A method and system for identifying dragon fruit planting areas by using multi-source night light remote sensing data

Abstract

The application discloses a kind of pitaya planting area identification method and system of application multi-source night light remote sensing data, belong to night light remote sensing technical field and agricultural monitoring technical field.The present application is based on multi-source remote sensing data, according to the screening rule based on spectrum, space and vegetation characteristics, extracts pitaya potential planting area from night light image, and further constructs and trains pitaya planting area identification model through pitaya potential planting area, avoid the shortcoming that it is difficult to capture the spatial characteristics of strip, dot matrix of pitaya planting area under low spatial resolution night light image, and improve the accuracy of pitaya planting area identification;Since multi-source data at least includes night light image, vegetation image and impervious surface data, from night light, vegetation, thermal environment, topography and city proximity etc. Multi-dimensional information is jointly included in model training, and time sequence consistency constraint is introduced, with high stability and adaptability.

Description

Technical Field

[0001] This invention relates to the fields of nighttime light remote sensing technology and agricultural monitoring technology, and in particular to a method and system for identifying dragon fruit growing areas using multi-source nighttime light remote sensing data. Background Technology

[0002] Currently, the technology for identifying dragon fruit growing areas is mainly achieved by integrating multi-source information with machine learning models. It is primarily based on medium-resolution NPP-VIIRS night light data, extracting seasonal variation features of "bright in winter and dark in summer" to distinguish between agricultural supplementary lighting areas and urban light sources, and using classification algorithms such as decision trees or random forests to generate regional distribution maps.

[0003] However, in actual research, technicians found that due to the low spatial resolution of the NPP-VIIRS luminous product, it was difficult to capture the strip-like and dot-like spatial features of dragon fruit growing areas when using the product to distinguish them, resulting in low accuracy of the identification results. Summary of the Invention

[0004] To address the problems of existing technologies, embodiments of the present invention provide a method and system for identifying dragon fruit growing areas using multi-source nighttime light remote sensing data. The technical solution is as follows:

[0005] On the one hand, a method for identifying dragon fruit growing areas using multi-source nighttime light remote sensing data is provided, the method comprising:

[0006] Acquire multi-source remote sensing data of the target area and perform temporal consistency preprocessing on the multi-source remote sensing data to obtain multi-source remote sensing data that is consistent in both spatial and temporal dimensions; the multi-source remote sensing data includes at least nighttime light images, vegetation images, and impervious surface data;

[0007] Agricultural light source regions are identified in the impermeable surface data, and the features corresponding to the identified agricultural light source regions are set as spectral screening rules; the spectral screening rules include agricultural light source regions.

[0008] Identify non-seasonal crop areas in the vegetation image and set the features of the identified non-seasonal crop areas as vegetation feature filtering rules;

[0009] Identify dot-like or strip-like distribution areas in the night-light image and set the identified dot-like or strip-like distribution areas as spatial filtering rules;

[0010] Based on the spectral screening rules, the spatial screening rules, and the vegetation feature screening rules, the nighttime images are identified, and the common areas identified are identified as potential dragon fruit planting areas.

[0011] Multiple features of the dragon fruit planting area are extracted from the multi-source remote sensing data corresponding to the potential dragon fruit planting area;

[0012] A model for identifying dragon fruit growing areas was obtained by modeling using a random forest regression model, and training samples were set.

[0013] Using the aforementioned features as input variables and the labels of the training samples as output variables, the dragon fruit planting area identification model is trained.

[0014] The trained dragon fruit growing area identification model is used to identify the area to be identified, and a dragon fruit distribution probability map is obtained.

[0015] Optionally, the step of acquiring multi-source remote sensing data of the target area and performing temporal consistency preprocessing on the multi-source remote sensing data includes:

[0016] Acquire multi-source remote sensing data with the same time series;

[0017] Spatial cropping and projection unification are performed on the multi-source remote sensing data to obtain spatially identical multi-source remote sensing data.

[0018] For multi-source remote sensing data with the same time and space, cloud mask removal, missing value repair and normalization are performed.

[0019] The identification of agricultural light source areas in the impermeable surface data includes:

[0020] Based on the impermeable surface data, the agricultural light source area and the non-agricultural light source area in the multi-source data are obtained. The non-agricultural light source area includes urban areas, roads, and industrial areas.

[0021] The identification of non-seasonal crop areas in the vegetation image includes:

[0022] By using the annual mean and annual standard deviation of vegetation images, seasonal crop areas in the target region are screened out to obtain the potential dragon fruit planting areas, and labels are set for the potential dragon fruit planting areas.

[0023] The identification of dot-like or strip-like distribution regions in the night-luminescent image includes:

[0024] Obtain the warm light characteristics of the dragon fruit supplemental lighting;

[0025] Based on the warm light characteristics and the night light image, target areas with red light intensity greater than green light and average brightness higher than a threshold are selected from the agricultural light source areas.

[0026] The dot-matrix or strip-shaped distribution areas in the target region are identified by local density and brightness variance.

[0027] 4. The method according to claim 3, characterized in that, extracting multiple features of the dragon fruit planting area from the multi-source remote sensing data corresponding to the potential dragon fruit planting area includes:

[0028] The multi-source remote sensing data is resampled;

[0029] Feature extraction is performed on the resampled multi-source remote sensing data to obtain the multiple features;

[0030] Among them, the multiple features include at least the night-light feature, vegetation feature, thermal environment feature, topographic feature, and urban proximity feature.

[0031] Optionally, the modeling using a random forest regression model to obtain a dragon fruit planting area identification model includes:

[0032] A random forest regression model was used for modeling, and the model parameters were set as follows: number of decision trees n, minimum number of leaf node samples 1, and out-of-bag sampling ratio 0.7.

[0033] The input to the dragon fruit planting area identification model is multiple feature values ​​corresponding to the area to be identified, and the output is the label of the area to be identified.

[0034] Optionally, setting the training samples includes:

[0035] From the multi-source remote sensing data, areas with concentrated distribution of dragon fruit plants were selected as training samples;

[0036] Extract multiple features from the training samples and set labels for the training samples;

[0037] The training samples are divided into a training set and a validation set according to a certain ratio.

[0038] Optionally, training the dragon fruit growing area identification model using the plurality of features as input variables and the labels of the training samples as output variables includes:

[0039] For a single training sample in the training set, the training process includes:

[0040] Input multiple features of the single training sample into the dragon fruit growing area recognition model, and output the label of the single training sample;

[0041] After all training samples have been trained, the importance index of the multiple features is calculated.

[0042] Optionally, the method further includes:

[0043] Set importance indicators corresponding to the region;

[0044] Based on the importance index, multiple feature weights corresponding to the region are generated, and the multiple feature weights correspond one-to-one with the multiple features.

[0045] Optionally, the method further includes:

[0046] Calculate the ROC curve and AUC value based on the validation set;

[0047] Based on the aforementioned importance indicators, key driving factors corresponding to time and space are set.

[0048] On the other hand, a dragon fruit planting area identification system based on multi-source nighttime light remote sensing data is also provided. The system includes a model training device and a model application device, wherein:

[0049] The model training device is used for:

[0050] Acquire multi-source remote sensing data of the target area and perform temporal consistency preprocessing on the multi-source remote sensing data to obtain multi-source remote sensing data that is consistent in both spatial and temporal dimensions; the multi-source remote sensing data includes at least nighttime light images, vegetation images, and impervious surface data;

[0051] Agricultural light source regions are identified in the impermeable surface data, and the features corresponding to the identified agricultural light source regions are set as spectral screening rules; the spectral screening rules include agricultural light source regions.

[0052] Identify non-seasonal crop areas in the vegetation image and set the features of the identified non-seasonal crop areas as vegetation feature filtering rules;

[0053] Identify dot-like or strip-like distribution areas in the night-light image and set the identified dot-like or strip-like distribution areas as spatial filtering rules;

[0054] Based on the spectral screening rules, the spatial screening rules, and the vegetation feature screening rules, the nighttime images are identified, and the common areas identified are identified as potential dragon fruit planting areas.

[0055] Multiple features of the dragon fruit planting area are extracted from the multi-source remote sensing data corresponding to the potential dragon fruit planting area;

[0056] A model for identifying dragon fruit growing areas was obtained by modeling using a random forest regression model, and training samples were set.

[0057] Using the aforementioned features as input variables and the labels of the training samples as output variables, the dragon fruit planting area identification model is trained.

[0058] The model application device is used for:

[0059] Acquire multi-source remote sensing data of the region to be identified, and obtain multiple features corresponding to the region to be identified;

[0060] The dragon fruit planting area identification model, trained and based on multiple features corresponding to the area to be identified, is used to identify the area and obtain a dragon fruit distribution probability map.

[0061] The beneficial effects of the technical solution provided by the embodiments of the present invention are as follows:

[0062] 1. Based on multi-source remote sensing data, and according to screening rules based on spectral, spatial and vegetation characteristics, potential dragon fruit planting areas are extracted from night light images. Furthermore, through the potential dragon fruit planting areas, a dragon fruit planting area identification model is constructed and trained. This avoids the disadvantage of difficulty in capturing the strip-like and dot-like spatial features of dragon fruit planting areas under low spatial resolution night light images, and improves the accuracy of dragon fruit planting area identification.

[0063] 2. Since multi-source data includes at least nighttime light images, vegetation images, and impervious surface data, it incorporates multi-dimensional information such as nighttime light, vegetation, thermal environment, topography, and urban proximity into the model training, and introduces temporal consistency constraints, thus exhibiting high stability and adaptability.

[0064] 3. Using the dragon fruit distribution probability map as the final identification result, without further thresholding or manual discretization of the probability values, the continuous probability distribution output by the model is directly retained in the map for subsequent use in fine mapping, regional comparison and time series analysis. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0066] Figure 1 This is a schematic diagram of a method for identifying dragon fruit growing areas using multi-source nighttime light remote sensing data, provided by an embodiment of the present invention.

[0067] Figure 2 This is a schematic diagram of a method for identifying dragon fruit growing areas using multi-source nighttime light remote sensing data, provided by an embodiment of the present invention.

[0068] Figure 3 This is a schematic diagram of an unidentified dragon fruit planting area provided in an embodiment of the present invention;

[0069] Figure 4 This is a schematic diagram of the dragon fruit distribution probability provided in an embodiment of the present invention;

[0070] Figure 5 This is a schematic diagram of a method system for identifying dragon fruit growing areas using multi-source nighttime light remote sensing data, provided by an embodiment of the present invention. Detailed Implementation

[0071] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0072] It should be noted that the dragon fruit planting area identification method using multi-source nighttime light remote sensing data in this embodiment of the invention is applicable to other crops that require nighttime supplemental lighting in practical applications.

[0073] In addition, the multi-source remote sensing data described in the embodiments of the present invention includes at least nighttime light images, vegetation images, and impervious surface data. Specifically, in practical applications, the aforementioned nighttime light images, vegetation images, and impervious surface data can be SDGSAT-1 nighttime light images, NASA Black Marble (VNP46A1) nighttime light products, Sentinel-2 vegetation images, MODIS land surface temperature (LST), SRTM digital elevation data, and GAIA impervious surface data, respectively.

[0074] The method described in this embodiment of the invention can also be applied to other crop planting areas in practical applications. Since other crop planting areas do not have nighttime light images, that is, there is no operation such as nighttime supplementary lighting, the above-mentioned multi-source remote sensing data may not include nighttime light images when applied to other crop planting areas.

[0075] Reference Figure 1 As shown in the figure, this invention provides a method for identifying dragon fruit growing areas using multi-source nighttime light remote sensing data. The method includes:

[0076] 101. Acquire multi-source remote sensing data of the target area and perform temporal consistency preprocessing on the multi-source remote sensing data to obtain multi-source remote sensing data that is consistent in both spatial and temporal dimensions.

[0077] Among them, multi-source remote sensing data includes at least nighttime light images, vegetation images, and impervious surface data;

[0078] 102. Identify agricultural light source regions in impermeable surface data, and set the features corresponding to the identified agricultural light source regions as spectral filtering rules; the spectral filtering rules include agricultural light source regions;

[0079] 103. Identify non-seasonal crop areas in vegetation images and set the features of the identified non-seasonal crop areas as vegetation feature filtering rules.

[0080] 104. Identify dot-like or strip-like distribution areas in nighttime light images and set the identified dot-like or strip-like distribution areas as spatial filtering rules;

[0081] It should be noted that steps 101 to 105 are the label construction process, and the potential dragon fruit planting areas generated at the end are used as label data for subsequent model training.

[0082] 105. Extract multiple features of dragon fruit planting areas from multi-source remote sensing data corresponding to potential dragon fruit planting areas;

[0083] Step 105 is the feature construction process, and the resulting multiple features are used to describe the dragon fruit growing area.

[0084] 106. A dragon fruit planting area identification model was obtained by modeling using a random forest regression model, and training samples were set.

[0085] 107. Using multiple features as input variables and the labels of training samples as output variables, train a dragon fruit growing area identification model.

[0086] 108. Using the trained dragon fruit planting area recognition model, the area to be identified is identified, and a dragon fruit distribution probability map is obtained.

[0087] Steps 105 to 108 are the process of model building and model usage.

[0088] It should be noted that in step 108, the trained dragon fruit planting area recognition model identifies the region to be identified, outputting a continuous dragon fruit occurrence probability value in the range of 0–1 for each pixel, and generating a dragon fruit distribution probability map. This probability value reflects the likelihood of dragon fruit planting in a pixel and is a comprehensive judgment result, not a simple binary classification; the model will give the corresponding occurrence probability for the pixel based on the comprehensive expression of multi-source information such as night light features, vegetation features, thermal environment, and topography.

[0089] Using the dragon fruit distribution probability map as the final identification result, without further thresholding or manual discretization of the probability values, the continuous probability distribution output by the model is directly retained in the map for subsequent use in fine mapping, regional comparison and time series analysis. In the above dragon fruit distribution probability map, the higher the probability value, the more likely the pixel is to be a dragon fruit growing area; areas with lower probability values ​​are considered potential areas with higher uncertainty.

[0090] Optionally, step 101, which involves acquiring multi-source remote sensing data of the target area and performing temporal consistency preprocessing on the multi-source remote sensing data, includes:

[0091] 201. Obtain multi-source remote sensing data with the same time series;

[0092] Based on the time parameters of the multi-source remote sensing data, acquire the multi-source remote sensing data at the same time. For the next time, continue to perform the above steps until multiple multi-source remote sensing data with the same time series are obtained.

[0093] 202. Spatial clipping and projection unification are performed on multi-source remote sensing data to obtain spatially identical multi-source remote sensing data;

[0094] For multi-source remote sensing data at the same time, spatial cropping and projection unification are performed to obtain spatially identical multi-source remote sensing data.

[0095] It should be noted that the "spatial sameness" mentioned in the embodiments of the present invention refers to the same space or region contained within the multi-source remote sensing.

[0096] 203. Perform cloud mask removal, missing value repair, and normalization on multi-source remote sensing data with the same time and space.

[0097] Specifically, the cloud mask removal described above can be achieved using optical data or SAR data, and the specific implementation process is not limited in the embodiments of the present invention.

[0098] The missing value repair mentioned above can be achieved by setting default values, and the normalization process mentioned above can be achieved by a normalization algorithm. The process and method of cloud mask removal, missing value repair and normalization process mentioned above in the embodiments of the present invention are not limited.

[0099] It should be noted that in step 101, for the same target area, the spatial consistency of the target area can be determined by determining the coordinates or latitude and longitude of the target area, as well as the coordinates or latitude and longitude of the multi-source remote sensing data, so as to ensure the consistency of the data in the spatial and temporal dimensions and provide standardized input for subsequent identification.

[0100] To further increase sample diversity and improve recognition accuracy, multiple target regions can be set, and multi-source remote sensing data of each target region can be acquired. After performing the above steps on a single target region, the above steps can be performed on the next target region.

[0101] Optional, refer to Figure 2 As shown, step 102, which involves using screening rules based on spectral, spatial, and vegetation characteristics to extract potential dragon fruit growing areas from nighttime light images, includes:

[0102] It should be noted that step 102 is performed based on nighttime light images. That is, it uses nighttime light images as a foundation, combined with parameters from other multi-source data, to filter the nighttime light images in order to extract potential dragon fruit planting areas. The specific process is as follows:

[0103] Identifying agricultural light source areas in impermeable surface data includes:

[0104] 301. Based on the impermeable surface data, the agricultural light source area and the non-agricultural light source area in the multi-source data are obtained. The non-agricultural light source area includes urban areas, roads and industrial areas.

[0105] Specifically, using GAIA impermeable surface data, agricultural light source regions and non-agricultural light source regions in multi-source data are obtained;

[0106] Remove non-agricultural light source areas from the nighttime light image to obtain a nighttime light image that does not contain non-agricultural light source areas.

[0107] The above identification process is achieved using GAIA impermeable surface data acquired at the same time as the nighttime imagery. In practical applications, GAIA impermeable surface data acquired at other times can also be used as a reference, specifically:

[0108] By using GAIA impermeable surface data collected at other times, agricultural light source regions and non-agricultural light source regions in the multi-source data were obtained;

[0109] Compare this non-agricultural light source region (denoted as the second non-agricultural light source region) with the non-agricultural light source region obtained in the above steps (denoted as the first non-agricultural light source region).

[0110] If the absolute value of the area difference between regions is greater than or equal to S1, then the step of identifying agricultural light source regions in the impermeable surface data is repeated.

[0111] Identifying dot-like or strip-like distribution areas in night-luminescent images includes:

[0112] 302. Obtain the warm light characteristics of the dragon fruit supplemental lighting;

[0113] In practical applications, this warm light characteristic can be the spectrum of a dragon fruit supplement light;

[0114] 303. Based on the warm light characteristics and night light images, select target areas from the agricultural light source areas where red light is stronger than green light and the average brightness is higher than the threshold.

[0115] Specifically, based on the spectrum of the dragon fruit supplemental lighting, target areas with stronger red light than green light and average brightness above a threshold were selected from nighttime images that do not include non-agricultural light sources.

[0116] 304. Identify lattice-like or strip-like distribution areas in the target region by using local density and brightness variance;

[0117] Step 304 is used to exclude urban centralized lighting areas.

[0118] Identifying non-seasonal crop areas in vegetation imagery includes:

[0119] 305. By using the annual mean and annual standard deviation of vegetation images, seasonal crop areas in the target area are screened out to obtain potential dragon fruit planting areas, and labels are set for potential dragon fruit planting areas.

[0120] This process can be achieved by using the annual mean (0.4–0.8) and annual standard deviation of the Sentinel-2 NDVI (<0.15) to screen out seasonal crops and retain only evergreen vegetation areas.

[0121] Optionally, step 103, which involves extracting multiple features of the dragon fruit planting area based on multi-source remote sensing data corresponding to the potential dragon fruit planting area, includes:

[0122] Resampling of multi-source remote sensing data;

[0123] Specifically, the sampling process can be achieved by uniformly resampling all data to a resolution of 500 m, thereby further ensuring the consistency of each feature in terms of spatial resolution.

[0124] Feature extraction is performed on the resampled multi-source remote sensing data to obtain multiple features;

[0125] Among these features are at least the following: nighttime light characteristics, vegetation characteristics, thermal environment characteristics, topographic features, and urban proximity characteristics.

[0126] In practical use, the above-mentioned features can be specifically described as follows:

[0127] Night light characteristics include: high light seasonal brightness (night_high), low light seasonal brightness (night_low), brightness amplitude (night_amp), brightness ratio (night_ratio), high light seasonal standard deviation (high_std), low light seasonal standard deviation (low_std), and current night light brightness (night_mar).

[0128] Vegetation characteristics include: NDVI mean (ndvi_mean) and NDVI standard deviation (ndvi_std);

[0129] Thermal environment characteristics include: land surface temperature (LST); it should be noted that the land surface temperature characteristics can be obtained from the land surface temperature corresponding to the target area in the MODIS land surface temperature (LST) of multi-source remote sensing data;

[0130] Topographic features include elevation and slope; it should be noted that surface temperature features can be obtained from the elevation and slope of the target area in the SRTM digital elevation data of multi-source remote sensing data.

[0131] Urban proximity characteristics include: distance from the built-up area (gaia_distance).

[0132] The above-mentioned features are merely exemplary. In practical applications, other features may be added. This embodiment of the invention does not limit the specific features.

[0133] Optionally, step 104, which involves modeling using a random forest regression model to obtain a dragon fruit planting area identification model, includes:

[0134] A random forest regression model was used for modeling, and the model parameters were set as follows: number of decision trees n, minimum number of leaf node samples 1, and out-of-bag sampling ratio 0.7.

[0135] The input to the dragon fruit growing area identification model is multiple feature values ​​corresponding to the area to be identified, and the output is the label of the area to be identified.

[0136] Optionally, setting the training samples in step 104 includes:

[0137] From multi-source remote sensing data, areas with concentrated distribution of dragon fruit plants were selected as training samples; specifically, typical areas with relatively concentrated dragon fruit distribution were selected from candidate areas of multi-source remote sensing data as training sample areas.

[0138] Extract multiple features from the training samples and set labels for the training samples;

[0139] Randomly select sample points within this area (the number of sample points depends on the size of the area; generally, the number of sample points is set to 1000-3000), extract 13 feature values ​​at a resolution of 500 m, and use candidate area labels (dragon fruit = 1, non-dragon fruit = 0) as supervision signals.

[0140] The training samples are divided into training and validation sets in a proportional ratio. This process can be done by dividing the sample set into training and validation sets in an 8:2 ratio for model learning and accuracy evaluation.

[0141] Optionally, step 105, which involves training a dragon fruit growing area identification model using multiple features as input variables and the labels of training samples as output variables, includes:

[0142] For a single training sample in the training set, the training process includes:

[0143] Input multiple features of a single training sample into the dragon fruit growing area recognition model, and output the label of the single training sample;

[0144] After all training samples have been trained, the importance indices of multiple features are calculated.

[0145] Optionally, the method also includes:

[0146] Set importance indicators corresponding to the region;

[0147] Based on the importance index, multiple feature weights corresponding to the region are generated, and each feature weight corresponds to a specific feature.

[0148] Optionally, the method also includes:

[0149] Based on the validation set, the ROC curve and AUC value are calculated. The ROC curve and AUC value are used to evaluate the prediction accuracy and stability of the model. The average AUC value of the model is greater than 0.9, which indicates that the method has high recognition reliability.

[0150] The model's average AUC value exceeds 0.9, indicating that the method has high recognition reliability.

[0151] Based on the importance indicators, set the main driving factors corresponding to time and space.

[0152] In practical applications, the method described in the embodiments of the present invention is applicable to... Figure 3 The identification process is performed on the regions to be identified in the image, and the resulting dragon fruit distribution probability map can be referenced. Figure 4 As shown, it can be seen that by using the method described in the embodiments of the present invention, a multi-scale radiometric consistency system is established by fusing high-resolution nighttime light images of SDGSAT-1 with Black Marble calibrated nighttime light data, taking into account both spatial accuracy and radiometric reliability. This system can more accurately depict the morphological characteristics of dragon fruit growing areas, making them unrestricted by spatial resolution and making it difficult to identify typical supplemental lighting agricultural structures such as strip-shaped and dot-shaped structures.

[0153] Multidimensional information such as nighttime light, vegetation, thermal environment, topography, and urban proximity is incorporated into the model training, and temporal consistency constraints are introduced to ensure that the model has high stability and adaptability in different years and regions.

[0154] Standardized features are constructed at a uniform resolution of 500 m, making them independent of specific regions, achieving the transferability of model structure and parameters, and enabling direct reuse in different countries and regions, thus facilitating cross-scale promotion.

[0155] Reference Figure 5 As shown, a dragon fruit planting area identification system using multi-source nighttime light remote sensing data is described. The system includes a model training device and a model application device, wherein:

[0156] Model training equipment is used for:

[0157] Acquire multi-source remote sensing data of the target area and perform temporal consistency preprocessing on the multi-source remote sensing data to obtain multi-source remote sensing data that is consistent in both spatial and temporal dimensions; the multi-source remote sensing data includes at least nighttime light images, vegetation images, and impervious surface data;

[0158] Identify agricultural light source regions in impermeable surface data and set the features corresponding to the identified agricultural light source regions as spectral screening rules; the spectral screening rules include agricultural light source regions;

[0159] Identify non-seasonal crop areas in vegetation images and set the features of the identified non-seasonal crop areas as vegetation feature filtering rules.

[0160] Identify dot-like or strip-like distribution areas in nighttime light images and set the identified dot-like or strip-like distribution areas as spatial filtering rules;

[0161] Based on spectral screening rules, spatial screening rules, and vegetation feature screening rules, the nighttime light images were identified, and the common areas identified were designated as potential dragon fruit planting areas.

[0162] Multiple features of dragon fruit planting areas were extracted from multi-source remote sensing data corresponding to potential dragon fruit planting areas.

[0163] A model for identifying dragon fruit growing areas was obtained by modeling using a random forest regression model, and training samples were set.

[0164] A dragon fruit growing area identification model was trained using multiple features as input variables and the labels of training samples as output variables.

[0165] Model application devices are used for:

[0166] Acquire multi-source remote sensing data of the area to be identified and obtain multiple features corresponding to the area to be identified;

[0167] By using the trained dragon fruit planting area identification model and multiple features corresponding to the area to be identified, the area to be identified is identified, and a dragon fruit distribution probability map is obtained.

[0168] Optionally, acquiring multi-source remote sensing data of the target area and performing temporal consistency preprocessing on the multi-source remote sensing data includes:

[0169] Acquire multi-source remote sensing data with the same time series;

[0170] Spatial clipping and projection unification of multi-source remote sensing data are performed to obtain spatially identical multi-source remote sensing data.

[0171] For multi-source remote sensing data with the same time and space, cloud mask removal, missing value repair and normalization are performed.

[0172] Optionally, the model training device is used for:

[0173] Based on the impermeable surface data, agricultural light source areas and non-agricultural light source areas were obtained from the multi-source data. Non-agricultural light source areas include urban areas, roads, and industrial areas.

[0174] Obtain the warm light characteristics of the dragon fruit supplemental lighting;

[0175] Based on warm light characteristics and night light images, target areas with red light intensity greater than green light and average brightness higher than the threshold are selected from the agricultural light source areas.

[0176] Identify lattice-like or strip-like distribution areas in the target region by using local density and brightness variance;

[0177] By using the annual mean and annual standard deviation of vegetation images, seasonal crop areas in the target region are screened out to obtain potential dragon fruit planting areas, and labels are set for these potential dragon fruit planting areas.

[0178] Optionally, the model training device is used for:

[0179] Resampling of multi-source remote sensing data;

[0180] Feature extraction is performed on the resampled multi-source remote sensing data to obtain multiple features;

[0181] Among these features are at least the following: nighttime light characteristics, vegetation characteristics, thermal environment characteristics, topographic features, and urban proximity characteristics.

[0182] Optionally, the model training device is used for:

[0183] A random forest regression model was used for modeling, and the model parameters were set as follows: number of decision trees n, minimum number of leaf node samples 1, and out-of-bag sampling ratio 0.7.

[0184] The input to the dragon fruit growing area identification model is multiple feature values ​​corresponding to the area to be identified, and the output is the label of the area to be identified.

[0185] Optionally, the model training device is used for:

[0186] From multi-source remote sensing data, areas with concentrated distribution of dragon fruit plants were selected as training samples;

[0187] Extract multiple features from the training samples and set labels for the training samples;

[0188] The training samples are divided into training set and validation set according to the ratio.

[0189] Optionally, the model training device is used for:

[0190] For a single training sample in the training set, the training process includes:

[0191] Input multiple features of a single training sample into the dragon fruit growing area recognition model, and output the label of the single training sample;

[0192] After all training samples have been trained, the importance indices of multiple features are calculated.

[0193] Optional model training equipment is used for:

[0194] Set importance indicators corresponding to the region;

[0195] Based on the importance index, multiple feature weights corresponding to the region are generated, and each feature weight corresponds to a specific feature.

[0196] Optionally, the model training device is used for:

[0197] Calculate the ROC curve and AUC value based on the validation set;

[0198] Based on the importance indicators, set the main driving factors corresponding to time and space.

[0199] All of the above-mentioned optional technical solutions can be combined in any way to form optional embodiments of the present invention, and will not be described in detail here.

[0200] It should be noted that the apparatus and system provided in the above embodiments are only illustrated by the division of the above functional modules when executing the corresponding methods. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the apparatus and system can be divided into different functional modules to complete all or part of the functions described above. In addition, the methods, apparatus and system embodiments provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiments, which will not be repeated here.

[0201] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware, or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0202] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying dragon fruit growing areas using multi-source nighttime light remote sensing data, characterized in that, The method includes: Acquire multi-source remote sensing data of the target area and perform temporal consistency preprocessing on the multi-source remote sensing data to obtain multi-source remote sensing data that is consistent in both spatial and temporal dimensions; the multi-source remote sensing data includes at least nighttime light images, vegetation images, and impervious surface data; Agricultural light source regions are identified in the impermeable surface data, and the features corresponding to the identified agricultural light source regions are set as spectral screening rules; the spectral screening rules include agricultural light source regions. Identify non-seasonal crop areas in the vegetation image and set the features of the identified non-seasonal crop areas as vegetation feature filtering rules; Identify dot-like or strip-like distribution areas in the night-light image and set the identified dot-like or strip-like distribution areas as spatial filtering rules; Based on the spectral screening rules, the spatial screening rules, and the vegetation feature screening rules, the nighttime images are identified, and the common areas identified are identified as potential dragon fruit planting areas. Multiple features of the dragon fruit planting area are extracted from the multi-source remote sensing data corresponding to the potential dragon fruit planting area; A model for identifying dragon fruit growing areas was obtained by modeling using a random forest regression model, and training samples were set. Using the aforementioned features as input variables and the labels of the training samples as output variables, the dragon fruit planting area identification model is trained. The trained dragon fruit growing area identification model is used to identify the area to be identified, and a dragon fruit distribution probability map is obtained.

2. The method according to claim 1, characterized in that, The process of acquiring multi-source remote sensing data of the target area and performing time-series consistency preprocessing on the multi-source remote sensing data includes: Acquire multi-source remote sensing data with the same time series; Spatial cropping and projection unification are performed on the multi-source remote sensing data to obtain spatially identical multi-source remote sensing data. For multi-source remote sensing data with the same time and space, cloud mask removal, missing value repair and normalization are performed.

3. The method according to claim 2, characterized in that, The identification of agricultural light source areas in the impermeable surface data includes: Based on the impermeable surface data, the agricultural light source area and the non-agricultural light source area in the multi-source data are obtained. The non-agricultural light source area includes urban areas, roads, and industrial areas. The identification of non-seasonal crop areas in the vegetation image includes: By using the annual mean and annual standard deviation of vegetation images, seasonal crop areas in the target region are screened out to obtain the potential dragon fruit planting areas, and labels are set for the potential dragon fruit planting areas. The identification of dot-like or strip-like distribution regions in the night-luminescent image includes: Obtain the warm light characteristics of the dragon fruit supplemental lighting; Based on the warm light characteristics and the night light image, target areas with red light intensity greater than green light and average brightness higher than a threshold are selected from the agricultural light source areas. The dot-matrix or strip-shaped distribution areas in the target region are identified by local density and brightness variance.

4. The method according to claim 3, characterized in that, The extraction of multiple features of the dragon fruit planting area from the multi-source remote sensing data corresponding to the potential dragon fruit planting area includes: The multi-source remote sensing data is resampled; Feature extraction is performed on the resampled multi-source remote sensing data to obtain the multiple features; Among them, the multiple features include at least the night-light feature, vegetation feature, thermal environment feature, topographic feature, and urban proximity feature.

5. The method according to claim 4, characterized in that, The modeling process using a random forest regression model to obtain a dragon fruit planting area identification model includes: A random forest regression model was used for modeling, and the model parameters were set as follows: number of decision trees n, minimum number of leaf node samples 1, and out-of-bag sampling ratio 0.

7. The input to the dragon fruit planting area identification model is multiple feature values ​​corresponding to the area to be identified, and the output is the label of the area to be identified.

6. The method according to claim 4, characterized in that, The training samples are set up as follows: From the multi-source remote sensing data, areas with concentrated distribution of dragon fruit plants were selected as training samples; Extract multiple features from the training samples and set labels for the training samples; The training samples are divided into a training set and a validation set according to a certain ratio.

7. The method according to claim 6, characterized in that, The process of training the dragon fruit growing area identification model using the multiple features as input variables and the labels of the training samples as output variables includes: For a single training sample in the training set, the training process includes: Input multiple features of the single training sample into the dragon fruit growing area recognition model, and output the label of the single training sample; After all training samples have been trained, the importance index of the multiple features is calculated.

8. The method according to claim 7, characterized in that, The method further includes: Set importance indicators corresponding to the region; Based on the importance index, multiple feature weights corresponding to the region are generated, and the multiple feature weights correspond one-to-one with the multiple features.

9. The method according to claim 8, characterized in that, The method further includes: Calculate the ROC curve and AUC value based on the validation set; Based on the aforementioned importance indicators, key driving factors corresponding to time and space are set.

10. A dragon fruit growing area identification system using multi-source nighttime light remote sensing data, characterized in that, The system includes a model training device and a model application device, wherein: The model training device is used for: Acquire multi-source remote sensing data of the target area and perform temporal consistency preprocessing on the multi-source remote sensing data to obtain multi-source remote sensing data that is consistent in both spatial and temporal dimensions; the multi-source remote sensing data includes at least nighttime light images, vegetation images, and impervious surface data; Agricultural light source regions are identified in the impermeable surface data, and the features corresponding to the identified agricultural light source regions are set as spectral screening rules; the spectral screening rules include agricultural light source regions. Identify non-seasonal crop areas in the vegetation image and set the features of the identified non-seasonal crop areas as vegetation feature filtering rules; Identify dot-like or strip-like distribution areas in the night-light image and set the identified dot-like or strip-like distribution areas as spatial filtering rules; Based on the spectral screening rules, the spatial screening rules, and the vegetation feature screening rules, the nighttime images are identified, and the common areas identified are identified as potential dragon fruit planting areas. Multiple features of the dragon fruit planting area are extracted from the multi-source remote sensing data corresponding to the potential dragon fruit planting area; A model for identifying dragon fruit growing areas was obtained by modeling using a random forest regression model, and training samples were set. Using the aforementioned features as input variables and the labels of the training samples as output variables, the dragon fruit planting area identification model is trained. The model application device is used for: Acquire multi-source remote sensing data of the region to be identified, and obtain multiple features corresponding to the region to be identified; The dragon fruit planting area identification model, trained and based on multiple features corresponding to the area to be identified, is used to identify the area and obtain a dragon fruit distribution probability map.

Images