Method and system for estimating spatial distribution information of under-forest staple food bamboos for giant pandas, and device

Abstract

Method of estimating spatial distribution of bamboo under forest cover for giant pandas, comprising: obtaining GF-1 satellite panchromatic and multispectral (PMS) images; obtaining digital elevation m

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of determining of spatial distribution information of plants, and in particular, to a method and system for estimating spatial distribution information of under-forest staple food bamboos for giant pandas and a device. BACKGROUND

[0002] Staple food bamboos for giant pandas are food on which giant pandas live and an important component of a giant panda habitat. Mastering a distribution and quantity of the staple food bamboos for giant pandas is of great significance to habitat selection, ecological carrying capacity evaluation, and population reproduction for giant pandas.

[0003] At present, distribution data of the staple food bamboos for giant pandas is obtained mainly by means of two methods: national surveys on giant pandas in China and routine monitoring in reserves. The national surveys on giant pandas in China are carried out about once every ten years, and data on bamboo species, plant height, basal stem, survival state, and the like of the staple food bamboos for giant pandas is obtained by arranging vegetation quadrats on a survey sample line. Routine monitoring is performed mostly on fixed sample lines, which increases a monitoring frequency, but obtained indexes are fewer than those obtained in national surveys on giant pandas in China.

[0004] Since the staple food bamboos for giant pandas are mostly distributed under forests and are covered by upper canopy trees, it is impossible to obtain distribution information of the staple food bamboos for giant pandas directly by remote sensing. A method of combining a sample line and quadrats used in manual surveys can only obtain sampling results, but cannot obtain relatively accurate spatial distribution information of the staple food bamboos for giant pandas. In addition, field surveys need a lot of manpower, material resources and time. Therefore, an efficient, accurate and low-cost method is urgently needed to obtain the spatial distribution information of the underforest staple food bamboos for giant pandas. SUMMARY

[0005] An obj ective of the present disclosure is to provide a method and system for estimating spatial distribution information of under-forest staple food bamboos for giant pandas and a device, which can efficiently, and accurately obtain the spatial distribution information of the under-forest staple food bamboos for giant pandas at a low cost.

[0006] To achieve the above objective, the present disclosure provides the following solutions.

[0007] In a first aspect, the present disclosure provides a method for estimating spatial distribution information of under-forest staple food bamboos for giant pandas, including:

[0008] obtaining GF-1 panchromatic and multispectral (PMS) images, digital elevation model (DEM) data, field survey data and a forest map of a giant panda reserve, where the field survey data is use to mark under-forest samples with bamboos and under-forest samples without bamboos in a field survey region based on whether there is a spatial distribution of under-forest staple food bamboos for giant pandas, and the field survey region is a part of the giant panda reserve;

[0009] segmenting the GF-1 PMS images and the DEM data to obtain a plurality of segmentation units, and extracting features from the data in each segmentation unit to obtain a feature set corresponding to each segmentation unit, where the feature set includes spectral features, texture features, vegetation indexes, and terrain features;

[0010] extracting, based on the forest map, a potential distribution region of the under-forest staple food bamboos for giant pandas, and superposing the field survey region, the potential distribution region of the under-forest staple food bamboos for giant pandas, and the segmentation units based on geographical coordinates to determine label data of each segmentation unit in the potential distribution region of the under-forest staple food bamboos for giant pandas, where the label data is under-forest samples with bamboos, under-forest samples without bamboos, and under-forest no samples;

[0011] building a spatial distribution information estimation model of under-forest staple food bamboos for giant pandas by using sample data and a deep learning algorithm, where the sample data includes feature sets corresponding to segmentation units whose label data is under-forest samples with bamboos and feature sets corresponding to segmentation units whose label data is under-forest samples without bamboos; and

[0012] inputting feature sets corresponding to all target segmentation units into the spatial distribution information estimation model of under-forest staple food bamboos for giant pandas to obtain spatial distribution information of the under-forest staple food bamboos for giant pandas corresponding to each target segmentation unit, where the target segmentation unit is a segmentation unit whose label data is under-forest no samples.

[0013] In a second aspect, the present disclosure provides a system for estimating spatial distribution information of under-forest staple food bamboos for giant pandas, including:

[0014] a giant panda reserve original information acquisition module, configured to obtain GF-1 PMS images, DEM data, field survey data and a forest map of a giant panda reserve, where the field survey data is use to mark under-forest samples with bamboos and under-forest samples without bamboos in a field survey region based on whether there is a spatial distribution of under-forest staple food bamboos for giant pandas, and the field survey region is a part of the giant panda reserve;

[0015] a feature extraction module, configured to segment the GF-1 PMS images and the DEM data to obtain a plurality of segmentation units, and extract features from the data in each segmentation unit to obtain a feature set corresponding to each segmentation unit, where the feature set includes spectral features, texture features, vegetation indexes, and terrain features;

[0016] a potential distribution region segmentation module, configured to extract, based on the forest map, a potential distribution region of the under-forest staple food bamboos for giant pandas, and superpose the field survey region, the potential distribution region of the under-forest staple food bamboos for giant pandas, and the segmentation units based on geographical coordinates to determine label data of each segmentation unit in the potential distribution region of the under-forest staple food bamboos for giant pandas, where the label data is under-forest samples with bamboos, under-forest samples without bamboos, and under-forest no samples;

[0017] an estimation model building module, configured to build a spatial distribution information estimation model of under-forest staple food bamboos for giant pandas by using sample data and a deep learning algorithm, where the sample data includes feature sets corresponding to segmentation units whose label data is under-forest samples with bamboos and feature sets corresponding to segmentation units whose label data is under-forest samples without bamboos; and

[0018] a module for estimating spatial distribution information of under-forest staple food bamboos for giant pandas, configured to input feature sets corresponding to all target segmentation units into the spatial distribution information estimation model of under-forest staple food bamboos for giant pandas to obtain spatial distribution information of the under-forest staple food bamboos for giant pandas corresponding to each target segmentation unit, where the target segmentation unit is a segmentation unit whose label data is under-forest no samples.

[0019] The system for estimating spatial distribution information of under-forest staple food bamboos for giant pandas further includes:

[0020] a module for drawing a spatial distribution map of under-forest staple food bamboos for giant pandas, configured to draw a spatial distribution map of the under-forest staple food bamboos for giant pandas in the giant panda reserve based on geographical coordinates of segmentation units whose label data is under-forest samples with bamboos, geographical coordinates of segmentation units whose label data is under-forest samples without bamboos, geographical coordinates of all the target segmentation units, and the spatial distribution information of the under-forest staple food bamboos for giant pandas.

[0021] In a third aspect, the present disclosure provides an electronic device. The electronic device includes a memory and a processor, where the memory is configured to store a computer program, and the processor runs the computer program to enable the electronic device to perform the method for estimating spatial distribution information of under-forest staple food bamboos for giant pandas according to the first aspect.

[0022] According to specific embodiments of the present disclosure, the present disclosure has the following technical effects:

[0023] According to the present disclosure, with remote sensing images as master data, with reference to DEM data, a forest map, and field survey data, a spatial distribution information estimation model of under-forest staple food bamboos for giant pandas is constructed by screening feature variables that affect a spatial distribution of staple food bamboos for giant pandas, and a spatial distribution map of the under-forest staple food bamboos for giant pandas is obtained efficiently and accurately at a low cost, which can provide effective data support for monitoring giant pandas and a habitat thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] To describe the technical solutions in embodiments of the present disclosure or in the prior art more clearly, the accompanying drawings required in the embodiments are briefly described below. Apparently, the accompanying drawings in the following description show merely some embodiments of the present disclosure, and other drawings can be derived from these accompanying drawings by those of ordinary skill in the art without creative efforts.

[0025] FIG. lisa schematic flowchart of a method for estimating spatial distribution information of under-forest staple food bamboos for giant pandas according to an embodiment of the present disclosure;

[0026] FIG. 2 is a spatial distribution diagram of field survey samples according to an embodiment of the present disclosure;

[0027] FIG. 3 is a receiver operating characteristic (ROC)-LV curve graph according to an embodiment of the present disclosure;

[0028] FIG. 4 is a classification feature importance score diagram according to an embodiment of the present disclosure;

[0029] FIG. 5 is an ROC curve graph according to an embodiment of the present disclosure; and

[0030] FIG. 6 is a spatial distribution map of under-forest staple food bamboos for giant pandas according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] The technical solutions of the embodiments of the present disclosure are clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Apparently, the described embodiments are merely some rather than all of the embodiments of the present disclosure. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present disclosure without creative efforts shall fall within the protection scope of the present disclosure.

[0032] In order to make the above obj ective, features and advantages of the present disclosure clearer and more comprehensible, the present disclosure is further described in detail below with reference to the accompanying drawings and specific implementations.

[0033] Embodiment 1

[0034] As shown in FIG. 1, this embodiment provides a method for estimating spatial distribution information of under-forest staple food bamboos for giant pandas, including the following steps.

[0035] Step 100: Obtain GF-1 PMS images, DEM data, field survey data and a forest map of a giant panda reserve, where the field survey data is use to mark under-forest samples with bamboos and under-forest samples without bamboos in a field survey region based on whether there is a spatial distribution of under-forest staple food bamboos for giant pandas, and the field survey region is a part of the giant panda reserve.

[0036] Step 200: Segment the GF-1 PMS images and the DEM data to obtain a plurality of segmentation units, and extract features from the data in each segmentation unit to obtain a feature set corresponding to each segmentation unit, where the feature set includes spectral features, texture features, vegetation indexes, and terrain features.

[0037] Step 300: Extract, based on the forest map, a potential distribution region of the under-forest staple food bamboos for giant pandas, and superpose the field survey region, the potential distribution region of the under-forest staple food bamboos for giant pandas, and the segmentation units based on geographical coordinates to determine label data of each segmentation unit in the potential distribution region of the under-forest staple food bamboos for giant pandas, where the label data is under-forest samples with bamboos, under-forest samples without bamboos, and under-forest no samples.

[0038] Step 400: Build a spatial distribution information estimation model of under-forest staple food bamboos for giant pandas by using sample data and a deep learning algorithm, where the sample data includes feature sets corresponding to segmentation units whose label data is under-forest samples with bamboos and feature sets corresponding to segmentation units whose label data is underforest samples without bamboos.

[0039] Step 500: Input feature sets corresponding to all target segmentation units into the spatial distribution information estimation model of under-forest staple food bamboos for giant pandas to obtain spatial distribution information of the under-forest staple food bamboos for giant pandas corresponding to each target segmentation unit, where the target segmentation unit is a segmentation unit whose label data is under-forest no samples.

[0040] Further, the method according to this embodiment further includes: drawing a spatial distribution map of the under-forest staple food bamboos for giant pandas in the giant panda reserve based on geographical coordinates of segmentation units whose label data is under-forest samples with bamboos, geographical coordinates of segmentation units whose label data is under-forest samples without bamboos, geographical coordinates of all the target segmentation units, and the spatial distribution information of the under-forest staple food bamboos for giant pandas.

[0041] In this embodiment, the DEM data is DEM data with a 15 m resolution. The forest map can be generated from forest resource data of a management department, and can also be obtained by interpretation of remote sensing images. The field survey data is close to an imaging time of the GF-1 PMS images. The field survey data falls into under-forest samples with bamboos and under-forest samples without bamboos based on whether there is a distribution of staple food bamboos for giant pandas below an arborous layer. Samples with a distribution area of under-forest staple food bamboos for giant pandas greater than 900 m2 and a canopy density greater than 20% are determined as effective under-forest samples with bamboos, and geographical coordinates of survey samples are recorded.

[0042] In a preferred implementation, the segmenting the GF-1 PMS images and the DEM data to obtain a plurality of segmentation units specifically includes the following substeps.

[0043] (1): Preprocess the GF-1 PMS images. For example, in the Environment for Visualizing Images (ENVI) 5.3, preprocessing such as radiometric calibration, atmospheric correction, orthorectification, and mosaic is performed on the GF-1 PMS images.

[0044] (2): Generate a slope map and an aspect map by means of raster surface analysis based on the DEM data. For example, the DEM data is used to generate a slope map and an aspect map by means of raster surface analysis in ArcGIS 10.8.

[0045] (3): Perform band synthesis on preprocessed GF-1 PMS images, the DEM data, the slope map, and the aspect map to obtain a synthetic image. For example, in ENVI 5.3, band synthesis is performed on the preprocessed GF-1 PMS images, the DEM data, the slope map, and the aspect map.

[0046] (4): Crop the synthetic image based on a boundary of the giant panda reserve.

[0047] (5): Perform multi-scale segmentation on the cropped synthetic image to obtain the plurality of segmentation units. For example, multi-scale segmentation is performed on the cropped synthetic image in eCognition 9.0, an optimal segmentation scale of the image is obtained by an ESP2 tool, and the cropped synthetic image is segmented based on this scale.

[0048] In a preferred implementation, the extracting features from the data in each segmentation unit to obtain a feature set corresponding to each segmentation unit specifically includes:

[0049] performing the following operations on the data in each segmentation unit.

[0050] 1): Extract features from the preprocessed GF-1 PMS images in the segmentation unit to obtain spectral features, texture features, and vegetation indexes. 2): Extract features from the DEM data, the slope map, and the aspect map in the segmentation unit to obtain terrain features.

[0051] The feature extraction process is described in detail below by means of an example.

[0052] For example, spectral features, texture features, vegetation indexes, and terrain features of each segmentation unit are extracted in eCognition 9.0, and output as shp files.

[0053] Spectral features SI: Means and standard deviations (StaDs) of a blue band (B), a green band (G), a red band (R), and a near infrared band (N) of the preprocessed GF-1 PMS images are extracted as spectral features.

[0054] Texture features S2: Eight texture features of four bands (the blue band (B), the green band (G), the red band (R), and the near infrared band (N)) of the preprocessed GF-1 PMS images are extracted by using a gray-level co-occurrence matrix (GLCM), and are homogeneity (Hom), contrast (Con), dissimilarity (Dis), entropy (Ent), angular second moment (Ang), mean (Mea), standard deviation (StdDev, Std), and correlation (Cor).

[0055] Vegetation indexes S3: Six vegetation indexes, namely a ratio vegetation index (RVI), a normalized difference vegetation index (NDVI), an enhanced vegetation index (EVI), a difference vegetation index (DVI), a green normalized difference vegetation index (GNDVI), and a visible atmospherically resistant index (VARI), are selected, with specific calculation formulas as follows:

[0056] Table 1 Vegetation index calculation formulas

[0057] Index name Calculation formula RVI I co CQ 1 Bq II <>a NDVI Bq — B3 NDVI = Bq + B3 EVI 2.5(B4 - B3) EVI =-----—----- Bq + 6.OS3 — 7 .SBq + 1 DVI DVI = Bq-B3 GNDVI Bq~B2 GNDVI = ---4 Bq + B2 VARI b2 — b3 VARI = -— B2 + B3 — Bq

[0058] Note: Bl, B2, B3, and B4 represent the blue band, the green band, the red band, and the near infrared band of the GF-1 PMS images respectively.

[0059] Terrain features S4: An elevation (Mean HB), a slope (Mean PD) and an aspect (Mean PX) are extracted from the DEM, the slope map and the aspect map as terrain features.

[0060] In a preferred implementation, the extracting, based on the forest map, a potential distribution region of the under-forest staple food bamboos for giant pandas is specifically as follows:

[0061] Since the potential distribution region of the under-forest staple food bamboos for giant pandas has a strong zonal law, and is mostly distributed in zones of coniferous forests, mixed coniferous and broad-leaved forests, and deciduous broad-leaved forests, regions of coniferous forests, mixed coniferous and broad-leaved forests and deciduous broad-leaved forests in the forest map are merged in ArcGIS 10.8 and output as the potential distribution region.

[0062] In a preferred implementation, the superposing the field survey region, the potential distribution region of the under-forest staple food bamboos for giant pandas, and the segmentation units based on geographical coordinates to determine label data of each segmentation unit in the potential distribution region of the under-forest staple food bamboos for giant pandas is specifically as follows:

[0063] In ArcGIS 10.8, the segmentation units are cropped by using the potential distribution region of the under-forest staple food bamboos for giant pandas, and spectral features, texture features, vegetation indexes, and terrain features of each segmentation unit in the potential distribution region of the under-forest staple food bamboos for giant pandas are extracted and used as sample features.

[0064] In ArcGIS 10 8, the field survey region is spatially connected to the segmentation units in the potential distribution region of the under-forest staple food bamboos for giant pandas to determine label data of each segmentation unit in the potential distribution region of the under-forest staple food bamboos for giant pandas.

[0065] The label data and corresponding sample features are output as xlsx files. Based on a number of samples, 70% of samples are used as training samples while 30% samples are used as verification samples. Other segmentation units in the potential distribution region are used as estimation samples.

[0066] In a preferred implementation, the building a spatial distribution information estimation model of under-forest staple food bamboos for giant pandas by using sample data and a deep learning algorithm specifically includes the following substeps.

[0067] (1): Use a regression decision tree to screen features in the feature set corresponding to each segmentation unit in the sample data to obtain optimized sample data. (2): Construct a gradient boosting - deep neural network model, and build the spatial distribution information estimation model of under-forest staple food bamboos for giant pandas based on the optimized sample data and the gradient boosting - deep neural network model.

[0068] In addition, precision of the spatial distribution information estimation model of under-forest staple food bamboos for giant pandas is evaluated by using an area under curve (AUC) value, and accuracy, recall and an Fl value of the spatial distribution information estimation model of underforest staple food bamboos for giant pandas are calculated by using a confusion matrix.

[0069] A model training process is described in detail below by means of an example.

[0070] In python 3.8, sample features are optimized, main features affecting a spatial distribution of under-forest staple food bamboos for giant pandas are selected, and a spatial distribution information estimation model of under-forest staple food bamboos for giant pandas is built by using a deep neural network (DNN), specifically as follows.

[0071] First, a regression decision tree (DecisionTreeRegressor) is used to analyze sample features and sample types (under-forest samples with bamboos or under-forest samples without bamboos), to obtain importance values of the sample features, and top 15 features that affect the distribution of the under-forest staple food bamboos for giant pandas are selected based on an importance order to form optimized samples. Then the optimized samples are input into a gradient boosting model (GradientBoostingClassifier), and the features are recombined by means of one-hot encoding (OneHotEncoder). 200 sample features and 10 sample types are set, and recombined features are input into a three-layer fully connected DNN to estimate the spatial distribution of the under-forest staple food bamboos for giant pandas. Furthermore, precision of the model is evaluated by using an AUC value, and accuracy, recall and an Fl value of the model are calculated by using a confusion matrix (confusion matrix).

[0072] In this embodiment, a method for drawing a distribution map of under-forest staple food bamboos for giant pandas was expounded with Baishuijiang National Nature Reserve in Gansu as an example.

[0073] Baishuijiang National Nature Reserve in Gansu is located at the southernmost tip of Gansu Province and at the junction of Shaanxi, Gansu and Sichuan provinces, and is a concentrated distribution region of wild giant pandas in Gansu Province. The nature reserve has a total area of 183,799 hm2 and a forest coverage rate of 87.3%. The vegetation in the nature reserve is obviously vertically distributed, including an evergreen and deciduous broad-leaved mixed forest, a deciduous broad-leaved forest, a mixed coniferous and broad-leaved forest, a coniferous forest, and an alpine shrub meadow from bottom to top.

[0074] Specific implementation steps were as follows:

[0075] SI: Data collection. With GF-1 PMS images on June 3, 2020 and September 6, 2020 as master data, a distribution map of under-forest staple food bamboos for giant pandas was drawn with reference to DEM data with a 15 m resolution, field survey data in June 2020 (FIG. 2), and a forest map obtained by remote sensing image interpretation.

[0076] S2: Image preprocessing. In ENVI 5.3, preprocessing such as radiometric calibration, atmospheric correction, orthorectification, and mosaic was performed on the GF-1 PMS images. The DEM data was used to generate a slope map and an aspect map by means of raster surface analysis in ArcGIS 10.8. In ENVI 5.3, band synthesis was performed on the above data, and cropping was performed by using a vector boundary of the reserve.

[0077] S3: Multi-scale segmentation. Multi-scale segmentation was performed on a cropped image in eCognition 9.0. It can be seen from an ROC-LV curve (FIG. 3) calculated by means of ESP2 software that optimal image segmentation parameters were as follows: a segmentation scale of 69, a shape factor 0.1, a color factor 0.9, and smoothness and compactness of 0.5.

[0078] S4: Feature set construction. Spectral features, texture features, vegetation indexes, and terrain features of each segmentation unit were extracted in eCognition 9.0 to construct a feature set (Table 2), and output as shp files.

[0079] Table 2 Feature list

[0080] Feature type Number of features Feature name Spectral features 8 Mean B, Mean G, Mean R, MeanN, StaDB, StaD G, StaDR, StaD_N Vegetation index features 6 RVI, NDVI, EVI, DVI, GNDVI, VARI Texture features 32 GL HOM B, GL CON B, GL ASM B, GL ENT B, GLDISB, GLCORB, GL STD B, GL MEA B, GL HOM G, GL CON G, GL^ASM^G, GLENTG, GL DIS G GL COR G, GL STD G, GLMEA^G GL HOM R. GL CON R, GLASM^R GL E^ GLDISR, GL COR R, GL STD R, GL MEA R, GL HOM N, GL CON N, GL ASM N, GL ENT N, GL DIS N, GL COR N, GL STD N, GL MEA N Terrain features 3 Mean HB, Mean PD, Mean PX

[0081] S5: Extraction of a potential distribution region of the under-forest staple food bamboos for giant pandas. Coniferous forests, mixed coniferous and broad-leaved forests, and deciduous broadleaved forests in the forest map were selected, and segmented images were cropped in ArcGIS 10.8 to obtain the potential distribution region (including a feature set of each segmentation unit) of the under-forest staple food bamboos for giant pandas.

[0082] S6: Survey sample processing. In ArcGIS 10.8, under-forest samples with bamboos and under-forest samples without bamboos were superimposed with images of the potential distribution region, and a sample distribution of each segmentation unit was obtained by spatial connection, and was output as an xlsx file. Based on field surveys in June, 2020, 371 under-forest samples with bamboos and 352 under-forest samples without bamboos were obtained. 70% of the sample were used as training samples while 30% of the samples were used as verification samples. Segmentation units without a sample distribution were estimation samples.

[0083] S7: Building of a spatial distribution information estimation model of under-forest staple food bamboos for giant pandas. In python 3.8, a regression decision tree (DecisionTreeRegressor) was used to analyze sample features and sample types, to obtain importance values of the sample features, and top 15 features (MeanHB, GL HOM B, StaD B, NDVI, StaD R, GL ENT B, Mean B, GNDVI, GLMEAN, GL STD B, GL ASM G, GLDISG, GLHOMN, VARI, and EVI) in an importance order were selected based on the importance order (FIG. 4) to form optimized samples, to participate in the building of the spatial distribution information estimation model of under-forest staple food bamboos for giant pandas.

[0084] The optimized samples were input into a gradient boosting model (GradientBoostingClassifier), and parameters were set as (n_estimators=200, random_state=2019, subsample=0.8, maxjiepth=5, min samplesJeaf=l, and ininsanrples^s The features were recombined by means of one-hot encoding (OneHotEncoder). 200 sample features and 10 sample types were set, and recombined features were input into a DNN to estimate the spatial distribution of the under-forest staple food bamboos for giant pandas.

[0085] The DNN included three fully connected layers, with a first-layer network having 200 neurons and an activation function being Relu, a second-layer network having 100 neurons and an activation function being Relu, and a third-layer network having 10 neurons and an activation function being softmax. An optimizer was provided as stochastic gradient descent (SGD), with a learning rate of 0.1.

[0086] After 20 epoch trainings, as shown in FIG. 5, the AUC value reached 0.89, and thus the model had high precision. A calculation result by means of a confusion matrix showed that the spatial distribution information estimation model of under-forest staple food bamboos for giant pandas had accuracy of 89.56%, recall of 93.38%, and an Fl value of 0.90.

[0087] S8: Drawing of a distribution map of under-forest staple food bamboos for giant pandas. The estimation samples were input into a debugged model to estimate the spatial distribution of the underforest bamboos, and the distribution map of the under-forest staple food bamboos for giant pandas (FIG. 6) was drawn in ArcGIS 10.8 based on estimation results.

[0088] Embodiment 2

[0089] In order to perform the corresponding method in Embodiment 1 to achieve corresponding functions and technical effects, a system for estimating spatial distribution information of under-forest staple food bamboos for giant pandas is provided below.

[0090] This embodiment provides a system for estimating spatial distribution information of underforest staple food bamboos for giant pandas, including:

[0091] a giant panda reserve original information acquisition module, configured to obtain GF-1 PMS images, DEM data, field survey data and a forest map of a giant panda reserve, where the field survey data is use to mark under-forest samples with bamboos and under-forest samples without bamboos in a field survey region based on whether there is a spatial distribution of under-forest staple food bamboos for giant pandas, and the field survey region is a part of the giant panda reserve;

[0092] a feature extraction module, configured to segment the GF-1 PMS images and the DEM data to obtain a plurality of segmentation units, and extract features from the data in each segmentation unit to obtain a feature set corresponding to each segmentation unit, where the feature set includes spectral features, texture features, vegetation indexes, and terrain features;

[0093] a potential distribution region segmentation module, configured to extract, based on the forest map, a potential distribution region of the under-forest staple food bamboos for giant pandas, and superpose the field survey region, the potential distribution region of the under-forest staple food bamboos for giant pandas, and the segmentation units based on geographical coordinates to determine label data of each segmentation unit in the potential distribution region of the under-forest staple food bamboos for giant pandas, where the label data is under-forest samples with bamboos, under-forest samples without bamboos, and under-forest no samples;

[0094] an estimation model building module, configured to build a spatial distribution information estimation model of under-forest staple food bamboos for giant pandas by using sample data and a deep learning algorithm, where the sample data includes feature sets corresponding to segmentation units whose label data is under-forest samples with bamboos and feature sets corresponding to segmentation units whose label data is under-forest samples without bamboos; and

[0095] a module for estimating spatial distribution information of under-forest staple food bamboos for giant pandas, configured to input feature sets corresponding to all target segmentation units into the spatial distribution information estimation model of under-forest staple food bamboos for giant pandas to obtain spatial distribution information of the under-forest staple food bamboos for giant pandas corresponding to each target segmentation unit, where the target segmentation unit is a segmentation unit whose label data is under-forest no samples.

[0096] Further, the system further includes:

[0097] a module for drawing a spatial distribution map of under-forest staple food bamboos for giant pandas, configured to draw a spatial distribution map of the under-forest staple food bamboos for giant pandas in the giant panda reserve based on geographical coordinates of segmentation units whose label data is under-forest samples with bamboos, geographical coordinates of segmentation units whose label data is under-forest samples without bamboos, geographical coordinates of all the target segmentation units, and the spatial distribution information of the under-forest staple food bamboos for giant pandas.

[0098] Embodiment 3

[0099] The embodiment of the present disclosure provides an electronic device. The electronic device includes a memory and a processor, where the memory is configured to store a computer program, and the processor runs the computer program to enable the electronic device to perform the method for estimating spatial distribution information of under-forest staple food bamboos for giant pandas according to embodiment 1.

[0100] Optionally, the electronic device described above may be a server.

[0101] In addition, an embodiment of the present disclosure further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for estimating spatial distribution information of under-forest staple food bamboos for giant pandas according to embodiment 1 is implemented.

[0102] Compared with the prior art, the present disclosure has the following advantages:

[0103] (1) A spatial distribution map of under-forest staple food bamboos for giant pandas is obtained at a low cost.

[0104] The staple food bamboos for giant pandas are mostly distributed under forests, so that it is impossible to obtain a spatial distribution of the bamboos directly by remote sensing. In addition, conventional surveys are based on sampling, and thus it is impossible to obtain a whole distribution of the under-forest staple food bamboos for giant pandas. According to the present disclosure, the spatial distribution map of the under-forest staple food bamboos for giant pandas is obtained at a low cost by building a spatial distribution information estimation model of under-forest staple food bamboos for giant pandas by means of remote sensing images, DEM data, a forest map, and field survey data.

[0105] (2) A spatial distribution of the under-forest staple food bamboos for giant pandas is estimated in high precision.

[0106] According to the present disclosure, a regression decision tree (DecisionTreeRegressor) is used to eliminate factors unrelated to the spatial distribution of the under-forest staple food bamboos for giant pandas. A gradient boosting - deep neural network model is built to predict the spatial distribution of the under-forest staple food bamboos for giant pandas. Field survey samples showed that the model had accuracy of 89.56%, recall of 93.38%, and an Fl value of 0.90, which can effectively predict the under-forest staple food bamboos for giant pandas.

[0107] In addition to the above embodiments, the prediction may be performed by means of other algorithms, specifically as follows.

[0108] (1) Feature optimization: In order to find influencing factors related to the distribution of the under-forest staple food bamboos for giant pandas and eliminate interference of irrelevant factors to the spatial distribution information estimation model of the under-forest staple food bamboos for giant pandas, according to the present disclosure, the regression decision tree (DecisionTreeRegressor) is selected to calculate important values of features, and features entering the spatial distribution information estimation model of under-forest staple food bamboos for giant pandas are selected. At present, there are many feature optimization algorithms, and feature selection can be achieved by means of a correlation filtering method, a wrapper method, and an embedding method.

[0109] (2) Gradient boosting algorithm: Gradient boosting is an algorithm to continuously improve model performance and gradually improve precision by learning previous model errors. According to the present disclosure, a gradient boosting decision tree (GBDT) is selected to optimize a loss function, but an eXtreme Gradient Boosting (XGBoost), a light gradient boosting machine (LightGBM), categorical boosting (CatBoost), and other methods can also achieve this goal.

[0110] (3) Deep neural network algorithm: a DNN can achieve fitting of highly complex functions. According to the present disclosure, a three-layer DNN model is built to predict the spatial distribution of the under-forest staple food bamboos for giant pandas, but an artificial neural network (ANN), a multi-layer perceptron (MLP), and other related neural networks can also achieve this goal to some extent.

[0111] Embodiments of this description are described in a progressive manner, each embodiment focuses on the difference from other embodiments, and for the same and similar parts between the embodiments, reference may be made to each other. Since the system disclosed in an embodiment corresponds to the method disclosed in an embodiment, the description is relatively simple, and for related contents, reference may be made to the description of the method.

[0112] Specific examples are used herein to explain the principles and implementations of the present disclosure. The foregoing description of the embodiments is merely intended to help understand the method of the present disclosure and its core ideas; besides, changes may be made by those of ordinary skill in the art to specific implementations and the scope of application in accordance with the ideas of the present disclosure. In conclusion, the content of this description shall not be construed as limitations to the present disclosure.

Claims

1. A method for estimating spatial distribution information of under-forest staple food bamboos for giant pandas, comprising:obtaining GF-1 panchromatic and multispectral (PMS) images, digital elevation model (DEM) data, field survey data and a forest map of a giant panda reserve, wherein the field survey data is use to mark under-forest samples with bamboos and under-forest samples without bamboos in a field survey region based on whether there is a spatial distribution of under-forest staple food bamboos for giant pandas, and the field survey region is a part of the giant panda reserve;segmenting the GF-1 PMS images and the DEM data to obtain a plurality of segmentation units, and extracting features from the data in each segmentation unit to obtain a feature set corresponding to each segmentation unit, wherein the feature set comprises spectral features, texture features, vegetation indexes, and terrain features;extracting, based on the forest map, a potential distribution region of the under-forest staple food bamboos for giant pandas, and superposing the field survey region, the potential distribution region of the under-forest staple food bamboos for giant pandas, and the segmentation units based on geographical coordinates to determine label data of each segmentation unit in the potential distribution region of the under-forest staple food bamboos for giant pandas, wherein the label data is under-forest samples with bamboos, under-forest samples without bamboos, and under-forest no samples;building a spatial distribution information estimation model of under-forest staple food bamboos for giant pandas by using sample data and a deep learning algorithm, wherein the sample data comprises feature sets corresponding to segmentation units whose label data is under-forest samples with bamboos and feature sets corresponding to segmentation units whose label data is under-forest samples without bamboos; andinputting feature sets corresponding to all target segmentation units into the spatial distribution information estimation model of under-forest staple food bamboos for giant pandas to obtain spatial distribution information of the under-forest staple food bamboos for giant pandas corresponding to each target segmentation unit, wherein the target segmentation unit is a segmentation unit whose label data is under-forest no samples.

2. The method for estimating spatial distribution information of under-forest staple food bamboos for giant pandas according to claim 1, further comprising:drawing a spatial distribution map of the under-forest staple food bamboos for giant pandas in the giant panda reserve based on geographical coordinates of segmentation units whose label data isunder-forest samples with bamboos, geographical coordinates of segmentation units whose label data is under-forest samples without bamboos, geographical coordinates of all the target segmentation units, and the spatial distribution information of the under-forest staple food bamboos for giant pandas.

3. The method for estimating spatial distribution information of under-forest staple food bamboos for giant pandas according to claim 1, wherein the segmenting the GF-1 PMS images and the DEM data to obtain a plurality of segmentation units specifically comprises:preprocessing the GF-1 PMS images;generating a slope map and an aspect map by means of raster surface analysis based on the DEM data;performing band synthesis on preprocessed GF-1 PMS images, the DEM data, the slope map, and the aspect map to obtain a synthetic image;cropping the synthetic image based on a boundary of the giant panda reserve; andperforming multi-scale segmentation on the cropped synthetic image to obtain the plurality of segmentation units.

4. The method for estimating spatial distribution information of under-forest staple food bamboos for giant pandas according to claim 3, wherein the extracting features from the data in each segmentation unit to obtain a feature set corresponding to each segmentation unit specifically comprises:performing the following operations on the data in each segmentation unit:extracting features from the preprocessed GF-1 PMS images in the segmentation unit to obtain spectral features, texture features, and vegetation indexes; andextracting features from the DEM data, the slope map, and the aspect map in the segmentation unit to obtain terrain features.

5. The method for estimating spatial distribution information of under-forest staple food bamboos for giant pandas according to claim 1, wherein the building a spatial distribution information estimation model of under-forest staple food bamboos for giant pandas by using sample data and a deep learning algorithm specifically comprises:using a regression decision tree to screen features in the feature set corresponding to each segmentation unit in the sample data to obtain optimized sample data;constructing a gradient boosting - deep neural network model; andbuilding the spatial distribution information estimation model of under-forest staple food bamboos for giant pandas based on the optimized sample data and the gradient boosting - deep neuralnetwork model.

6. The method for estimating spatial distribution information of under-forest staple food bamboos for giant pandas according to claim 1, further comprising:evaluating precision of the spatial distribution information estimation model of under-forest staple food bamboos for giant pandas by using an area under curve (AUC) value, and calculating accuracy, recall and an Fl value of the spatial distribution information estimation model of underforest staple food bamboos for giant pandas by using a confusion matrix.

7. A system for estimating spatial distribution information of under-forest staple food bamboos for giant pandas, comprising:a giant panda reserve original information acquisition module, configured to obtain GF-I PMS images, DEM data, field survey data and a forest map of a giant panda reserve, wherein the field survey data is use to mark under-forest samples with bamboos and under-forest samples without bamboos in a field survey region based on whether there is a spatial distribution of under-forest staple food bamboos for giant pandas, and the field survey region is a part of the giant panda reserve;a feature extraction module, configured to segment the GF-1 PMS images and the DEM data to obtain a plurality of segmentation units, and extract features from the data in each segmentation unit to obtain a feature set corresponding to each segmentation unit, wherein the feature set comprises spectral features, texture features, vegetation indexes, and terrain features;a potential distribution region segmentation module, configured to extract, based on the forest map, a potential distribution region of the under-forest staple food bamboos for giant pandas, and superpose the field survey region, the potential distribution region of the under-forest staple food bamboos for giant pandas, and the segmentation units based on geographical coordinates to determine label data of each segmentation unit in the potential distribution region of the under-forest staple food bamboos for giant pandas, wherein the label data is under-forest samples with bamboos, under-forest samples without bamboos, and under-forest no samples;an estimation model building module, configured to build a spatial distribution information estimation model of under-forest staple food bamboos for giant pandas by using sample data and a deep learning algorithm, wherein the sample data comprises feature sets corresponding to segmentation units whose label data is under-forest samples with bamboos and feature sets corresponding to segmentation units whose label data is under-forest samples without bamboos; anda module for estimating spatial distribution information of under-forest staple food bamboos for giant pandas, configured to input feature sets corresponding to all target segmentation units into the spatial distribution information estimation model of under-forest staple food bamboos for giantpandas to obtain spatial distribution information of the under-forest staple food bamboos for giant pandas corresponding to each target segmentation unit, wherein the target segmentation unit is a segmentation unit whose label data is under-forest no samples.

8. The system for estimating spatial distribution information of under-forest staple food bamboos for giant pandas according to claim 7, further comprising:a module for drawing a spatial distribution map of under-forest staple food bamboos for giant pandas, configured to draw a spatial distribution map of the under-forest staple food bamboos for giant pandas in the giant panda reserve based on geographical coordinates of segmentation units whose label data is under-forest samples with bamboos, geographical coordinates of segmentation units whose label data is under-forest samples without bamboos, geographical coordinates of all the target segmentation units, and the spatial distribution information of the under-forest staple food bamboos for giant pandas.

9. An electronic device, comprising a memory and a processor, wherein the memory is configured to store a computer program, and the processor runs the computer program to enable the electronic device to perform the method for estimating spatial distribution information of under-forest staple food bamboos for giant pandas according to any one of claims 1 to 6.19

Images