Mosquito distribution prediction method and apparatus based on semantic segmentation and ensemble learning

Abstract

Provided in the embodiments of the present invention are a mosquito distribution prediction method and apparatus based on semantic segmentation and ensemble learning. The method comprises: selecting a target research point, and acquiring mosquito data of the target research point; acquiring a remote sensing image map of the target research point, and inputting same into a trained Deeplabv3 model for semantic segmentation, so as to obtain a semantic segmentation result; and performing correlation analysis on ground object feature data of the target research point and the mosquito data of the target point, if there is a typical correlation relationship, obtaining ground object feature data related to mosquito distribution, and training an XGBoost classification model by using a training set with the ground object feature data related to the mosquito distribution as an independent variable and the mosquito data of the target point as a dependent variable, so as to predict the mosquito distribution of the target research point. The method allows for mosquito distribution prediction for large-area regions, saves labor, improves efficiency, and accurately predicts mosquito distribution, thus improving the accuracy and reliability of prediction results.

Description

Mosquito distribution prediction method and device based on semantic segmentation and ensemble learning TECHNICAL FIELD

[0001] The present application relates generally to the field of environmental science and public health, and more specifically, to a mosquito distribution prediction method and device based on semantic segmentation and ensemble learning. BACKGROUND

[0002] Mosquito-borne diseases are a class of infectious diseases caused by female mosquitoes sucking human blood to spread pathogenic microorganisms, such as dengue fever and malaria, which pose a serious threat to human life and health. The spread of mosquito-borne diseases is related to different types of mosquitoes, so studying mosquito distribution is of great significance to understanding the risk distribution area of mosquito-borne diseases. Mosquito distribution is closely related to environmental characteristics, for example, Culex pipiens pallens is positively correlated with wetlands, Culex tritaeniorhynchus is positively correlated with cultivated land, Aedes albopictus is positively correlated with building land, etc. By studying the distribution of regional land features, it is helpful to understand the distribution characteristics of mosquito species. Currently, the interpretation of land features usually adopts manual methods, but this method is time-consuming and difficult to complete large-scale mosquito distribution research.

[0003] In recent years, machine deep learning has been widely used in semantic segmentation of high-resolution remote sensing images. Among them, Deeplabv3 is a classic semantic segmentation deep learning model. DeepLabv3 model uses deep convolutional neural network (CNN) architecture, combined with techniques such as atrous convolution and spatial pyramid pooling, to improve segmentation accuracy and receptive field effectiveness. In addition, the model also uses conditional random field (CRF) to further optimize the segmentation results to obtain more accurate boundaries and details. For tabular data mining prediction, xgboost is an excellent algorithm. XGBoost (Extreme Gradient Boosting Machine) is a machine learning method based on gradient framework ensemble learning algorithm, which creates a new model to predict the residual of the current model, and adds all the models in series as the final prediction. In recent years, XGBoost classification model has been widely used in the medical field, including the classification of cardiovascular diseases and cancers.

[0004] Applying deep learning and machine learning to mosquito distribution research has multiple implications. First, these methods can learn and identify patterns based on large amounts of data, providing high-precision mosquito distribution predictions. Compared with traditional statistical methods, deep learning models have stronger non-linear modeling capabilities, better capturing complex mosquito distribution patterns and improving prediction accuracy. Second, deep learning and machine learning methods can quickly process large-scale remote sensing image data and geographic information data, extracting useful features and accelerating research progress. Third, accurately predicting mosquito distribution and activity areas is crucial for early warning and prevention of mosquito-borne diseases. Deep learning and machine learning methods can help researchers identify high-risk areas in a timely manner and take effective prevention measures to reduce the spread of mosquito-borne diseases and protect human health. Finally, the mosquito distribution prediction results generated by deep learning and machine learning methods can be visualized to help decision-makers and the public better understand and respond to the threat of mosquito-borne diseases.

[0005] For the problem of mosquito distribution prediction using deep learning and machine learning, the difficulty lies in accurately analyzing the relevant environment and performing accurate high-precision remote sensing semantic segmentation and data relationship mining on the detected small amount of data. First, since the single area radius of the detection statistics is only 250 meters, the range is small, and some detailed ground feature characteristics need to be extracted in a high-precision manner within the range. Second, due to the high cost of insect monitoring in terms of time and economic cost, only a few points are usually collected, resulting in a small amount of data. SUMMARY

[0006] According to an embodiment of the present application, a mosquito distribution prediction scheme based on semantic segmentation and ensemble learning is provided. This scheme can predict mosquito distribution in a large area, solve the problem that manual prediction in a large area is not possible, improve efficiency, and solve the problems of high-precision remote sensing semantic segmentation and small data volume of small sample data, achieving mosquito distribution prediction.

[0007] In a first aspect of the present application, a mosquito distribution prediction method based on semantic segmentation and ensemble learning is provided. The method comprises:

[0008] selecting a target research point in the target research area and obtaining mosquito data of the target research point;

[0009] using a deep learning image semantic segmentation dataset as a first training set, training a Deeplabv3 model through an Adadelta optimizer, and obtaining a trained Deeplabv3 model;

[0010] Acquire a remote sensing image map of the target research point, input the trained Deeplabv3 model for semantic segmentation, and obtain a semantic segmentation result, wherein the semantic segmentation result contains ground feature data of the target research point;

[0011] The ground feature data of the target research point and the mosquito data of the target point are used as feature variables for correlation analysis. If there is a typical correlation, the ground feature data related to mosquito distribution is obtained. The ground feature data related to mosquito distribution is used as the independent variable, and the mosquito data of the target point is used as the dependent variable to train an XGBoost classification model. The mosquito distribution of the target research point is predicted by the trained XGBoost classification model.

[0012] Further, the Deeplabv3 model includes a first sub-model and a second sub-model; the first sub-model uses a ResNet50 network structure as a feature extraction network; and the second sub-model includes a ResNet50 network structure and a multi-scale block arranged after the ResNet50 network structure.

[0013] Further, the deep learning image semantic segmentation dataset is subjected to data enhancement in a random flip and stretching manner to obtain a data enhanced dataset as a first training set.

[0014] Further, the training process of the XGBoost classification model includes:

[0015] Common species and dominant species are selected from the training set, and the XGBoost classification model is trained using the common species and the dominant species to obtain a common species model and a dominant species model.

[0016] Further, the training set is subjected to data optimization to obtain an optimized training set; and the XGBoost classification model is trained by using the optimized training set.

[0017] Further, the data optimization includes:

[0018] Lake surface data is added to the ground feature data related to mosquito distribution in the training set to obtain a first dataset;

[0019] The first dataset is grouped by using pseudo-labels to generate a second dataset containing multiple groups of data;

[0020] The second dataset is subjected to weighted processing to obtain a weighted second dataset;

[0021] Screening the data set containing multiple mosquito type combinations in the weighted second data set, splitting it into multiple data sets containing only a single mosquito type as the optimized training set.

[0022] Further, after predicting the mosquito distribution of the target research point, the prediction result is labeled to generate the mosquito distribution characteristics of the target research point.

[0023] Further, the mosquito distribution prediction method based on semantic segmentation and ensemble learning is performed on the research point in the target research area to generate the pixel of the mosquito distribution in the target research area.

[0024] In the second aspect of the present application, a mosquito distribution prediction device based on semantic segmentation and ensemble learning is provided. The device comprises:

[0025] The data acquisition module is used to select the target research point in the target research area and acquire the mosquito data of the target research point.

[0026] The training module is used to take the deep learning image semantic segmentation data set as the first training set, train the Deeplabv3 model through the Adadelta optimizer, and obtain the trained Deeplabv3 model.

[0027] The semantic segmentation module is used to acquire the remote sensing image of the target research point, input the trained Deeplabv3 model for semantic segmentation, obtain the semantic segmentation result, and screen the feature data of the ground object related to the mosquito distribution from the semantic segmentation result.

[0028] The prediction module is used to take the feature data of the ground object related to the mosquito distribution and the mosquito data of the target research point as the characteristic variable for correlation analysis. If there is a typical correlation between the feature data of the ground object related to the mosquito distribution and the mosquito data of the target research point, the XGBoost classification model is trained by taking the feature data of the ground object related to the mosquito distribution as the independent variable and the mosquito data of the target research point as the dependent variable, and the mosquito distribution of the target research point is predicted by the trained XGBoost classification model.

[0029] In the third aspect of the present application, an electronic device is provided. The electronic device comprises at least one processor; and a memory connected with the at least one processor in communication; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method of the first aspect of the present application.

[0030] It is to be understood that the description in the summary is not intended to identify key or essential features of embodiments of the application, nor is it intended to limit the scope of the application. Other aspects of the application will be evident from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0031] The above and other features, aspects, and advantages of embodiments of the present application will become more apparent from the following description in conjunction with the accompanying drawings. In the drawings:

[0032] FIG. 1 shows a flowchart of a mosquito distribution prediction method based on semantic segmentation and ensemble learning according to an embodiment of the present application;

[0033] FIG. 2 shows a block diagram of a mosquito distribution prediction apparatus based on semantic segmentation and ensemble learning according to an embodiment of the present application;

[0034] FIG. 3 shows a block diagram of an exemplary electronic device capable of implementing an embodiment of the present application;

[0035] FIG. 4 shows a schematic diagram of a semantic segmentation method capable of implementing an embodiment of the present application;

[0036] In the figure, 300 is an electronic device, 301 is a computing unit, 302 is a ROM, 303 is a RAM, 304 is a bus, 305 is an I / O interface, 306 is an input unit, 307 is an output unit, 308 is a storage unit, and 309 is a communication unit. DETAILED DESCRIPTION

[0037] To make the objectives, technical solutions, and advantages of embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0038] In addition, the term “and / or” herein merely describes an association relationship of associated objects, and can represent three relationships, for example, A and / or B can represent three cases of existence of A alone, existence of A and B together, and existence of B alone. In addition, the character “ / ” herein generally represents an “or” relationship between the front and rear associated objects.

[0039] FIG. 1 shows a flowchart of a mosquito distribution prediction method based on semantic segmentation and ensemble learning according to an embodiment of the present application.

[0040] The method comprises:

[0041] S101. Select a target research point in a target research area, and obtain mosquito data of the target research point.

[0042] In this embodiment, the target research area is a selected geographical area, which can be represented by latitude and longitude coordinates in a world coordinate system. It can also be a district range, such as Wuxi City. The target research point is a selected monitoring point in the target research area. The monitoring data is the mosquito data monitored in the buffer zone around each target research point. The mosquito data refers to the species of mosquitoes collected at each point, and the number of individuals of each species.

[0043] S102. Take the deep learning image semantic segmentation dataset as the first training set, train the Deeplabv3 model through the Adadelta optimizer, and obtain the trained Deeplabv3 model.

[0044] In this embodiment, the deep learning image semantic segmentation dataset is the loveda dataset. The loveda dataset is the work of Wuhan University State Key Laboratory of Surveying and Mapping and Remote Sensing Information Engineering in the field of land cover classification, which contains 5987 0.3-meter high-resolution images and 166768 semantic annotation objects from three different cities, and has been accepted by NeurlPS 2021 Datasets and Benchmarks Track.

[0045] In some embodiments of the present application, the deep learning image semantic segmentation dataset can also be subjected to data enhancement in the form of random flipping and stretching to obtain a data-enhanced dataset as the first training set. In addition, considering the color difference between the remote sensing image and the loveda dataset, the color can be randomly changed to achieve good accuracy.

[0046] Data enhancement of the deep learning image semantic segmentation dataset can achieve good accuracy.

[0047] In this embodiment, the Deeplabv3 model includes a first sub-model and a second sub-model; the first sub-model adopts a traditional Deeplabv3 structure, that is, a ResNet50 network structure is used as a feature extraction network (backbone); the second sub-model includes a ResNet50 network structure and a feature pyramid network (FPN) arranged after the ResNet50 network structure as a multi-scale block. Parameters are optimized in the reverse direction by predicting results from different scales, aiming to achieve the effect of spatial pyramid multi-scale feature extraction. This embodiment uses the method of adding a spatial pyramid module to optimize the DeepLabv3 module to solve the problem of high-precision remote sensing semantic segmentation difficulty, and improves the accuracy and precision of the segmentation result.

[0048] In this embodiment, the Deeplabv3 model is trained using the Adadelta optimizer, and the initial learning rate is set to 0.1. For the weight of the loss function, [0.6, 0.9, 2.0, 2.8, 1.7, 1.9, 1.5, 1.6] is used to balance the importance of different losses. The setting of the weight should be retrained according to the specificity between the regional remote sensing image and the loveda dataset, and customized according to the loss condition.

[0049] By using the deep learning algorithm Deeplabv3 and combining the method of extracting features of ground objects from remote sensing images, the distribution of mosquitoes can be efficiently predicted. This method uses a deep learning model to analyze and predict remote sensing images to obtain accurate prediction results of ground object features.

[0050] S103, acquire the remote sensing image of the target research point, input the trained Deeplabv3 model for semantic segmentation, obtain the semantic segmentation result, and select the ground feature data related to the distribution of mosquitoes from the semantic segmentation result.

[0051] Specifically, in inference, the Deeplabv3 model uses the TTA (Test-Time Augmentation) method for data enhancement to predict the results of the remote sensing image of the target research point, votes multiple results, and takes the maximum value of the voting as the semantic segmentation result for output.

[0052] In this embodiment, the ground feature data related to the distribution of mosquitoes is selected from the semantic segmentation result, including:

[0053] The semantic segmentation result is judged, and if the result probability P in the semantic segmentation result is not greater than 0.05, it is considered that the ground feature data in the semantic segmentation result is the ground feature data related to the distribution of mosquitoes.

[0054] S104, the land feature data related to the mosquito distribution and the mosquito data of the target research point are taken as feature variables to perform correlation analysis, if there is a typical correlation between the land feature data related to the mosquito distribution and the mosquito data of the target research point, an XGBoost classification model is trained by taking the land feature data related to the mosquito distribution as an independent variable and the mosquito data of the target research point as a dependent variable, and the mosquito distribution of the target research point is predicted by the trained XGBoost classification model.

[0055] In the embodiment, by performing feature engineering, the correlation analysis is performed by using the canonical correlation analysis, the land feature of the research point is taken as a vector X (x1, x2, x3, …), and the mosquito community is taken as a vector Y (y1, y2, y3, …), and the SPSS19.0 program is used to analyze the canonical correlation between the mosquito and the land feature.

[0056] In the embodiment, the training process of the XGBoost classification model includes:

[0057] First, the common species and the dominant species are selected from the training set, and the XGBoost classification model is trained by using the common species and the dominant species respectively to obtain a common species model and a dominant species model.

[0058] In the embodiment, the dominant species accounts for more than 20% of the total number, and the common species accounts for 5% to 20% of the total number.

[0059] In terms of configuring the training parameters, the 10-fold cross-validation and the grid search method are used to configure the training parameters of the model.

[0060] In some embodiments, the dominant species and the common species need to be processed differently when training the model parameters by using the grid search method.

[0061] In the embodiment, the training set needs to be optimized to obtain an optimized training set, and the XGBoost classification model is trained by using the optimized training set.

[0062] The XGBoost machine learning algorithm is used to establish a classification model, which can accurately predict the dominant species and the common species of mosquitoes. Through the modeling analysis of the land feature data and the mosquito data, the method can accurately predict the mosquito distribution, and provide an important reference for environmental scientific analysis and public health prevention and control.

[0063] The data optimization includes four optimization methods:

[0064] The first kind is to add lake surface data to the terrain feature data related to mosquito distribution in the training set to obtain a first data set; the first data set contains mosquito data of the target research point and the terrain feature data related to mosquito distribution with added lake surface data.

[0065] The lake surface data is data in the semantic segmentation result obtained after semantic segmentation, which represents the terrain feature data of the water surface at the target research point.

[0066] The second kind is to participate in training through pseudo-labeling, group the original data, for example, into 12 groups, and gradually include multiple groups of data into the training set to generate a training set containing multiple groups of data to gradually improve the accuracy of training.

[0067] The third kind is to perform weighted processing on the original data, for example, to weight the original data by 3 times, to ensure that the original data has a greater impact on the model in the first few rounds of training, thereby improving the accuracy of the pseudo-labeling.

[0068] The fourth kind, because the time and economic cost of insect monitoring are both high, generally only a few point positions are collected, resulting in more target classification data and less data quantity, the data group containing multiple combinations of mosquito types is split into multiple data groups containing only a single mosquito type to reduce the number of classifications and improve the accuracy.

[0069] It can be seen that the method of processing data using a large number of small samples in the embodiment splits the combination, so that the prediction is only a single mosquito, and finally expresses the combination result of the prediction by probability, reduces the number of classifications, improves the precision, solves the problem of small sample data prediction difficulty in mosquito monitoring, and improves the accuracy and reliability of the prediction result.

[0070] In order to further illustrate the beneficial effects of the present application, the accuracy, recall rate, precision and F1 are used as evaluation indexes to evaluate the experimental results in the embodiment. F1 represents F1 score;

[0071]

[0072] Compared with the artificial results, the results are shown in Table 1 as follows:

[0073]

[0074] It can be seen that the indicators of the dominant population and the conventional population are relatively high, and the model effect is excellent.

[0075] In the embodiment, after predicting the mosquito distribution of the target point, the prediction result is obtained, and the result of the model prediction result probability of the dominant mosquito species being greater than 0.15 is regarded as the existence of the dominant species; and the result of the model prediction result probability of the common mosquito species being greater than 0.25 is regarded as the existence of the common species.

[0076] Optionally, after predicting the mosquito distribution of the target research point, the prediction result can also be labeled to generate the mosquito distribution characteristics of the target research point, that is, the dominant species and the common species.

[0077] In some optional implementation manners of the embodiment, the pixel map of the mosquito distribution in the target research area can also be constructed according to the prediction result and the mosquito distribution characteristics.

[0078] Taking Wuxi City as an example, a grid of 129 rows and 151 columns is established in the administrative area of Wuxi City according to the span and geographical distance of the longitude and latitude of Wuxi City, and 12647 points are obtained. The ground feature segmentation mode is shown in FIG. 4. For each point, the ground feature data in the 250-meter radius buffer area thereof is extracted, and the mosquito distribution characteristics of each buffer area are predicted by using the above model. According to the prediction result, different colors are used to represent the mosquito species characteristics in the buffer area, and the inscribed square of each buffer area is taken as a pixel point to form a pixel of mosquito distribution in the whole Wuxi City. According to the evaluation standard of mosquito experts, the result of the model prediction result probability of the dominant mosquito species being greater than 0.15 is regarded as the existence of the dominant species; and the result of the model prediction result probability of the common mosquito species being greater than 0.25 is regarded as the existence of the common species. Then, for the two model structures, the pixel map of the mosquito distribution is drawn respectively.

[0079] And there are two categories of mosquito data that need to be focused on in the field of public health, one is the dominant mosquito species, and the other is the common mosquito species. The prediction of the present method includes both aspects. According to the evaluation of mosquito experts, the result of the model prediction result probability of the dominant species being greater than 0.15 is regarded as the existence of the dominant species; and the result of the model prediction result probability of the common mosquito species being greater than 0.25 is regarded as the existence of the common species. Then, for the two model structures, the pixel map of the mosquito distribution is drawn respectively.

[0080] In order to illustrate the effectiveness of the classification method, the present application sets the following examples for experimental data comparison: algorithm one is the method provided by the present application (xgboost). Algorithm two is a clustering algorithm (kmeans). Algorithm three is a classic gradient boosting decision tree (gbdt).

[0081] The experiment uses the mosquito distribution data in Wuxi City and the corresponding regional terrain feature data as the experimental data set, and solves by using algorithms one, two and three respectively. In order to make a fair comparison, the control variable method is used to set the parameters of algorithms one, two and three, and the grid search method is used to combine and traverse the parameters of xgboost and gbdt. In the clustering algorithm, the advantage species clustering number of kmeans is 3, and the conventional species clustering number is 4, both of which use the kmeans++ algorithm.

[0082] Specifically, the parameter settings of the xgboost algorithm are shown in Table 2 as follows:

[0083]

[0084] The parameter settings of the Gbdt algorithm are shown in Table 3 as follows:

[0085]

[0086] The experimental results of the three algorithms are shown in Table 4 as follows:

[0087]

[0088] The experimental results show that the method of the present application can obtain better results than the comparative algorithm in different mosquito species prediction and different category prediction. It can be found from the experimental results that the experimental data does not show obvious clustering under clustering, and the contour coefficient of the clustering obtained by the experiment is 0.58, indicating that the clustering result is not very ideal. Comparing the experimental results of Xgboost and gbdt, the accuracy of xgboost is higher, although the precision is slightly lower, but the F1 value is higher, reflecting better prediction stability, indicating that the xgboost algorithm can better mine data relationships and predict data results.

[0089] It should be noted that, for the foregoing method embodiments, in order to simply describe, they are all expressed as a series of action combinations, but those skilled in the art should know that the present application is not limited by the action sequence described, because according to the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present application.

[0090] The above is the introduction of the method embodiment, and the following will further illustrate the scheme of the present application through the device embodiment with the same inventive concept as the method in the foregoing embodiment.

[0091] As shown in Figure 2, the device 200 comprises:

[0092] The data acquisition module 210 is configured to select a target research point in a target research area, and acquire mosquito data of the target research point.

[0093] The training module 220 is configured to take a deep learning image semantic segmentation dataset as a first training set, train a Deeplabv3 model through an Adadelta optimizer, and obtain a trained Deeplabv3 model.

[0094] The semantic segmentation module 230 is configured to acquire a remote sensing image of the target research point, input the trained Deeplabv3 model for semantic segmentation, and obtain a semantic segmentation result, wherein the semantic segmentation result comprises feature data of a ground object of the target research point.

[0095] The prediction module 240 is configured to take the feature data of the ground object of the target research point and mosquito data of the target point as characteristic variables, perform correlation analysis, obtain feature data of a ground object related to mosquito distribution if there is a typical correlation, take the feature data of the ground object related to mosquito distribution as an independent variable, and take mosquito data of the target point as a dependent variable to train an XGBoost classification model, and predict mosquito distribution of the target research point through the trained XGBoost classification model.

[0096] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described modules can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0097] In the technical solution of the present application, the acquisition, storage and application of user personal information comply with relevant laws and regulations and do not violate public order and good customs.

[0098] According to the embodiments of the present application, the present application further provides an electronic device.

[0099] Figure 3 shows a schematic block diagram of an electronic device 300 that can be used to implement embodiments of the present application. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not intended to limit implementations of the present application described and / or claimed in this document.

[0100] The electronic device 300 includes a computing unit 301 that can perform various appropriate actions and processes in accordance with a computer program stored in a read-only memory (ROM) 302 or a computer program loaded from a storage unit 308 into a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 can also be stored. The computing unit 301, the ROM 302, and the RAM 303 are connected to each other through a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0101] A plurality of components in the electronic device 300 are connected to the I / O interface 305, including: an input unit 306, such as a keyboard, a mouse, and the like; an output unit 307, such as various types of displays, a speaker, and the like; the storage unit 308, such as a magnetic disk, an optical disk, and the like; and a communication unit 309, such as a network card, a modem, a wireless communication transceiver, and the like. The communication unit 309 allows the electronic device 300 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0102] The computing unit 301 can be various general-purpose and / or special-purpose processing components having processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, and the like. The computing unit 301 performs various methods and processes described above, such as the methods S101-S104. For example, in some embodiments, the methods S101-S104 can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 308. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 300 via the ROM 302 and / or the communication unit 309. When the computer program is loaded into the RAM 303 and executed by the computing unit 301, one or more steps of the methods S101-S104 described above can be performed. Alternatively, in other embodiments, the computing unit 301 can be configured to perform the methods S101-S104 by any other appropriate means, such as by means of firmware.

[0103] The various embodiments of the systems and techniques described above can be implemented in digital electronic circuitry, integrated circuitry, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system on a chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0104] Program code for carrying out methods of the present application can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces a means for implementing the functions / acts specified in the flowcharts and / or block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, or entirely on a remote machine or server.

[0105] In the context of the present application, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), optical fibers, portable compact disc read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0106] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0107] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0108] The computer system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server is generally established by computer programs running on the respective computers and having a client-server relationship to each other. The servers can be cloud servers, servers of a distributed system, or servers combined with a blockchain.

[0109] It should be understood that the various forms of flow shown above can be re-ordered, added to, or deleted from without departing from the spirit and scope of the present disclosure. For example, the steps recited in the present disclosure can be performed in parallel, in series, or in different orders, as long as the desired results of the present disclosure are achieved, and are not limited herein.

[0110] The specific embodiments described above are not intended to limit the scope of the present disclosure. Those skilled in the art will understand that various modifications, combinations, sub-combinations, and alternatives can be made to the specific embodiments described above without departing from the spirit and principles of the present disclosure. Any further modifications, changes, improvements, and the like that come within the spirit and principles of the present disclosure should also be included within the scope of the present disclosure.

Claims

1. A mosquito distribution prediction method based on semantic segmentation and ensemble learning, characterized in that, include: Select a target research point in the target research area and obtain mosquito data at the target research point; The Deeplabv3 model is trained using the deep learning image semantic segmentation dataset as the first training set using the Adadelta optimizer to obtain the trained Deeplabv3 model. Obtain a remote sensing image of the target research point, input the image into the trained Deeplabv3 model for semantic segmentation, and obtain a semantic segmentation result, wherein the semantic segmentation result includes the ground feature data of the target research point; The feature data of the target research site and the mosquito data of the target research site are used as feature variables for correlation analysis. If a typical correlation exists, the feature data related to the mosquito distribution is obtained. The feature data related to the mosquito distribution is used as the independent variable and the mosquito data of the target research site is used as the training set for the dependent variable to train an XGBoost classification model. The mosquito distribution of the target research site is predicted by the trained XGBoost classification model. The training process of the XGBoost classification model includes: Common species and dominant species are selected from the training set, and the XGBoost classification model is trained using the common species and the dominant species, respectively, to obtain a common species model and a dominant species model; Performing data optimization on the training set to obtain an optimized training set; and training the XGBoost classification model using the optimized training set. The data optimization includes: Adding lake surface data to the ground feature data related to mosquito distribution in the training set to obtain a first data set; Grouping the first data set by pseudo labels to generate a second data set containing multiple groups of data; performing weighted processing on the second data set to obtain a weighted second data set; The data group containing a combination of multiple mosquito types in the weighted second data set is screened out, and the data group is split into multiple data groups containing only a single mosquito type as the optimized training set.

2. The method of claim 1, wherein, The Deeplabv3 model includes a first sub-model and a second sub-model; the first sub-model uses the ResNet50 network structure as the feature extraction network; the second sub-model includes the ResNet50 network structure and a multi-scale block set after the ResNet50 network structure.

3. The method of claim 1, wherein, The deep learning image semantic segmentation dataset is subjected to data enhancement by random flipping and scaling, and the data-enhanced dataset is obtained as the first training set.

4. The method of claim 1, wherein, After predicting the mosquito distribution at the target research point, the prediction results are marked to generate mosquito distribution characteristics at the target research point.

5. The method of claim 1, wherein, The mosquito distribution prediction method based on semantic segmentation and ensemble learning as claimed in claim 1 is executed on several research points within the target research area to generate pixels of mosquito distribution within the target research area.

6. A mosquito distribution prediction device based on semantic segmentation and ensemble learning, characterized by, include: A data acquisition module is used to select a target research point in the target research area and obtain mosquito data at the target research point; The training module is configured to train the Deeplabv3 model by using an Adadelta optimizer and taking a deep learning image semantic segmentation dataset as a first training set, to obtain a trained Deeplabv3 model; The semantic segmentation module is configured to acquire a remote sensing image of the target research point, input the trained Deeplabv3 model to perform semantic segmentation, obtain a semantic segmentation result, and filter out feature data of a ground object related to mosquito distribution from the semantic segmentation result; The prediction module is configured to take the feature data of the ground object related to mosquito distribution and mosquito data of the target research point as feature variables, perform correlation analysis, and if a typical correlation exists between the feature data of the ground object related to mosquito distribution and the mosquito data of the target research point, train an XGBoost classification model by taking the feature data of the ground object related to mosquito distribution as an independent variable and the mosquito data of the target research point as a dependent variable, and predict mosquito distribution of the target research point by using the trained XGBoost classification model. The training process of the XGBoost classification model includes: Filtering out common species and dominant species from the training set, training the XGBoost classification model by using the common species and the dominant species respectively, and obtaining a common species model and a dominant species model; Optimizing the training set to obtain an optimized training set, and training the XGBoost classification model by using the optimized training set; The data optimization includes: Adding lake surface data to the feature data of the ground object related to mosquito distribution in the training set to obtain a first data set; Grouping the first data set by using pseudo-labels to generate a second data set containing multiple groups of data; Performing weighted processing on the second data set to obtain a weighted second data set; Filtering out data groups containing multiple mosquito type combinations in the weighted second data set, splitting the data groups into multiple data groups containing only a single mosquito type, and taking the data groups as an optimized training set.

7. An electronic device, comprising at least one processor; and A memory connected in communication with the at least one processor; characterized in that The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-5.

Images