Malaria high-risk area identification method and device based on segmented arbitrary model

Abstract

The invention discloses a malaria high-risk region identification method and device based on a segmented arbitrary model, and the method comprises the steps: obtaining a meteorological data set of a research region in a preset year, determining a high-risk month of malaria outbreak, obtaining Sentinel-1 data of the research region, building a Sentinel-1 monthly image, and calculating an SWI index; detecting a water body on an SWI graph corresponding to the SWI index to obtain a Sentinel-1 water extraction model, identifying a high-risk area corresponding to a high-risk month based on majority voting and decision fusion rules, and evaluating the accuracy of the high-risk area by adopting a water body mask generated by any segmented model to complete malaria high-risk area identification. According to the method, medium-spatial-resolution Sentinel-1 images summarized monthly for a long time are utilized, rainfall and temperature parameters are used, the process is limited in a high-risk month, the process amount can be reduced, temporary water bodies and small depressions with high risks formed by anopheles larva habitats are effectively detected by adopting a majority voting method of monthly, annual and overall analysis, and the anopheles larva habitats are effectively detected. And the malaria high-risk area identification precision is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of malaria high-risk area identification, and in particular relates to a malaria high-risk area identification method and device based on a segmented arbitrary model. Background Art

[0002] Although malaria has been successfully eradicated in many parts of the world, the risk of transmission still exists in areas where Anopheles mosquitoes inhabit. Targeting Anopheles larvae has been shown to be effective in the fight against malaria. Surface water bodies such as lakes and ponds are the main breeding sites for mosquitoes that carry malaria. By using maps of surface water bodies along with meteorological data such as precipitation and temperature, the presence of mosquito larvae can be predicted. Identifying the most productive water bodies can help improve targeted mosquito larval control.

[0003] Mapping surface waters traditionally involves field observations and recording, which is time-consuming and expensive in terms of human and budgetary resources, especially for large areas. Some researchers have successfully used drone technology to study Anopheles breeding sites. However, their resources and equipment are prohibitive for large-scale water resource mapping. Therefore, satellite imagery is a more practical method that is applicable to a wide range of research areas. Satellite observations have been used for habitat surveys for more than two decades. Mapping water bodies using satellite earth observation technology is a routine practice, providing spatial information for monitoring and management of aquatic habitats. Advances in remote sensing technology allow for rapid extraction of surface water information from data sets. Although useful, optical images are hampered by fog, clouds, and bad weather, limiting data collection to daytime. In addition, due to their temporary nature, they have difficulty accurately locating surface water bodies and Anopheles habitats. Therefore, it is urgent to provide a method and device for identifying high-risk areas for malaria based on segmented arbitrary models to solve the above-mentioned technical problems. Summary of the invention

[0004] In view of this, the present invention provides a method and device for identifying high-risk areas for malaria based on a segmented arbitrary model, which can improve the accuracy of detecting surface water bodies, use precipitation and temperature parameters and limit the process to high-risk months to reduce the process volume, and effectively detect temporary water bodies and small depressions with high risks of Anopheles larvae habitat formation by adopting a majority voting method for monthly, annual and overall analysis, which is specifically achieved by the following technical solutions.

[0005] In a first aspect, the present invention provides a method for identifying high-risk areas for malaria based on a segmented arbitrary model, comprising the following steps: Obtaining a meteorological data set of a study area in a preset year, and determining a high-risk month for malaria outbreak based on the meteorological data set, wherein the meteorological data set includes a long-term series of precipitation parameters and temperature parameters; Acquire Sentinel-1 data of the study area according to the high-risk month and establish a Sentinel-1 monthly image, calculate the SWI index according to the Sentinel-1 monthly image, and detect water bodies on the SWI map corresponding to the SWI index to obtain a Sentinel-1 water extraction model; Based on majority voting and decision fusion rules, high-risk water bodies in the Sentinel-1 water extraction model are identified to obtain high-risk areas corresponding to the high-risk months; The accuracy of the high-risk area is evaluated by using a water body mask generated by a segmented arbitrary model to complete the identification of high-risk areas for malaria, wherein the water body mask includes a plurality of water-containing depressions and a plurality of water-free depressions.

[0006] As a preferred embodiment of the above technical solution, a meteorological data set of a study area in a preset year is obtained, and a high-risk month for malaria outbreak is determined according to the meteorological data set, including: Predetermining a study period for the study area and obtaining a meteorological dataset of malaria transmission risk during the study period; The rainy season suitable for the growth and survival of Anopheles mosquitoes is extracted from the meteorological data set to obtain precipitation parameters and temperature parameters.

[0007] As a preferred embodiment of the above technical solution, before detecting the water body on the SWI map corresponding to the SWI index to obtain the Sentinel-1 water extraction model, it includes: Sentinel-1 images during the study period were acquired through GEE; Temporary water bodies corresponding to the Sentinel-1 images were detected using multi-temporal data.

[0008] As a preferred embodiment of the above technical solution, the high-risk water bodies in the Sentinel-1 water extraction model are identified based on majority voting and decision fusion rules to obtain the high-risk areas corresponding to the high-risk months, including: Three different analyses were used to detect and classify high-risk areas, the first annual analysis identified areas with water during the four study months each year to determine where water storage persisted year-round; the second monthly analysis identified areas with water storage over the years; A four-level classification map was used to label areas with water impoundment every seven years as high risk, areas with water impoundment between four and six years as relatively risky, areas with water impoundment less than four years as low risk, and a fourth category was assigned to areas with no water in any of the study months, where the study period was seven years with preset years and study months.

[0009] As a preferred embodiment of the above technical solution, majority voting and weighted majority voting are used to detect high-risk areas. Majority voting includes each classification map voting for a specific category, and all votes have the same weight; In weighted majority voting, each classification map is assigned a weight based on its perceived reliability or performance, and the votes of each classification map are weighted according to its assigned weight, with the weights determined based on the precipitation in the study month to weigh the voting results.

[0010] As a preferred embodiment of the above technical solution, the water body mask generated by the segmented arbitrary model is used to evaluate the accuracy of the high-risk area to complete the identification of the high-risk area for malaria, including: Use multi-point feature hints to segment water bodies and convert points in geographic data frames into coordinates List; Set SamGeo as a predictor for a single coordinate and predict the features of the single coordinate in the instance to obtain a mask, and save the mask to a unique file for each point; After generating a mask for each point, all masks are merged into a single mosaic raster file, providing a complete representation of all segmented regions from the multi-point feature cues.

[0011] As a preferred embodiment of the above technical solution, obtain downloaded tiles and use tile map service to create georeferenced images, select satellite images for segmentation, and set paths to define SAM models, where SamGeo has been initialized with specified parameters and shapefile has been converted from geographic data frame to point coordinate list; By looping through the list of coordinate points, each point is predicted, segmented and saved into a unique Geotiff file to generate a unique mask for each point; All masks were resampled to 10m resolution and merged into one file as a mosaic mask containing information of all segmented regions; The multiple masks generated by SAM were used to evaluate the accuracy of water bodies extracted from the SWI index to obtain evaluation results, wherein the confusion matrix and comprehensive indicators were used to evaluate the performance of the model, and the comprehensive indicators included precision, recall, F1 score, overall accuracy OA or Kappa.

[0012] As a preferred embodiment of the above technical solution, after calculating the surface water distribution map corresponding to all geographic reference images, the flooding frequency IF of the depression is calculated using the expression: , WOT means water observation time, and TOT means total observation time.

[0013] As a preferred embodiment of the above technical solution, the evaluation results are compared with four classification algorithms to complete the classification of four types of land cover: vegetation, urban areas, wasteland and water bodies. The four classification algorithms include RF model, SVM model, SGTB model and SC model.

[0014] In a second aspect, the present invention further provides a device for identifying high-risk areas for malaria based on a segmented arbitrary model, which is applied to the above-mentioned method for identifying high-risk areas for malaria based on a segmented arbitrary model, comprising: A data acquisition unit, used to acquire a meteorological data set of a study area in a preset year, and determine a high-risk month for malaria outbreak according to the meteorological data set, wherein the meteorological data set includes a long-term series of precipitation parameters and temperature parameters; A model extraction unit, used for acquiring Sentinel-1 data of the study area according to the high-risk month and establishing a Sentinel-1 monthly image, calculating a SWI index according to the Sentinel-1 monthly image, and detecting water bodies on a SWI map corresponding to the SWI index to obtain a Sentinel-1 water extraction model; A region identification unit, used for identifying high-risk water bodies in the Sentinel-1 water extraction model based on majority voting and decision fusion rules to obtain high-risk regions corresponding to the high-risk months; The model evaluation unit is used to evaluate the accuracy of the high-risk area by using a water body mask generated by a segmented arbitrary model to complete the identification of high-risk areas for malaria, wherein the water body mask includes a plurality of water-containing depressions and a plurality of water-free depressions.

[0015] The present invention provides a method and device for identifying high-risk areas for malaria based on a segmented arbitrary model. The method comprises the following steps: obtaining a meteorological data set of a study area in a preset year, determining a high-risk month for malaria outbreak according to the meteorological data set, obtaining Sentinel-1 data of the study area according to the high-risk month and establishing a Sentinel-1 monthly image, calculating a SWI index according to the Sentinel-1 monthly image, detecting water bodies on a SWI map corresponding to the SWI index to obtain a Sentinel-1 water extraction model, identifying high-risk water bodies in the Sentinel-1 water extraction model based on majority voting and decision fusion rules to obtain a high-risk area corresponding to the high-risk month, using a water body mask generated by the segmented arbitrary model to evaluate the accuracy of the high-risk area to complete the identification of high-risk areas for malaria, utilizing medium spatial resolution Sentinel-1 images aggregated monthly over a long period of time, and combining temperature and precipitation data from a meteorological data set. Using precipitation and temperature parameters and limiting the process to high-risk months can reduce the process volume. By adopting the majority voting method for monthly, annual and overall analysis, the proposed method effectively detects temporary water bodies and small depressions with high risk of Anopheles larval habitat formation, improving the accuracy of identifying high-risk areas for malaria. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 A flow chart of a method for identifying high-risk areas for malaria based on a segmented arbitrary model provided by the present invention; Figure 2 A map of surface water bodies produced during the study period provided for the present invention; Figure 3 A physical map of the earth's surface drawn after a comprehensive analysis provided by the present invention; Figure 4 This is an effect diagram of the point feature prompt processing image set provided by the present invention; Figure 5 The effect diagram of the generated mask and the corresponding detected water body provided by the present invention; Figure 6 This is a structural block diagram of the device for identifying high-risk areas for malaria based on a segmented arbitrary model provided by the present invention. DETAILED DESCRIPTION

[0018] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0019] See also Figure 1 The present invention provides a method for identifying high-risk areas for malaria based on a segmented arbitrary model, comprising the following steps: S1: Obtain a meteorological data set for a study area in a preset year, and determine a high-risk month for malaria outbreak based on the meteorological data set, wherein the meteorological data set includes a long-term series of precipitation parameters and temperature parameters; S2: Acquire the Sentinel-1 data of the study area according to the high-risk month and establish a Sentinel-1 monthly image, calculate the SWI index according to the Sentinel-1 monthly image, detect water bodies on the SWI map corresponding to the SWI index, and obtain a Sentinel-1 water extraction model; S3: Identify the high-risk water bodies in the Sentinel-1 water extraction model based on majority voting and decision fusion rules to obtain the high-risk areas corresponding to the high-risk months; S4: Using a water body mask generated by a segmented arbitrary model to evaluate the accuracy of the high-risk area to complete the identification of malaria high-risk areas, wherein the water body mask includes a plurality of water-containing depressions and a plurality of water-free depressions.

[0020] In this embodiment, a meteorological data set of a study area in a preset year is obtained, and a high-risk month for malaria outbreak is determined based on the meteorological data set, including: presetting a study period for the study area, and obtaining a meteorological data set of malaria transmission risk during the study period; extracting a rainy season suitable for the growth and survival of Anopheles mosquitoes from the meteorological data set to obtain precipitation parameters and temperature parameters. Before detecting water bodies on the SWI map corresponding to the SWI index to obtain the Sentinel-1 water extraction model, it includes: obtaining a Sentinel-1 image during the study period through GEE; and using multi-temporal data to detect temporary water bodies corresponding to the Sentinel-1 image. Based on majority voting and decision fusion rules, high-risk water bodies in the Sentinel-1 water extraction model are identified to obtain high-risk areas corresponding to the high-risk months, including: three different analyses are used to detect and classify high-risk areas. The first annual analysis determines the areas with water in four study months each year to determine the locations of continuous water storage throughout the year; the second monthly analysis determines the areas with water storage for many years; a four-level classification map is used to mark areas with water storage once every seven years as high-risk areas, areas with water storage for four to six years as relatively dangerous areas, areas with water storage for less than four years as low-risk areas, and the fourth category is assigned to areas without water in any study month, where the study period is seven years, with preset years and study months.

[0021] Optionally, majority voting and weighted majority voting are used to detect high-risk areas, where majority voting includes each classification map voting for a specific category and all votes have the same weight; In weighted majority voting, each classification map is assigned a weight based on its perceived reliability or performance, and the votes of each classification map are weighted according to its assigned weight, with the weights determined based on the precipitation in the study month to weigh the voting results.

[0022] In this embodiment, the water body mask generated by the segmented arbitrary model is used to evaluate the accuracy of the high-risk area to complete the identification of the high-risk area for malaria, including: using multi-point feature prompts to segment the water body and converting the points in the geographic data frame into coordinates list; set SamGeo as the predictor for a single coordinate and predict the features of that single coordinate in the instance to obtain a mask, saving the mask to a unique file for each point; after generating the mask for each point, all masks are merged into a mosaic raster file, providing a complete representation of all segmented regions from the multi-point feature hints.

[0023] Get downloaded tiles and create georeferenced images using tile map services, select satellite images for segmentation, and set paths to define the SAM model, where SamGeo has been initialized with specified parameters and shapefiles have been converted from geographic data frames to point coordinate lists; predict, segment, and save each point into a unique Geotiff file by looping through the list of coordinate points to generate a unique mask for each point; resample all masks to 10m resolution and merge them into one file as a mosaic mask containing information about all segmented regions; use multiple masks generated by SAM to evaluate the accuracy of water bodies extracted from the SWI index to obtain evaluation results, where confusion matrices and comprehensive indicators are used to evaluate the performance of the model, and the comprehensive indicators include precision, recall, F1 score, overall accuracy OA or Kappa.

[0024] It should be noted that after calculating the surface water distribution map corresponding to all georeferenced images, the inundation frequency IF of the depression is calculated using the expression: , WOT represents the water observation time, and TOT represents the total observation time. That is, after calculating the surface water distribution map corresponding to all georeferenced images, the corresponding calculation of the flooding frequency IF of the depression is: , WOT represents water observation time, and TOT represents total observation time. The evaluation results are compared with four classification algorithms to complete the classification of four types of land cover, namely vegetation, urban areas, wasteland, and water bodies. The four classification algorithms include RF model, SVM model, SGTB model, and SC model.

[0025] Among them, regarding the identification of mosquito-producing water bodies and aquatic habitats with large seasonal differences, remote sensing images can effectively map the spatial variation of transmission risks, which is crucial for effective larval source management due to the short life cycle of mosquitoes (10-14 days). The temporary nature of almost all depressions makes reference datasets such as the Global Surface Water Explorer (GSWE) dataset unhelpful. A simple and effective solution is to use an index. The Modified Normalized Difference Water Index (MNDWI) is a spectral water index derived from optical images, which can reliably identify water bodies based on exceeding a specific threshold. It is necessary to dynamically monitor water surface areas based on SAR data using easily operational indicators such as MNDWI. A new index based on Sentinel-1 is introduced, which uses a multivariate stepwise regression analysis called the SWI-based Water Extraction Model (SWIM). SWIM is a simple but robust model for extracting water areas from Sentinel-1 images. SWI (Sentinel Water Index) has advantages over other machine learning models such as speed and efficiency. When applied to different regions at different times, the SWI threshold classification method showed high quantitative accuracy (about 99%) and overall accuracy (about 97%).

[0026] Some studies have been conducted on using GEE to make temporal or spatial predictions of habitats and disease high-risk areas. GEE used a series of environmental factors that affect the growth of Anopheles mosquitoes in a certain area over a seven-year period to estimate the high-risk period. In addition, compared with using raw time series data, using the median as a temporal aggregation method in GEE can effectively reduce the amount of data while maintaining high accuracy. The processing tasks of the study were all completed through a JavaScript application programming interface (API).

[0027] Among them, small and large depressions are mainly temporary water bodies affected by precipitation, evaporation and hydrological systems. The proposed method identifies high-risk areas that may become Anopheles habitats. Months with high risk of malaria outbreaks are identified by analyzing long-term series of precipitation and temperature parameters, and the SWI index is applied separately to Sentinel-1 monthly images. The results of each month are merged using majority voting and decision fusion rules to take advantage of the strengths of each classification map. The results are evaluated using masks generated by the segmented arbitrary model (SAM) and compared with other methods.

[0028] Specifically, most malaria cases occurred in Provinces A and B, and southern parts of Provinces C and D in the southeastern corner of the region. According to the Köppen-Geiger climate classification, the region belongs to the hot desert climate category (BWh). Although the annual rainfall is about 110 mm, the region is prone to heavy rains. An unexpected malaria outbreak in 2021 continued into 2022, with ten times the number of cases in 2021, with an estimated total of 3,000 cases by November 2022. Factors contributing to the outbreak included heavy rains and insufficient case detection. The datasets used in this study include Sentinel-1 imagery, PERSIANN-CDR precipitation data, and temperatures extracted from meteorological datasets.

[0029] Specifically, Sentinel-1, launched by a space agency in 2014 and 2016, is a polar-orbiting satellite providing dual-polarization data in the C-band. It provides a revisit time of 6 days via two satellites. 79 Sentinel-1 tiles are provided monthly for Provinces A and B, a number that has increased to 2212 tiles in four months over seven years of analysis, making GEE the best choice for handling such a large area processing load. The data are collected in GEE as “COPERNICUS / S1_GRD” and include the Interferometric Wideband (IW) mode of the Ground Based Sounding (GRD) scene. GEE developers have applied preprocessing steps including thermal noise removal, radiometric calibration, and terrain correction, taking advantage of the orthogonal corrected backscatter coefficient product of the individual VV and VH channels of Sentinel-1 in ascending and descending orbits at a spatial resolution of 10 m, and the only remaining preprocessing is the removal of speckle noise using a median filter with a kernel size of 3×3.

[0030] PERSIANN-CDR provides a 40-year precipitation dataset, providing daily near-global data at a resolution of 0.25°, making it valuable for global climate research, especially extreme weather events. Unlike the PERSIANN technique, PERSIANN-CDR uses an ANN model trained with preset parameters to estimate historical data. The dataset can detect water bodies using the monthly PERSIANN-CDR dataset from 2017 to 2023. A seven-year time series of meteorological data obtained from the Regional Meteorological Organization (IMO) has been used to consider the temperature factor, which includes daily average temperature summary data measured at weather stations in Provinces A and B. The monthly average temperature is then calculated based on the average temperature of each day. The operation was performed for seven consecutive years from 2017 to 2023.

[0031] The present invention mapped water bodies in Provinces A and B in a certain region over a period of seven years to identify high-risk areas for Anopheles breeding. To predict potential breeding sites, the study period was first extracted, and focusing on the high-risk period of the year, a seven-year time series was selected to balance capturing long-term environmental changes. Seven years is long enough to include different climate conditions, seasonal patterns, and abnormal weather events, thereby enhancing the robustness and generalizability of the model. This duration helps to identify consistent trends and patterns in parameters that are critical for accurately detecting high-risk mosquito habitats. In contrast, using shorter time series may not capture the full range of climate change, resulting in less reliable predictions and possible neglect of long-term trends. Overall, the seven-year period provides a practical compromise and provides sufficient data to build a reliable model for effective mosquito habitat detection.

[0032] The use of effective environmental parameters helps to accurately predict the time of malaria outbreaks. Precipitation and temperature are key factors that increase the number of Anopheles mosquitoes, affect larval and mosquito populations, and the risk of malaria transmission. Therefore, the monthly PERSIANN-CDR dataset from 2017 to 2023 is used to determine the rainy season for analysis. Temperature is another key environmental factor, from lowest to highest. These temperature limits are called the lower and upper developmental temperatures (LDT and UDT). Anopheles eggs require temperatures between 15°C and 35°C and an average of 611 degree days to mature into adults. The degree-day model means that the temperature must be kept within the appropriate range and the cumulative temperature for several consecutive days must reach 611 degree days. Higher temperatures accelerate the growth rate of Anopheles larvae and shorten their maturation time. To take this factor into account, the average daily temperature of the seven IMO stations in the study area was used to determine the monthly average temperature. Finally, by summarizing the precipitation and temperature monitoring results from 2017 to 2023, the rainy season with a temperature suitable for the growth and survival of Anopheles mosquitoes, such as 15-35°C, was extracted, and the images corresponding to this period were used for the water extraction model.

[0033] After the study period was determined, Sentinel-1 imagery needed to be acquired. Using multi-temporal data allowed for improved detection of temporary water bodies and reduced speckle noise. A temporal aggregation method was implemented instead of using time series images directly. Therefore, during the study period, the median operator of this method was used to acquire Sentinel-1 imagery through GEE. Areas with the potential for ponding during high-risk periods, exceeding a threshold of 611 degrees, can serve as habitats for Anopheles larvae. After calculating the SWI index and extracting water bodies monthly over seven years (2017-2023), the maps were fused at the decision level to identify high-risk areas. These areas provide the necessary surface water for the growth and development of Anopheles.

[0034] Three different analyses were used to detect and classify high-risk areas. The first annual analysis identified areas that had water in all four study months of each year to identify locations with continuous water storage throughout the year. The second monthly analysis identified areas with water storage for many years. Using a four-level classification map, areas with water storage once every seven years were designated as high-risk areas, areas with water storage for four to six years were labeled as relatively dangerous areas, and areas with water storage for less than four years were labeled as low-risk areas. The fourth category was assigned to areas that had no water in any of the study months. This analysis aimed to identify high-risk areas with continuous water storage for many years, indicating its long-term durability. The third analysis used all the generated maps to map high-risk water bodies within the study area. A hierarchical approach was used to detect high-risk areas using majority voting and weighted majority voting. The most popular method was majority voting, where each classification map voted for a specific class, all votes were equally weighted, and the final label was the most common label in the classifier output. Majority voting was used for monthly and annual analyses. However, for the overall analysis, weighted majority voting is influenced by the number of votes cast and the assigned weights. In weighted majority voting, each classified map is assigned a weight based on its perceived reliability or performance. The final decision is made by majority vote, with votes for each map weighted according to its assigned weight.

[0035] Votes are weighed using the amount of precipitation in each month to determine weights, which directly indicates potential waterlogging. By assigning weights based on precipitation levels, periods with higher rainfall are given more importance, which is more likely to result in waterlogging, allowing the model to adapt to temporal variations in precipitation, making it more adaptable to real environmental conditions. Focusing on high precipitation periods helps filter out noise from low or no precipitation months, reducing false positives. Therefore, precipitation levels, which reflect sensitivity and specificity, are reliable indicators of algorithm performance, with high sensitivity and specificity enhancing the impact of decisions. On the other hand, this approach may bias detection towards rainy seasons, potentially underestimating the risk of dry periods, as small-scale waterlogging remains problematic. Abnormal weather patterns, such as unseasonal rainfall, may also have a disproportionate impact on the model's output if not managed properly. In addition, the effectiveness of the weighted voting system depends on the accuracy of precipitation data, as inaccurate data can lead to incorrect weights and flawed predictions. These contributions are calculated by dividing monthly precipitation by the sum of the corresponding monthly precipitation.

[0036] Specifically, a satellite dataset with a spatial resolution of 0.5 meters from the SkySat ESA archive was used to mask the water bodies. For the masks, the classes were divided into separate classifications (“depression” vs. “background”). The geospatial analysis was performed with the help of a custom tool “SamGeo”, which is an integral part of the original module. SAM has several models available, which have different computational requirements and their underlying architectures. The ViT-H SAM model, which is the most advanced and complex model currently available, was used to bring most of the SAM capabilities to our operations. SAM provides various hints, and the present invention utilizes the multi-point feature hint for water body segmentation. This hint converts the points in the geographic data frame into a list of coordinates [x,y]. SamGeo acts as a predictor for the individual coordinates, predicting their features in the instance and saving the resulting mask to a unique file for each point. After generating the mask for each point, all masks are merged into a mosaic raster file, providing a complete representation of all segmented areas from the multi-point hints. The centroids of the 260 water bodies were initially located as points in Google Earth Pro and then exported as shapefiles to mask the water bodies. These points were selected based on visual interpretation because during the dry season, these depressions show unique characteristics due to the structure of silty clay layers caused by soil erosion caused by runoff, as observed during the field assessment. On the other hand, during the rainy season, these places are prone to water accumulation. Of the 260 depressions, 31 were observed in the field.

[0037] The satellite image was selected for segmentation by downloading tiles and creating a large georeferenced image using the tile map service (TMS2GeoTIFF); then the model of SAM was defined by setting the path, SamGeo was initialized with the specified parameters, and the shapefile was converted from a geographic data frame to a list of point coordinates; during the prediction process, the coordinate reference system was defined and set to "EPSG:4326"; subsequently, by looping through the list of coordinate points, each point was predicted, segmented, and saved into a unique Geotiff file. This process generated a unique mask for each point. Finally, all masks were resampled to 10m resolution and merged into one file as a mosaic mask containing information on all segmented areas. All parameters defined in the SAM model are listed in Table 1.

[0038] Table 1: SAM model parameters

[0039] All SAM-related processes were completed using the code available on GitHub: AI Remote Sensing Repository, which was modified according to the case. Masks of 260 depressions generated by SAM were used to evaluate the accuracy of water volume extracted from the SWI index. In addition, 260 samples representing non-water areas were randomly selected in Google Earth Pro. To evaluate the performance of the model, confusion matrices (Table 2) and comprehensive metrics such as precision, recall, F1 score, overall accuracy (OA), and Kappa were used. In addition, pixel-level accuracy criteria, especially the intersection of union (IoU) commonly used in object detection, were considered for depression detection validation.

[0040] Table 2: Confusion matrix for binary classification

[0041] In another evaluation, in order to better understand and know the spatiotemporal distribution characteristics of water bodies, the inundation frequency (IF) of depressions was calculated using the above equation after calculating the surface water distribution map for all 28 images. The IF also reflects the topography, and lower areas are more prone to flooding. In order to evaluate the performance of the proposed method, the results were compared with four classification algorithms: Random Forest (RF), Support Vector Machine (SVM), Smile Gradient Tree Boost (SGTB), and Smile Cart (SC). For this purpose, 4 land cover categories of vegetation, urban areas, wasteland, and water bodies were classified by using each of the above machine learning models. The training samples associated with the water body category were the samples located in Google Earth Pro.

[0042] See also Figure 6 The present invention also provides a device for identifying high-risk areas for malaria based on a segmented arbitrary model, which is applied to the above-mentioned method for identifying high-risk areas for malaria based on a segmented arbitrary model, comprising: A data acquisition unit, used to acquire a meteorological data set of a study area in a preset year, and determine a high-risk month for malaria outbreak according to the meteorological data set, wherein the meteorological data set includes a long-term series of precipitation parameters and temperature parameters; A model extraction unit, used for acquiring Sentinel-1 data of the study area according to the high-risk month and establishing a Sentinel-1 monthly image, calculating a SWI index according to the Sentinel-1 monthly image, and detecting water bodies on a SWI map corresponding to the SWI index to obtain a Sentinel-1 water extraction model; A region identification unit, used for identifying high-risk water bodies in the Sentinel-1 water extraction model based on majority voting and decision fusion rules to obtain high-risk regions corresponding to the high-risk months; The model evaluation unit is used to evaluate the accuracy of the high-risk area by using a water body mask generated by a segmented arbitrary model to complete the identification of high-risk areas for malaria, wherein the water body mask includes a plurality of water-containing depressions and a plurality of water-free depressions.

[0043] In this example, four high-risk periods were identified by combining mean monthly precipitation and temperature data, and during these months, areas with waterlogging potential were identified using the SWI index on Sentinel-1 imagery. The overall analysis was evaluated using an object-based approach using masks generated by the SAM model. Furthermore, our approach outperformed other classifiers, including machine learning models, in terms of accuracy. It was applied to Sentinel-1 imagery for the selected study period between 2017 and 2023. Figure 2 The final classification results of a small area of ​​SRVN at a spatial resolution of 10 meters in the high-risk month of 2017 are shown, where Figure 2 Maps (blue) of a small portion of the southeastern part of S City using the SWI method during the 2017 study period: (a) March 2017; (b) April 2017; (c) May 2017; and (d) July 2017.

[0044] As mentioned above, during high-risk times of the year, depressions with the potential for stagnant water pass the 611 degree threshold and can become habitat for Anopheles larvae. Three different analyses were performed to detect high-risk areas for forming larval habitats. The results are shown below: 1) Monthly analysis This analysis was performed to identify areas with water in the four months from 2017 to 2023. Therefore, it was necessary to extract the areas that were commonly identified as water in the classification output maps for March, April, May, and July. For this purpose, the logical operator of intersection (AND operator) was used to perform this analysis. Results of the analysis of a small area of ​​S City, where the detection of high-risk areas formed by the habitat of Anopheles larvae was analyzed monthly. The areas depicted in blue are common areas with water in all four months of the year studied, and high-risk areas in a small area of ​​the southeastern part of S City: (a) 2017; (b) 2018; (c) 2019; (d) 2020; (e) 2021; (f) 2022; (g) 2023.

[0045] 2) Annual analysis In the second analysis, the areas with water were classified according to the frequency of occurrence during the study period, and the classification results of some areas in City S. The process of this classification is that areas with water in all seven years of the study are classified as dangerous areas if they have water for four to six years out of the seven years, and are classified as medium-risk areas, while areas with water for less than four years are classified as low-risk areas.

[0046] 3) In addition to the monthly and annual analyses, the third analysis also considered all 28 classification maps for four months in the seven-year study. For this purpose, weighted majority voting was used to calculate weights based on precipitation. The output maps of this analysis were used to evaluate accuracy. The analysis results for a small area of ​​S City are shown in Figure 2. Figure 3 Blue areas indicate depressions with water, based on an overall analysis of the 28-month study results.

[0047] The accuracy validation included evaluation using SAM, IF coefficients, and classification algorithms. 260 masks were generated using a point-based approach by performing SAM on marked points representing high potential areas of Anopheles breeding habitat. By focusing on a singular point, SAM was able to provide accurate segmentation results. These points acted as precise spatial cues sufficient for the model to accurately identify and segment depressions. This approach may be particularly relevant to applications that require accurate identification and segmentation of individual objects in an image. Figure 4 The results of masked depressions for some samples in the northern part of S City are shown. After generating the masks, they were resampled to 10 meters, which is the same as the mapping results of Sentinel-1 images, where Figure 4 Illustrations of images representing three samples in the northern part of S city processed using point hints. The first column shows a Google Earth image with the selected points as the centroid of the depression. The second column shows the prediction results of SAM. In order to accurately verify the proposed method and make a meaningful and reliable quantitative analysis of the final map, the masks generated by the SAM model and the sampling points in the non-water body area are superimposed on the output map of the weighted majority voting method. The samples of depression detected are shown in Figure 5 As shown, Figure 5 Representation of the generated masks and the corresponding detected water bodies: (a) Google EarthPro image used to create the mask; (b) SAM masked depression; (c) detection of water bodies using the proposed method. Then, the confusion matrix is ​​obtained by comparing the prediction results with the reference mask. Six indicators are calculated based on the confusion matrix, and the results are shown in Table 3.

[0048] Table 3: Detection accuracy of the proposed method

[0049] Among them, the intensity-frequency (IF) distribution map corresponding to the monthly SAR data of a small area in the study area is presented. The water body detection performance of the proposed method is compared with the other four classification algorithms RF, SVM, SGTB and SC by implementing land cover classification. The classification results are evaluated using 520 collected samples.

[0050] Table 4: Summary of the detection results of all classifiers

[0051] By comparing the results shown in Table 3 and Table 4, the Kappa of the proposed method is 17% higher on average compared to RF, SVM, GTB and SC.

[0052] In addition, January, March, April, May, July and October had rainfall in at least four of the seven years and were significantly wetter than the other months. Although some months had rainfall in consecutive years, these months had the highest rainfall. As mentioned earlier, temperatures between 15°C and 35°C are most suitable for the growth of Anopheles larvae. Most Anopheles species cannot complete their life cycle and survive below 15°C, while the growth of immature stages to adults is halted at temperatures above 35°C. High temperatures also reduce the stability of pits and depressions after heavy rains, leading to surface evaporation and drying of the soil, which is not conducive to the growth of Anopheles larvae. In all seven years, the average temperatures from March to November were within the ideal range.

[0053] To select the study period, we analyzed the trend of monthly contributions to annual precipitation and temperature. In most years, March, April, May, and July were favorable for the formation of Anopheles larval habitat in terms of precipitation and temperature. Under suitable precipitation and temperature conditions, aquatic habitats became the main sites for the growth and development of Anopheles larvae after reaching 611 degree days, eventually leading to the disease peak. In addition, the authorities noticed two separate epidemic peaks in the study area, the first peak occurred in late winter to late spring, and the second peak occurred in midsummer to mid-autumn, which coincided with the selected months. Therefore, the study period was limited to four months (March, April, May, and July) from 2017 to 2023, which are necessary conditions for the formation of larval habitats. However, the growth and larvalization processes stopped in cases where the maximum temperature reached 35°C or higher.

[0054] The present invention visually demonstrates that the method effectively delineates water areas, highlighting depressions with high water retention potential (e.g. Figure 2). The ability to identify these areas depends on factors such as the water storage capacity of the depression, the amount of water, the size of the depression, the slope of the area, and the connectivity of the waterways. The index also displays detailed information due to its fine spatial resolution. Comparing the water maps extracted for different months shows that the changes in surface water cover are affected by precipitation, evaporation, and infiltration. In addition, depressions persisted in all four studied months of the year, with blue areas indicating high-risk areas for that year. The results show that larger and deeper depressions retain water. The presence of water in these depressions in consecutive years may be due to runoff filling the depressions after rainfall or accumulation of water from previous rainfall. The lifespan of water in depressions depends on factors such as depression area and volume, evaporation rate, and infiltration rate. Longer water retention time provides suitable conditions for the formation and growth of Anopheles habitats, after reaching the 611 degree threshold. Comparing the maps from different years shows that smaller depressions have greater changes, which may be due to different rainfall and subsequent water flow and accumulation in surrounding depressions. Therefore, these specific areas have the highest risk for Anopheles habitat formation in the annual survey.

[0055] Small depressions (shown in blue) store water less frequently (less than 4 years), while larger depressions are seen as water bodies more frequently. The frequent presence of water in larger depressions indicates a high risk, as they have a greater capacity to retain water and can fill up with the first rainfall in the risk month. Additionally, red areas, representing high-risk areas, appear in deeper boreholes where runoff is directed when rainfall occurs. These deep areas are the first to fill up even when rainfall is low, keeping them consistently wet and conducive to the growth of Anopheles mosquitoes. Surrounding the red areas (high risk level) are green areas (medium risk level), representing shallower parts of the depressions that can store water for 4 to 6 years. These areas are also likely to be covered with water due to relatively high rainfall, which can help health officials prioritize high-risk areas for monitoring and control.

[0056] The results highlight the consistent detection of larger high-risk depressions as water bodies, demonstrating the effectiveness of the approach in identifying important water features associated with potential risk (e.g. Figure 3). However, a refinement was observed in this analysis, where smaller depressions (usually caused by low precipitation and widely distributed throughout the region) were systematically excluded and not classified as water bodies. This subtle refinement improves the accuracy of water detection by filtering out insignificant depressions, resulting in a more accurate representation of water bodies in the study area. The map of the overall analysis will be used for accuracy assessment. In addition, the spatial distribution of IF highlights the water storage capacity of depressions based on the time and location of water in the hydrological system. The results of this index confirm the analysis we conducted previously. Areas with IF values ​​close to 100 (shown in red) indicate consistent water cover throughout the study period, and they are highly suitable habitats for Anopheles mosquito larvae due to their water retention capacity. In contrast, non-water areas (blue areas with IF values ​​close to 0) do not store water and therefore cannot support Anopheles mosquito habitat. In addition, some areas (bright blue, white, or bright red) were flooded for several months during the study period. Although not as high risk as the red dots, these areas may become potential habitats as precipitation rates increase and the hydrological system changes. The low IoU accuracy of 34.1% can be attributed to several factors in the classification and evaluation process. First, the complexity of the classified objects can significantly affect the accuracy of the classification. The sensitivity of the IoT metric to the boundaries of objects should be carefully considered as it plays a crucial role in evaluating the accuracy of boundary delineation. The amount of water in a depression depends on the precipitation and the depression structure. Each depression may contain several concave points, which can be visually compared to a mine pit. This indicates that these depressions contain multiple cavities and they are more likely to accumulate runoff than other areas. This leads to a reduction in the IoT metric: (a) Overview; (b) Ground Profile. After rainfall, runoff fills these points first due to the slope. Increased rainfall can cause more flooding in other areas of the depression. Evaporation and permeability also affect the amount of water. In the SAM model, the entire depression is considered as a water body and the depression is fully extracted, but since some holes may not be filled with water, it is difficult to align the SAM-generated ground truth mask with the predicted water body, which reduces the intersection area and IoU value.

[0057] Specifically, the model achieved 98.7% precision and 93.4% recall, demonstrating its accuracy in effectively identifying water bodies. This balance ensures accurate extraction while minimizing false positives, which is critical for reliable water resource monitoring. The overall performance of the model is strong, as confirmed by the high F1 score, with an overall accuracy of 96.1% and a kappa coefficient of 92.2%. Key factors contributing to this success include the suitability of Sentinel-1 imagery for distinguishing water bodies from other land cover and the robustness of the classification model. Object-based assessment and the model's ability to adapt to temporal changes in hydrodynamics further improve its performance. Limitations in image resolution, especially for smaller depressions, can lead to misclassification. Addressing this issue may involve higher-resolution data sources or post-processing techniques. However, the synergy of suitable satellite data and a well-optimized model allows for extremely high accuracy in delineating water bodies using Sentinel-1 imagery.

[0058] Freely available radar imagery and environmental parameters from meteorological datasets were utilized, and all processing was done in cloud-based services such as GEE to efficiently manage extensive time series analysis, eliminating the need for physical data storage and processing resources. In addition, it is possible to significantly reduce the need for extensive fieldwork and labor operations, and by utilizing medium-resolution data, high-risk areas over large, widespread areas can be identified. By using long-time series of Sentinel-1 data combined with environmental parameters such as precipitation and temperature, the kappa coefficient is consistent with Ovakoglou's study of non-vegetated wetlands and rice fields. As an alternative to machine learning models that have not shown good accuracy, our approach is a flexible solution that demonstrates its efficiency in detecting water in a wide range of study areas and contributes to the overall progress in identifying Anopheles larval habitats. Mapping of water bodies should never be relied upon alone to characterize malaria risk and develop disease control strategies, instead, the relationship between the location of identified habitats and human population density should be carefully considered.

[0059] Remote sensing has been widely used in disease control and prevention over the past two decades, including for malaria. The proposed method differs from earlier studies in that it utilizes medium spatial resolution Sentinel-1 imagery aggregated monthly over a long period of time and combines it with temperature and precipitation data from a meteorological dataset. The amount of the process can be reduced by using precipitation and temperature parameters and limiting the process to high-risk months. By adopting a majority voting approach for monthly, annual, and overall analyses, the proposed method effectively detects temporary water bodies and small depressions with a high risk of Anopheles larval habitat formation. To overcome the challenges associated with limited training data, this approach provides a flexible solution that can effectively identify and monitor high-risk Anopheles habitats over large study areas. A key advantage of the proposed method is its ability to overcome the challenges associated with obtaining sufficient training data and defining robust features for machine learning models. The proposed method is more accurate than classification models, and the design of the method ensures generalizability and scalability across different geographical regions. The method can be effectively localized by adjusting parameters based on local environmental conditions, such as different precipitation patterns and temperature ranges. This flexibility makes the method applicable to a variety of environments, enhancing its usefulness in global disease surveillance and control efforts. It is a valuable tool for detecting potential breeding sites of Anopheles mosquitoes, especially in areas prone to vector-borne diseases where field work is expensive, time-consuming and challenging. This invention not only promotes the application of remote sensing technology in the field of public health, but also provides a practical means for disease monitoring and control.

[0060] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limiting, and thus other examples of the exemplary embodiments may have different values.

[0061] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0062] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A method for identifying high-risk areas for malaria based on a segmented arbitrary model, characterized in that: The following steps are involved: Obtaining a meteorological data set of a study area in a preset year, and determining a high-risk month for malaria outbreak based on the meteorological data set, wherein the meteorological data set includes a long-term series of precipitation parameters and temperature parameters; Acquire Sentinel-1 data of the study area according to the high-risk month and establish a Sentinel-1 monthly image, calculate the SWI index according to the Sentinel-1 monthly image, and detect water bodies on the SWI map corresponding to the SWI index to obtain a Sentinel-1 water extraction model; Based on majority voting and decision fusion rules, high-risk water bodies in the Sentinel-1 water extraction model are identified to obtain high-risk areas corresponding to the high-risk months; The accuracy of the high-risk area is evaluated by using a water body mask generated by a segmented arbitrary model to complete the identification of high-risk areas for malaria, wherein the water body mask includes a plurality of water-containing depressions and a plurality of water-free depressions.

2. The method for identifying high-risk areas for malaria based on a segmented arbitrary model according to claim 1, obtaining a meteorological data set of a study area in a preset year, and determining a high-risk month for malaria outbreaks based on the meteorological data set, comprising: Predetermining a study period for the study area and obtaining a meteorological dataset of malaria transmission risk during the study period; The rainy season suitable for the growth and survival of Anopheles mosquitoes is extracted from the meteorological data set to obtain precipitation parameters and temperature parameters.

3. The method for identifying high-risk areas for malaria based on a segmented arbitrary model according to claim 2, before detecting water bodies on the SWI map corresponding to the SWI index to obtain the Sentinel-1 water extraction model, comprises: Sentinel-1 images during the study period were acquired through GEE; Temporary water bodies corresponding to the Sentinel-1 images were detected using multi-temporal data.

4. The method for identifying high-risk areas for malaria based on a segmented arbitrary model according to claim 1, identifying high-risk water bodies in the Sentinel-1 water extraction model based on majority voting and decision fusion rules to obtain high-risk areas corresponding to the high-risk months, comprising: Three different analyses were used to detect and classify high-risk areas, the first annual analysis identified areas with water during the four study months each year to determine where water storage persisted year-round; the second monthly analysis identified areas with water storage over the years; A four-level classification map was used to label areas with water impoundment every seven years as high risk, areas with water impoundment between four and six years as relatively risky, areas with water impoundment less than four years as low risk, and a fourth category was assigned to areas with no water in any of the study months, where the study period was seven years with preset years and study months.

5. The method for identifying high-risk areas for malaria based on a segmented arbitrary model according to claim 4, further comprising: Majority voting and weighted majority voting are used to detect high-risk areas. Majority voting includes each classification map voting for a specific category, and all votes have the same weight; In weighted majority voting, each classification map is assigned a weight based on its perceived reliability or performance, and the votes of each classification map are weighted according to its assigned weight, with the weights determined based on the precipitation in the study month to weigh the voting results.

6. The method for identifying high-risk areas for malaria based on a segmented arbitrary model according to claim 1, wherein the water mask generated by the segmented arbitrary model is used to evaluate the accuracy of the high-risk area to complete the identification of the high-risk area for malaria, comprising: Use multi-point feature hints to segment water bodies and convert points in geographic data frames into coordinates List; Set SamGeo as a predictor for a single coordinate and predict the features of the single coordinate in the instance to obtain a mask, and save the mask to a unique file for each point; After generating a mask for each point, all masks are merged into a single mosaic raster file, providing a complete representation of all segmented regions from the multi-point feature cues.

7. The method for identifying high-risk areas for malaria based on a segmented arbitrary model according to claim 1, further comprising: Get downloaded tiles and create georeferenced images using the tile map service, select satellite images for segmentation, and set paths to define the SAM model, where SamGeo has been initialized with the specified parameters and the shapefile has been converted from a geodataframe to a list of point coordinates; By looping through the list of coordinate points, each point is predicted, segmented and saved into a unique Geotiff file to generate a unique mask for each point; All masks were resampled to 10m resolution and merged into one file as a mosaic mask containing information of all segmented regions; The accuracy of water bodies extracted from the SWI index was evaluated using multiple masks generated by SAM to obtain evaluation results, wherein the performance of the model was evaluated using a confusion matrix and comprehensive indicators, including precision, recall, F1 score, overall accuracy OA or Kappa.

8. According to the method for identifying high-risk areas for malaria based on a segmented arbitrary model in claim 7, after calculating the surface water distribution map corresponding to all geographic reference images, the corresponding calculated flooding frequency IF of the depression is: , WOT means water observation time, and TOT means total observation time.

9. The method for identifying high-risk areas for malaria based on a segmented arbitrary model according to claim 8, wherein the evaluation result is compared with four classification algorithms to complete the classification of four types of land cover: vegetation, urban areas, wasteland and water bodies, wherein: The four classification algorithms include RF model, SVM model, SGTB model and SC model.

10. A device for identifying high-risk areas for malaria based on a segmented arbitrary model, characterized in that: The method for identifying high-risk areas for malaria based on a segmented arbitrary model as described in any one of claims 1 to 9 comprises: A data acquisition unit, used to acquire a meteorological data set of a study area in a preset year, and determine a high-risk month for malaria outbreak according to the meteorological data set, wherein the meteorological data set includes a long-term series of precipitation parameters and temperature parameters; A model extraction unit, used for acquiring Sentinel-1 data of the study area according to the high-risk month and establishing a Sentinel-1 monthly image, calculating a SWI index according to the Sentinel-1 monthly image, and detecting water bodies on a SWI map corresponding to the SWI index to obtain a Sentinel-1 water extraction model; A region identification unit, used for identifying high-risk water bodies in the Sentinel-1 water extraction model based on majority voting and decision fusion rules to obtain high-risk regions corresponding to the high-risk months; The model evaluation unit is used to evaluate the accuracy of the high-risk area by using a water body mask generated by a segmented arbitrary model to complete the identification of high-risk areas for malaria, wherein the water body mask includes a plurality of water-containing depressions and a plurality of water-free depressions.

Images