A rice remote sensing identification method and system based on random forest and regional growth

Abstract

The present invention discloses a method and system for remote sensing identification of rice based on random forest and regional growth, wherein the method steps include: obtaining remote sensing image data of rice and preprocessing to obtain preprocessed remote sensing image data; based on the preprocessed remote sensing image data, obtaining a remote sensing image spectral feature set, a spectral index feature set and a texture feature set, and constructing a first feature space; based on the first feature space and the preprocessed remote sensing image data, obtaining a rice classification result; based on the preprocessed remote sensing image data, constructing a second feature space; based on the rice classification result and the second feature space, obtaining a rice identification result; based on the first feature space, obtaining an image segmentation result; based on the image segmentation result and the rice identification result, completing rice identification. The present invention makes up for the problems of incomplete remote sensing identification of rice area and inaccurate boundaries in the traditional random forest method, and improves the rice identification rate.

Description

Technical Field

[0001] The invention belongs to the technical field of remote sensing images, and in particular relates to a rice remote sensing identification method and system based on random forest and regional growth. Background Art

[0002] Rice is the main grain crop in my country and is grown in both the north and south of my country. Accurately understanding its planting area is of great significance for grain production estimation and grain subsidies. Traditionally, the statistics of rice planting area are obtained by local governments through statistical analysis of reported data from various villages and government spot checks. Due to the interference of human factors and spot checks, the statistical results have certain errors and waste a lot of manpower, material resources, financial resources and time. With the rapid development of remote sensing technology and image interpretation methods, it has become possible to extract rice planting areas on a large scale, quickly and accurately through remote sensing technology.

[0003] There are two main types of traditional machine learning methods for rice identification: (1) pixel-based classification, such as decision tree classification, maximum likelihood classification, random forest classification, support vector machine classification, etc. This method uses a single pixel as the basic processing unit and mainly classifies according to the spectral information presented by the pixel. It rarely uses other features and spatial information, such as the texture and shape of the data type. This method is difficult to solve salt and pepper noise, and there are problems such as different objects with the same spectrum or different spectra for the same object. (2) Object-oriented classification method. This algorithm is divided into three steps: data segmentation, feature selection, and image classification. Compared with pixel-based classification methods, this method has higher accuracy, completeness, and efficiency. Due to the complex terrain and fragmented plots in my country's rice-growing areas, both methods have problems such as incomplete rice identification and inaccurate identification of boundaries. Summary of the invention

[0004] In order to solve the above problems, the present invention combines the regional growing technology with the results of random forest classification to extract rice. Multi-layer feature information is used to improve the accuracy of late rice extraction, and a dynamic threshold is used to achieve local adaptability, so as to accurately extract crop areas of different shapes, sizes and conditions.

[0005] To achieve the above object, the present invention provides a rice remote sensing identification method based on random forest and regional growth, the method comprising:

[0006] Acquire remote sensing image data of rice and perform preprocessing to obtain preprocessed remote sensing image data;

[0007] Based on the preprocessed remote sensing image data, a remote sensing image spectral feature set, a spectral index feature set and a texture feature set are obtained, and a first feature space is constructed;

[0008] Based on the first feature space and the preprocessed remote sensing image data, obtaining a rice classification result;

[0009] constructing a second feature space based on the preprocessed remote sensing image data;

[0010] Based on the rice classification result and the second feature space, obtaining a rice recognition result;

[0011] Based on the first feature space, obtaining an image segmentation result;

[0012] Rice recognition is completed based on the image segmentation result and the rice recognition result.

[0013] Preferably, after the second feature space is constructed, the grayscale value variance of late rice in each band in the second feature space is unified based on the rice classification result:

[0014]

[0015] Wherein, i represents the i-th band in the second feature space; G i Indicates the characteristic information pixel value after calculation; g i Represents the pixel value of feature information before calculation; σ i Represents the standard deviation of the feature pixel value; σ0 represents the standard deviation of the first feature space pixel value.

[0016] Preferably, the method for obtaining the rice identification result includes:

[0017] Performing region growing on a single object in the rice classification result to obtain a region growing result;

[0018] Based on the regional growth result, obtaining the rice identification result;

[0019] The single object means that the classification result of each pixel in the object is rice, and at least one of the four neighboring pixels of any pixel in the object belongs to the pixel set.

[0020] Preferably, the method for performing the region growing comprises:

[0021] (1) Take a single object in the classification result of crop M as the initial region D0, and calculate the mean value m of the initial region D0 respectively. i，0 and variance σ i，0 , where i represents the i-th band in the second feature space;

[0022] (2) Determine σ i，0 ≤σ max , if true, the object proceeds to the next step, otherwise stops growing;

[0023] (3) In the nth iteration, according to the mean m of the previous iteration i，n-1 and variance σ i，n-1 Calculate the new threshold range Ω i，n =[m i，n-1 -θσ i，n-1 , m i，n-1 +θσ i，n-1 ];

[0024] (4) According to Ω i,n , σ i,n-1 ≤εσ i,0 and the area D after the previous iteration n-1 Calculate the new area D n , which is composed of n-1 Connected and meets the threshold condition Ω i,n The pixel composition of

[0025] (5) Calculate the new area D n The mean value m i,n and variance σ i,n ;

[0026] (6) Repeat steps (3) to (5) until the range no longer expands.

[0027] Preferably, the method for rice identification includes: performing image segmentation on the first feature space, and calculating the area ratio of the rice identification result to a single area after image segmentation; if the area ratio is greater than a preset threshold, the area is rice; otherwise, the area is non-rice, and rice identification is completed.

[0028] Preferably, the method for calculating the area ratio includes:

[0029]

[0030] Where k represents the area ratio of rice recognition results to a single region after image segmentation; X represents the area occupied by rice after regional growth; Y represents a single region after image segmentation; S X∩Y represents the area of ​​the intersection of region X and region Y; S Y Represents the area of ​​region Y.

[0031] The present invention also provides a rice remote sensing identification system based on random forest and regional growth, the system is used to implement the above method, including: an acquisition module, a calculation module, a random forest module, a construction module, a regional growth module, an image segmentation module and an identification module;

[0032] The acquisition module is used to obtain remote sensing image data of rice and perform preprocessing;

[0033] The calculation module is used to obtain a remote sensing image spectral feature set, a spectral index feature set and a texture feature set based on the pre-processed remote sensing image data, and construct a first feature space;

[0034] The region growing module is used to obtain a rice classification result based on the first feature space and the pre-processed remote sensing image data;

[0035] The construction module is used to construct a second feature space based on the pre-processed remote sensing image data;

[0036] The region growing module is used to obtain a rice recognition result based on the rice classification result and the second feature space;

[0037] The image segmentation module is used to obtain an image segmentation result based on the first feature space;

[0038] The recognition module is used to complete rice recognition based on the image segmentation result and the rice recognition result.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention discloses a rice remote sensing identification method and system based on the combination of random forest and regional growth, which makes up for the problems of incomplete rice area identification and inaccurate boundaries of the traditional random forest method, improves the area extraction accuracy of the rice planting area, and has good economic value. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0042] Figure 1 A schematic diagram of a method flow of an embodiment of the present invention;

[0043] Figure 2 A schematic diagram of a study area according to an embodiment of the present invention;

[0044] Figure 3 Schematic diagram of the system structure of an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] Embodiment 1

[0048] like Figure 1 FIG. 1 is a schematic diagram of the method flow of this embodiment, and the steps include:

[0049] S1. Acquire remote sensing image data of rice and perform preprocessing to obtain preprocessed remote sensing image data.

[0050] In this embodiment, the acquired remote sensing image data may be single-phase or multi-phase remote sensing image data; specifically, it includes:

[0051] Study area: The central and eastern part of Dongkou County, Hunan Province was selected as the study area. Figure 2 shown.

[0052] Image data: Two phases of Planet remote sensing image data are selected, namely the SuperDove satellite data on September 6, 2022 and the Cove data on September 27, 2022; and the remote sensing image data are preprocessed by radiation calibration, atmospheric correction, image registration, etc.

[0053] Sample data: Samples were obtained through field surveys, drone aerial photography, and GF-2 remote sensing images on September 27, 2022. According to the ground features in the study area, the classified samples were divided into six types, namely late rice, woodland, bare land or mid-season rice, water bodies, greenhouses, and building land as shown in Table 1.

[0054] Table 1

[0055]

[0056] S2. Based on the preprocessed remote sensing image data, a spectral feature set, a spectral index feature set and a texture feature set of the remote sensing image are obtained, and a first feature space is constructed.

[0057] After processing the multispectral remote sensing data in S1, the remote sensing image spectral feature set, spectral index feature set and texture feature set are obtained by calculation.

[0058] The spectral index feature set includes: the normalized vegetation index and the normalized water index of remote sensing image data. Specifically, the normalized vegetation index and the normalized water index calculation formula are used to calculate the normalized vegetation index and the normalized water index:

[0059]

[0060]

[0061] In the formula, NDVI represents the normalized vegetation index; NDWI represents the normalized water index; NIR represents the reflectance value of the near-infrared band; R represents the reflectance value of the red light band; and G represents the reflectance value of the green light band.

[0062] The method of obtaining the texture feature set includes: selecting a 3×3 window and extracting five statistics: mean, variance, homogeneity, entropy, and contrast. The calculation formula is as follows:

[0063]

[0064] Variance=∑ i ∑ j (iu) 2 P(i, j)

[0065]

[0066] Entropy=-∑ i ∑ j P(i, j)log[P(i, j)]

[0067] Contrast=∑ i ∑ j P(i, j)·(ij) 2

[0068] Where n is the order of grayscale values, p(i, j) is the n×n normalized co-occurrence matrix, and u is the mean of p(i, j).

[0069] S3. Obtain rice classification results based on the first feature space and the preprocessed remote sensing image data.

[0070] A random forest classifier is used to classify and identify rice to obtain a rice classification result. In this embodiment, a random forest classification tool in ENVI5.3 is used, wherein the number of random forest trees is 100, and the number of features is the arithmetic square root of the number of bands.

[0071] S4. Construct a second feature space based on the preprocessed remote sensing image data.

[0072] In this implementation, the second feature space includes the red band, near infrared band, red edge band, and normalized vegetation index in the remote sensing image on September 6, 2022, and the red band, near infrared band, and normalized vegetation index in the remote sensing image on September 27, 2022. After constructing the second feature space, it is necessary to use the rice classification results to unify the grayscale value variance of late rice in each band in the second feature space:

[0073]

[0074] Where i represents the i-th band in the second feature space; G i Indicates the characteristic information pixel value after calculation; g i Represents the pixel value of feature information before calculation; σ i Represents the standard deviation of the characteristic pixel value; σ0 represents the standard deviation of the first characteristic space pixel value.

[0075] S5. Based on the rice classification result and the second feature space, obtain the rice recognition result.

[0076] Perform region growing on a single object in the rice classification result. A single object is specifically: the classification result of each pixel in the object is rice, and at least one of the four neighboring pixels of any pixel a in the object belongs to the pixel set; at the same time, the number of pixels in the object must satisfy: n≤N, where N represents the minimum threshold number of pixels, which is set to 20 in this embodiment.

[0077] The specific steps of region growing include:

[0078] (1) Take the single object in the rice recognition result as the initial area D0, and calculate the mean value m of the initial area D0 respectively. i，0 and variance σ i，0 , where i represents the i-th band in the second feature space;

[0079] (2) Determine σ i，0 ≤σ max , if true, the object proceeds to the next step, otherwise stops growing;

[0080] (3) In the nth iteration, according to the mean m of the previous iteration i，n-1 and variance σ i，n-1 Calculate the new threshold range Ω i，n =[m i，n-1 -θσ i，n-1 , m i，n-1 +θσ i，n-1 ];

[0081] (4) According to Ω i，n , σ i，n-1≤εσ i，0 and the area D after the previous iteration n-1 Calculate the new area D n , which is composed of n-1 Connected and meets the threshold condition Ω i，n The pixel composition of

[0082] (5) Calculate the new area D n The mean value m i，n and variance σ i，n ;

[0083] (6) Repeat steps 3 to 5 until the range stops expanding.

[0084] Among them, σ max , θ, ε, are all regional growth model parameters, which are set to 2, 2.3, and 1.1 respectively. σ max is the variance threshold. Since the random forest classification may mistakenly identify some non-late rice areas as late rice, the variance threshold strategy helps to start the growth process from the late rice area and effectively suppress the non-late rice area. θ is the growth factor, which can be configured by increasing / decreasing its value. The larger θ is, the more fully the segmented area is, but it is likely to be over-segmented; and the smaller θ is, the more likely it is to cause under-segmentation. The spectral characteristics of late rice may be similar to those of other nearby vegetation, and the regional growing model may misjudge the non-late rice area near late rice as late rice. The parameter ε can suppress this phenomenon. The smaller ε is, the better the suppression effect is, but incomplete extraction of late rice areas may occur.

[0085] S6. Based on the first feature space, obtain the image segmentation result.

[0086] The first feature space is segmented, and the area ratio k of the rice recognition result in a single area after image segmentation is calculated; the first feature space is segmented into several specific areas with unique properties by the edge segmentation method in ENVI5.3, where the segmentation threshold is 25, the merging threshold is 75, and the size of the texture kernel is 3. The method for calculating the area ratio k of the rice recognition result in a single area after image segmentation includes:

[0087]

[0088] Where k represents the area ratio of rice recognition results to a single region after image segmentation; X represents the area occupied by rice after regional growth; Y represents a single region after image segmentation; S X∩Y represents the area of ​​the intersection of region X and region Y; S Y Represents the area of ​​region Y.

[0089] S7. Complete rice recognition based on the image segmentation results and the rice recognition results.

[0090] Based on the above image segmentation results, if k is greater than the threshold K, the area is rice, otherwise the area is non-rice, and the rice recognition is completed.

[0091] In this embodiment, the threshold K is set to 0.4, and the rice remote sensing recognition method based on the combination of random forest and regional growth and the extraction accuracy of random forest and object-oriented random forest are shown in Table 2. Finally, based on the obtained image segmentation results and the rice recognition results obtained in the above steps, rice recognition is completed.

[0092] Table 2

[0093] Classification method PA(%) UA(%) OA (%) Kappa (%) Random Forest 80.4 98.0 89.4 78.8 Object-Oriented Random Forest 81.2 98.1 89.8 79.6 Combination of random forest and region growing 90.4 98.3 94.4 88.8 .

[0094] Embodiment 2

[0095] like Figure 3 As shown, it is a schematic diagram of the system structure of this embodiment, including: an acquisition module, a calculation module, a random forest module, a construction module, a regional growth module, an image segmentation module and a recognition module. Among them, the acquisition module is used to obtain remote sensing image data of rice and perform preprocessing; the calculation module is used to obtain the remote sensing image spectral feature set, spectral index feature set and texture feature set based on the preprocessed remote sensing image data, and construct a first feature space; the random forest module is used to obtain the rice classification result based on the first feature space and the preprocessed remote sensing image data; the construction module is used to construct the second feature space based on the preprocessed remote sensing image data; the regional growth module is used to obtain the rice recognition result based on the rice classification result and the second feature space; the image segmentation module is used to obtain the image segmentation result based on the first feature space; the recognition module is used to complete the rice recognition based on the image segmentation result and the rice recognition result.

[0096] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A rice remote sensing identification method based on random forest and regional growth, characterized in that: Methods include: Acquire remote sensing image data of rice and perform preprocessing to obtain preprocessed remote sensing image data; Based on the preprocessed remote sensing image data, a remote sensing image spectral feature set, a spectral index feature set and a texture feature set are obtained, and a first feature space is constructed; Based on the first feature space and the pre-processed remote sensing image data, a rice classification result is obtained; the method for obtaining the rice classification result comprises: based on the first feature space and the pre-processed remote sensing image data, a random forest classifier is used to perform remote sensing identification on rice to obtain the rice classification result; Based on the preprocessed remote sensing image data, a second feature space is constructed; after the second feature space is constructed, based on the rice classification result, the grayscale value variance of late rice in each band in the second feature space is unified: Where i represents the i-th band in the second feature space; G i Indicates the characteristic information pixel value after calculation; g i Represents the pixel value of feature information before calculation; σ i Represents the standard deviation of the characteristic pixel value; σ0 represents the standard deviation of the first characteristic space pixel value; Based on the rice classification result and the second feature space, a rice recognition result is obtained; the method for obtaining the rice recognition result includes: performing region growing on a single object in the rice classification result to obtain a region growing result; based on the region growing result, the rice recognition result is obtained; the single object indicates that: the classification result of each pixel in the object is rice, and at least one of the four neighboring pixels of any pixel in the object belongs to a pixel set; wherein the method for performing the region growing includes: (1) Take a single object in the classification result of crop M as the initial region D0, and calculate the mean value m of the initial region D0 respectively. i，0 and variance σ i，0 , where i represents the i-th band in the second feature space; (2) Determine σ i，0 ≤σ max , if true, the object proceeds to the next step, otherwise stops growing; (3) In the nth iteration, according to the mean m of the previous iteration i，n-1 and variance σ i，n-1 Calculate the new threshold range Ω i，n =[m i，n-1 -θσ i，n-1 , m i，n-1 +θσ i，n-1 ]; (4) According to Ω i,n , σ i,n-1 ≤εσ i,0 and the area D after the previous iteration n-1 Calculate the new area D n , which is composed of n-1 Connected and meets the threshold condition Ω i,n The pixel composition of (5) Calculate the new area D n The mean value m i,n and variance σ i,n ; (6) Repeat steps (3) to (5) until the range no longer expands; Based on the first feature space, obtaining an image segmentation result; Rice recognition is completed based on the image segmentation result and the rice recognition result.

2. The rice remote sensing identification method based on random forest and regional growth according to claim 1 is characterized in that: The method for rice identification includes: performing image segmentation on the first feature space, and calculating the area ratio of the rice identification result to the single area after image segmentation. If the area ratio is greater than a preset threshold, the area is rice, otherwise the area is non-rice, and rice identification is completed.

3. The rice remote sensing identification method based on random forest and regional growth according to claim 2 is characterized in that: Methods for calculating the area ratio include: Where k represents the area ratio of rice recognition results to a single region after image segmentation; X represents the area occupied by rice after regional growth; Y represents a single region after image segmentation; S X∩Y represents the area of ​​the intersection of region X and region Y; S Y Represents the area of ​​region Y.

4. A rice remote sensing identification system based on random forest and regional growth, the system being used to implement the method according to any one of claims 1 to 3, characterized in that: include: Acquisition module, calculation module, random forest module, construction module, region growing module, image segmentation module and recognition module; The acquisition module is used to acquire remote sensing image data of rice and perform preprocessing to obtain preprocessed remote sensing image data; The calculation module is used to obtain a remote sensing image spectral feature set, a spectral index feature set and a texture feature set based on the pre-processed remote sensing image data, and construct a first feature space; The random forest module is used to obtain a rice classification result based on the first feature space and the pre-processed remote sensing image data; The construction module is used to construct a second feature space based on the pre-processed remote sensing image data; The region growing module is used to obtain a rice recognition result based on the rice classification result and the second feature space; The image segmentation module is used to obtain an image segmentation result based on the first feature space; The recognition module is used to complete rice recognition based on the image segmentation result and the rice recognition result.

Images