Normalized Difference Vegetation Index (NDVI) calculated in Step 1, constructing an inner window and an outer window for each traversed pixel. The outer window is larger than the inner window, and the traversed pixel is located at the center of the inner and outer windows. In this embodiment, a 3×3 window and a 5×5 window are set centered on each traversed pixel (hereinafter referred to as the 3×3 window as the inner window and the 5×5 window as the outer window). A schematic diagram of the window settings is shown below. Figure 1 As shown:

[0044] Step 3: Calculate the normalized vegetation index variation coefficient for the pixels contained in the inner and outer windows in Step 2. The calculation formula is as follows:

[0045] (2)

[0046] in, and These are the normalized vegetation index variation coefficients for the pixels contained in the inner and outer windows, respectively. and These are the normalized standard deviations of the vegetation index for the pixels contained in the inner and outer windows, respectively. and These are the average normalized vegetation indices of the pixels contained in the inner and outer windows, respectively.

[0047] Step 4: Vegetation Canopy Determination. The normalized coefficient of variation (NCV) of the vegetation index of the pixels included in the inner and outer windows, calculated in Step 3, is used to determine whether the currently traversed pixel is a vegetation canopy pixel. The determination criteria are as follows:

[0048] like If so, then the currently traversed pixels are vegetation canopy pixels;

[0049] like If so, then the currently traversed pixels are non-vegetation canopy pixels;

[0050] Step 5: Additional Determination of Canopy Pixels. Step 4 extracted canopy pixels based on the normalized vegetation index (NVI) variation coefficients of the pixels contained in the inner and outer windows. This step is used to distinguish canopies from other vegetation types. If the NVI corresponding to a canopy pixel is greater than or equal to 0.15, then the canopy pixel is a canopy pixel.

[0051] By iterating through all pixels of the normalized vegetation index calculated in step 1 and repeating steps 1-5, all canopy pixels within the scene can be obtained, resulting in canopy samples within the scene. The canopy recognition results are shown in Figures 2(a)-2(d). As can be seen from Figures 2(c) and 2(d), this invention demonstrates good generalization ability for addressing the problem of poor canopy sample quality in high spatial resolution remote sensing images caused by terrain shadows and canopy overlap. It can accurately identify canopies at the single-tree scale for different scenes, improving the quality of canopy samples. In one embodiment, a computer device is also provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0052] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0053] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0054] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods.

[0055] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for automatically selecting a tree crown sample in a high spatial resolution remote sensing image, characterized in that, The method comprises the following steps: Step 1: calculating a normalized vegetation index according to digital quantization value data; Step 2: traversing each pixel of the normalized vegetation index, constructing an inner window and an outer window of the traversed pixel, the outer window being larger than the inner window, and the traversed pixel being located at the center of the inner window and the outer window; Step 3: calculating a normalized vegetation index coefficient of variation of the pixels contained in the inner window and the outer window; Step 4: determining whether the currently traversed pixel is a vegetation canopy pixel based on the normalized vegetation index coefficient of variation of the pixels contained in the inner window and the outer window; Step 5: for the determined vegetation canopy pixel, if the normalized vegetation index corresponding to the vegetation canopy pixel is greater than or equal to a set threshold, the vegetation canopy pixel is a tree canopy pixel, The normalized vegetation index coefficient of variation of the pixels contained in the inner window and the outer window is based on the following formula: wherein, and are the normalized difference vegetation index coefficient of variation for the pixels contained in the inner and outer windows, respectively, and are the normalized difference vegetation index standard deviation for the pixels contained in the inner and outer windows, respectively, and are the normalized difference vegetation index mean for the pixels contained in the inner and outer windows, respectively, The determination of whether the currently traversed pixel is a vegetation canopy pixel comprises the following steps: If then the current traversed pixel is a vegetation canopy pixel; If then the current traversed pixel is a non-vegetation canopy pixel.

2. The method according to claim 1, wherein, The normalized difference vegetation index Based on the following equation: wherein, and are digital quantized value data in near-infrared and red wave bands, respectively.

3. The method according to claim 1, wherein, The set threshold in the step 5 is 0.

15.

4. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to realize the steps of the method for automatically selecting a tree canopy sample in a high-spatial-resolution remote sensing image according to any one of claims 1 to 3.

5. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the method for automatically selecting a tree canopy sample in a high-spatial-resolution remote sensing image according to any one of claims 1 to 3.

6. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the method for automatically selecting a tree canopy sample in a high-spatial-resolution remote sensing image according to any one of claims 1 to 3.

Images