WO2021169464A1 - 养殖场风险预测方法、装置、设备及存储介质 - Google Patents

养殖场风险预测方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2021169464A1
WO2021169464A1 PCT/CN2020/134032 CN2020134032W WO2021169464A1 WO 2021169464 A1 WO2021169464 A1 WO 2021169464A1 CN 2020134032 W CN2020134032 W CN 2020134032W WO 2021169464 A1 WO2021169464 A1 WO 2021169464A1
Authority
WO
WIPO (PCT)
Prior art keywords
farm
information
risk
map
breeding
Prior art date
Application number
PCT/CN2020/134032
Other languages
English (en)
French (fr)
Inventor
蔡林
周古月
吴泽衡
徐倩
杨强
Original Assignee
深圳前海微众银行股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳前海微众银行股份有限公司 filed Critical 深圳前海微众银行股份有限公司
Publication of WO2021169464A1 publication Critical patent/WO2021169464A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/29—Geographical information databases
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0635—Risk analysis of enterprise or organisation activities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/02—Agriculture; Fishing; Forestry; Mining

Definitions

  • This application relates to the field of artificial intelligence technology, and in particular to a method, device, equipment and storage medium for risk prediction of a farm.
  • African swine fever is an acute, febrile, and highly contact animal infectious disease of pigs caused by the African swine fever virus, with a related fatality rate as high as 100%. Since the first case of African swine fever in China was diagnosed by the China Animal Health and Epidemiology Center, African swine fever epidemics have occurred successively in some provinces in China, causing serious losses. Once this animal infectious disease similar to African swine fever is discovered, it is necessary to quickly and accurately take preventive and control measures to avoid the continuous expansion of the epidemic, which will cause adverse effects in all aspects of society. In order to make rapid and accurate prevention and control measures, it is necessary to quickly and accurately understand the risk situation of each farm.
  • the current situation is the risk assessment of the farm, which requires experts to conduct an on-site inspection of the situation of the farm, or rely on the information reported by the farmer.
  • the experts' on-site inspection program is obviously inefficient and difficult to respond to sudden and rapidly spreading epidemics; if the farmers actively report, the farmers may conceal the breeding information, resulting in low risk assessment accuracy. The inability to quickly and accurately obtain the risk information of the farm will result in poor prevention and control work.
  • the main purpose of this application is to provide a method, device, equipment, and storage medium for risk prediction of farms, aiming to solve the problem of current animal epidemics that cannot quickly and accurately obtain the risk conditions of farms, resulting in poor prevention and control work. problem.
  • the breeding information and the other influencing factor information are input into a preset risk prediction model to obtain the risk prediction result of the farm to be predicted.
  • the present application also provides a risk prediction device for a breeding farm
  • the risk prediction device for a breeding farm includes:
  • the input module is used to input the map to be detected into the farm detection model to obtain the farm information in the map to be detected;
  • the determining module is used to determine the farm to be predicted according to the farm information, determine the farming information within a preset range around the farm to be predicted, and obtain other influencing factors in the preset range except the farm information;
  • the prediction module is used to input the breeding information and the other influencing factor information into a preset risk prediction model to obtain the risk prediction result of the farm to be predicted.
  • this application also provides a farm risk prediction device, the farm risk prediction device comprising: a memory, a processor, and a farm risk stored in the memory and capable of running on the processor
  • a prediction program which implements the steps of the method for predicting the risk of a farm as described above when the program for predicting the risk of a farm is executed by the processor.
  • this application also proposes a computer-readable storage medium with a farm risk prediction program stored on the computer-readable storage medium.
  • the farm risk prediction program is executed by a processor, the above The steps of the farm risk prediction method described.
  • the farm information in the map to be detected is obtained; the farm to be predicted is determined according to the farm information and the predetermined range around the farm to be predicted is determined Farming information, and obtain other influencing factor information within a preset range; input the farming information and other influencing factor information into the preset risk detection model to obtain the risk prediction result of the farm to be predicted.
  • epidemic prevention and control work can be carried out more quickly and in a targeted manner, avoiding the out of control of the epidemic, improving the quality and safety of agricultural products, and improving the ecological environment.
  • FIG. 1 is a schematic structural diagram of a hardware operating environment involved in a solution of an embodiment of the present application
  • Figure 2 is a schematic flowchart of the first embodiment of the risk prediction method for a farm under the application
  • Fig. 3 is a functional schematic block diagram of a preferred embodiment of a risk prediction device for a farm of the application
  • Figure 4 is a schematic diagram of a risk prediction process involved in an embodiment of this application.
  • FIG. 1 is a schematic diagram of the device structure of the hardware operating environment involved in the solution of the embodiment of the present application.
  • the risk prediction equipment of the farm in the embodiment of the present application may be a smart phone, a personal computer, a server, and other equipment, which is not specifically limited here.
  • the farm risk prediction equipment may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002.
  • the communication bus 1002 is used to implement connection and communication between these components.
  • the user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface.
  • the network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).
  • the memory 1005 may be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a magnetic disk memory.
  • the memory 1005 may also be a storage device independent of the aforementioned processor 1001.
  • FIG. 1 does not constitute a limitation on the risk prediction equipment of the farm.
  • the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a farm risk prediction program.
  • the user interface 1003 is mainly used to communicate with the client;
  • the network interface 1004 is mainly used to establish a communication connection with the server;
  • the processor 1001 can be used to call the farm risk stored in the memory 1005 Predict the program and do the following:
  • the breeding information and the other influencing factor information are input into a preset risk prediction model to obtain the risk prediction result of the farm to be predicted.
  • FIG. 2 is a schematic flowchart of a first embodiment of a method for predicting a farm risk in this application.
  • the embodiment of the application provides an embodiment of the method for predicting the risk of a farm. It should be noted that although the logical sequence is shown in the flowchart, in some cases, the sequence shown here can be executed in a different order. Or the steps described.
  • the execution subject of the various embodiments of the method for predicting farm risk in this application may be devices such as smart phones, personal computers, and servers. For ease of description, the execution subject is omitted in the following embodiments for illustration.
  • the method for predicting the risk of a farm includes:
  • Step S10 input the map to be detected into a farm detection model to obtain farm information in the map to be detected, wherein the farm detection model is obtained by pre-training;
  • a map of the area can be obtained as the map to be inspected.
  • the risk can refer to the extent to which the farm may be affected in the case of the spread of swine fever and other animal infectious diseases.
  • a map can be a map downloaded from a map software, or a remote sensing satellite map can be obtained through a remote sensing satellite.
  • a farm detection model can be trained in advance to detect the farm in the picture.
  • the farm detection model can use commonly used image target detection models, such as the semantic image segmentation model DeepLab-v3+.
  • the training method of the farm detection model can use the existing training method of the machine learning model.
  • the output result of the farm detection model can be the location of the farm in the map to be detected, and then according to the scale of the map to be detected and the actual geographic location corresponding to the map to be detected, the farm in the area to be detected can be calculated Physical location.
  • the acquired farm information can include at least the actual geographic location of the farm.
  • the farm can be viewed as a point (the center point of the farm), and the actual geographic location of the farm is the location of the point; or the farm can be viewed as an area, the actual geographic location of the farm That is, the location of the area.
  • the actual geographic location can be expressed in latitude and longitude, or expressed in coordinates under a predetermined coordinate system, and so on.
  • Step S20 Determine the farm to be predicted according to the farm information, and determine the farming information within a preset range around the farm to be predicted, and obtain information about other influencing factors within the preset range except the farm;
  • the farm to be predicted can be determined according to the farm information. Specifically, according to the farm information, it is possible to determine how many farms are included in the area to be detected, and the actual geographic location of each farm. One of the multiple farms in the detected area to be detected can be selected as the farm to be predicted. Specifically, each farm can be used as the farm to be predicted in turn, or the farm to be predicted can be determined based on the user's selection.
  • the risk of the farm should be predicted.
  • the principle adopted in this embodiment is that the relevant data of the area around the farm can be used as the basis for judging the risk of the farm. Specifically, a preset range around the farm to be predicted can be determined first.
  • a radius value can be preset, with the farm as the center of the circle, and the radius as the radius, to divide a circular area, which can be used as the preset range of the farm
  • the actual geographic location of the farm and the radius value can be determined; it is also possible to use the farm as the center of the rectangle and preset the side length value of the matrix to divide A matrix area is used as the preset range of the farm; there are other feasible methods. According to different actual application scenarios, different methods can be adopted to determine the preset range, which will not be listed here.
  • the farm information within the preset range can be determined according to the farm information in the map to be detected. Specifically, according to the actual geographic location of the preset range and the actual geographic location of each farm in the farm information, it can be detected whether each farm falls within the preset range, and the farm falling within the preset range is determined as the target Farm, then the information of the target farm can be extracted from the farm information as the farming information within the preset range.
  • the preset range of positions can be composed of a longitude interval and a latitude interval. Then it is necessary to determine whether the longitude of each farm falls within the longitude interval and whether the latitude falls within the latitude interval. If they all fall, it is determined that the farm falls within the preset range.
  • water network can refer to rivers and lakes of various sizes
  • road network can refer to various types of roads, such as national highways and highways.
  • water network and the road network are collectively referred to as the "water network", which can mean that the water network and the road network are treated in a unified manner, or they can be treated separately.
  • information about other risk influencing factors other than the breeding farm within the preset range can be obtained, that is, information about other influencing factors.
  • other influencing factor information in the area to be detected may be collected in advance, wherein the method of obtaining other influencing factor information may be extraction from electronic map data, or manual collection.
  • the information of other influencing factors can specifically include the location and size of other risk influencing factors, and the data varies according to the type of risk influencing factors.
  • the collected village information can include the location information, area, etc. of the village, and the collected waterway network information. It can include the location information and length of the waterway network.
  • Step S30 Input the breeding information and the other influencing factor information into a preset risk prediction model to obtain the risk prediction result of the farm to be predicted.
  • a risk prediction model can be set in advance.
  • the risk prediction model can be a linear model or a neural network model. If it is a neural network model, the neural network model needs to be trained in advance, and the model obtained after training is used to For risk prediction, the training process of the neural network model can refer to the existing neural network model training process, which will not be described in detail here.
  • the setting principle of the risk prediction model is: when the input data characterizes the more and denser the target influencing factors within the preset range of the farm to be predicted, the output risk prediction result characterizes the higher the risk of the farm to be predicted.
  • the input data of the risk prediction model can be set according to the situation.
  • the input data can be set to the total area of the breeding farm and the total area of the waterway network within the preset range.
  • the aquaculture information and other influencing factor information can be processed into a form that matches the input data of the risk prediction model, and then the processed data is input into the risk prediction model.
  • the output of the farm to be predicted is output.
  • the risk prediction result can be in various forms, for example, it can be a score or a level.
  • the degree of risk corresponding to various scores or levels can be specified in advance, for example, the score is 0 to 10, and the degree of risk gradually increases from 0 to 10.
  • the farm information in the map to be detected is obtained; the farm to be predicted is determined according to the farm information and the preset range around the farm to be predicted is determined And obtain other influencing factor information within the preset range; input the aquaculture information and other influencing factor information into the preset risk detection model to obtain the risk prediction result of the farm to be predicted.
  • the farm detection model by training a farm detection model to detect each farm from the map and obtain farm information as a data source for risk prediction, it reduces the dependence on the information reported by the farmers and improves the farm’s performance.
  • Management efficiency has also improved the efficiency of risk prediction; accurate and timely statistics of farm information, so as to provide farmers with production and management plans, and improve the efficiency of animal husbandry operations. And by setting up a risk prediction model to process the breeding information and other influencing factors within the preset range of the farm to be predicted, the risk prediction results of the farm to be predicted are obtained, without the need for experts to inspect the situation of each farm.
  • the efficiency of risk prediction for farms is improved, and it is more able to cope with the rapid spread of the epidemic; compared with the plan of farmers actively reporting risk conditions, the unified risk prediction model is adopted in this embodiment to more accurately predict the risk of each farm.
  • Risk situation Based on rapid and accurate risk prediction results, epidemic prevention and control work can be carried out more quickly and in a targeted manner, avoiding the loss of control of the epidemic, improving the quality and safety of agricultural products, and improving the ecological environment.
  • different prevention and control plans can be set in advance for different risk prediction results.
  • the appropriate prevention and control plans can be matched from each prevention and control plan according to the risk prediction results, and the matched
  • the prevention and control plan is determined as the target prevention and control plan of the farm to be predicted, and the target prevention and control plan can be output.
  • the output mode can be output to the display screen or output in voice mode.
  • the output mode can be set differently according to specific application scenarios.
  • the actual application can be carried out in the form of mobile phone or computer application software, so that the user can input the relevant information of the farm to be predicted based on the installed application, for example, the location information. Obtain the risk prediction result of the farm to be predicted, and even the prevention and control plan.
  • the step S10 includes:
  • Step S102 input the map to be detected into the farm detection model to obtain the classification category of each pixel in the map to be detected, wherein the classification category is used to indicate whether the corresponding pixel belongs to the farm;
  • the farm detection model can be a two-classification model. After the map to be detected is input into the farm detection model, the farm detection model can output the classification result of each pixel in the map to be detected.
  • the classification result is used to indicate whether the corresponding pixel belongs to The farm is still not a farm.
  • the classification result can be 0 and 1. 1 indicates that the corresponding pixel belongs to a farm, and 0 indicates that the corresponding pixel does not belong to a farm.
  • Step S103 Determine the farm information in the map to be detected according to the classification category.
  • the farm information in the map to be detected can be determined. Specifically, it can be based on the classification of each pixel to determine the target pixel belonging to the farm from each pixel; then according to the scale of the map to be tested, it can be determined how much area each pixel represents in reality, and then according to the Detect the latitude and longitude interval of the map to determine the latitude and longitude interval of each target pixel. Since each target pixel corresponds to the farm area, the longitude and latitude interval of the target pixel can be used as the latitude and longitude of the corresponding farm area. Interval.
  • step S103 includes:
  • Step S1031 Determine target pixels belonging to the breeding farm in the map to be detected according to the classification category
  • the target pixel belonging to the farm is determined from each pixel. For example, when there are two classification categories, 0 and 1, all pixels with the classification category 1 are regarded as the target pixel.
  • Step S1032 clustering the target pixels by using a preset clustering algorithm to obtain each cluster group;
  • a clustering algorithm can be set in advance, and the target pixels can be clustered using the clustering algorithm to obtain each clustering group. It should be noted that when the actual area corresponding to a pixel in the map to be detected is relatively small, one pixel may represent a breeding house (that is, a unit smaller than a breeding farm), and there may be multiple areas of a breeding farm. Target pixels. At this time, clustering algorithm can be used to cluster multiple identified pixels, or cluster the latitude and longitude intervals corresponding to multiple pixels, and cluster multiple breeding houses that are close together.
  • the class is a cluster group.
  • the clustering algorithm can use a commonly used clustering algorithm, such as the nearest neighbor clustering method.
  • Step S1033 Determine the farm information in the map to be detected according to each of the cluster groups.
  • Each cluster group obtained by clustering is regarded as a breeding farm.
  • the latitude and longitude of the breeding farm can be the center of the cluster group, and the area of the breeding farm can be the sum of the area of all breeding houses within the cluster group.
  • the farm information in the farm to be detected may at least include the location and area of each farm.
  • step S10 it further includes:
  • Step S40 Preliminarily training the model to be trained using positive training data, where the positive training data includes multiple pre-collected remote sensing satellite maps containing farms, and farm label data corresponding to each remote sensing satellite map;
  • the training method of the farm detection model may be as follows:
  • the annotation data of the farm can be the mask map corresponding to the remote sensing satellite image.
  • the mask map has the classification category corresponding to each pixel of the remote sensing satellite image. For example, 0 means the pixel does not belong to the farm, and 1 means the pixel. Belonging to a farm, the mask map can use different colors to indicate different categories. Use multiple remote sensing satellite maps containing farms and farm data as positive training data.
  • the structure of the model to be trained can use commonly used semantic segmentation models, such as the semantic image segmentation model DeepLab-v3+.
  • the initial training process can use the existing machine learning model training process.
  • Step S50 using negative training data, or using the negative training data and the positive training data to adjust the model to be trained after preliminary training, wherein the negative training data includes a plurality of pre-collected Does not include remote sensing satellite maps of farms;
  • the ratio of negative training data to positive training data can be set according to actual experience.
  • the negative training data can be tenths of the positive training data. one.
  • the adjustment can be fine-tuning. Specifically, fine-tuning is to use positive training data and negative training data together for training, and the training process can also use the training process of an existing machine learning model.
  • the negative training data can be the positive training data.
  • the learning rate in the fine-tuning stage can also be set to one-tenth of the initial training stage.
  • Step S60 When it is detected that the adjusted model to be trained meets the preset model conditions, the adjusted model to be trained is used as the farm detection model; otherwise, the step is executed again based on the adjusted model to be trained: Example of training data for preliminary training of the model to be trained.
  • the adjusted model to be trained is used as a farm detection model, so that the farm detection model can be subsequently used for farm detection. Otherwise, if the preset model conditions are not met, the positive training data is used to perform preliminary training on the adjusted model to be trained, and then adjustment training is performed until it is detected that the preset model conditions are met.
  • the preset model conditions can be conditions set in advance according to the performance requirements of the model, such as the convergence of the model's loss function as a condition, or the commonly used objective indicators for detecting model performance as conditions, such as accuracy, Recall rate and IOU (Intersection objective indicators such as over Union.
  • the calculation method of each objective indicator can refer to the existing indicator calculation method, which will not be described in detail here.
  • the map to be detected is detected and identified, and the farm information in the map to be detected is obtained, so that the farm information in each area can be obtained intelligently and quickly, thereby improving
  • the risk prediction efficiency of the farm to be predicted is improved; and compared with the method that relies on the farmers to actively report, the scheme in this embodiment can obtain farm information more comprehensively, so that the data on which the risk prediction is based is more accurate and comprehensive. Make the risk prediction result more accurate.
  • step S40 includes:
  • Step S401 Perform a data augmentation operation on each remote sensing satellite map in the positive training data to obtain an augmented map, where the data augmentation operation includes at least a warping operation, a flipping operation, and a noise adding operation;
  • each remote sensing satellite map in the positive training data can be subjected to data augmentation operations.
  • the data augmentation operations can be operations such as distortion operations, flip operations, and noise addition operations on the remote sensing satellite images. These operations are similar to the existing data augmentation operations and will not be described in detail here.
  • the data augmentation operation on the one hand, when the number of remote sensing satellite images is small, the number of remote sensing satellite images can be increased, thereby increasing the training data, so that the model can be fully trained.
  • the trained model can be Identifying various farms also improves the versatility of the model.
  • Step S402 using the positive training data and the augmented map to perform preliminary training on the model to be trained.
  • An augmented map obtained by augmenting data from a remote sensing satellite map is the same as the labeled data of the remote sensing satellite.
  • Both the remote sensing satellite map and the augmented map in the positive training data are used as the input data of the model to be trained, and the labeled data is used to correct the output of the model to be trained, so as to perform preliminary training on the model to be trained.
  • training data of negative examples can also be augmented by this augmentation operation.
  • a feasible risk prediction process in this embodiment includes three stages: data preparation, target detection, and risk prediction.
  • the data preparation stage is to prepare the training data used to train the farm detection model, as well as other influencing factors information, such as waterway network information and village information.
  • the latitude and longitude of villages, road networks (including highways, national highways, etc.) and water networks in the province are used to assist in epidemic risk prediction after conversion; use ArcGIS geographic satellite labeling software to label 34 remote sensing satellite maps to indicate specific breeding Generate annotated geographic satellite images as experimental comparison, and generate annotated mask with a black background as the experimental input; crop the annotated map, 32 training maps overlapped and cut into multiple small pieces of 1024*1024 , Other maps are cropped into multiple small blocks with a size of 8192*8192 without overlap; select small blocks with farms and small labeled masks as positive training data, and select small blocks without farms as negative examples Training data.
  • the target detection stage mainly uses the farm detection model to detect the farm in the map of each area, and obtain the mask map of the farm.
  • the density value can specifically include stocking density, high-speed density, village density, water network density, and national highway density. Then input these density values into the risk prediction model, and finally output the risk prediction results of the farm to be predicted.
  • the other influencing factor information includes waterway network information
  • the step S30 includes:
  • Step S301 determining the characteristic value of the waterway network density of the preset range according to the waterway network information and the preset range;
  • the other influencing factor information may be waterway network information. Then, predict the risk situation of the farm to be predicted based on the breeding information and waterway network information within the preset range.
  • the waterway network density characteristic value of the preset range is determined according to the waterway network information and the preset range.
  • the characteristic value of waterway network density can be the result of dividing the total length of the waterway network within the preset range by the total area of the preset range, or it can be the total area of the waterway network within the preset range divided by the total area of the preset range The results obtained.
  • the calculation process of area and length can be as follows: the total area of the preset range can be determined according to the location information of the preset range, for example, the total area of the preset range can be calculated according to the latitude and longitude interval of the preset range; according to the information in the waterway network
  • the location information of the waterway network determines the length of each waterway network.
  • a river or a road can be represented by the latitude and longitude of two points.
  • the length of the river or road is calculated based on the latitude and longitude of the two points. Add the length of the road to get the length of the target waterway network; you can also determine the total area of the waterway network according to the location information of the waterway network in the waterway network information.
  • a river or a road can be represented by the latitude and longitude interval. Calculate the area of the river or road according to the latitude and longitude interval, and add the area of all the rivers or roads to get the total area of the waterway network.
  • Step S302 Determine the density characteristic value of the farm in the preset range according to the farming information and the preset range;
  • the density characteristic value of the preset range is determined.
  • the characteristic value of the density of the farm may be the result of dividing the total area of the farm within the preset range by the total area of the preset range.
  • the calculation method for the total area of the farms within the preset range can be: each farm can be represented by the latitude and longitude interval, and the area of each farm is calculated according to the latitude and longitude interval, and then the various areas are added together to obtain the target farm The total area.
  • the characteristic value of the density of the breeding farm may also be the result of dividing the total breeding amount of the breeding farm in the preset range by the total area of the preset range, that is, the characteristic value of the density of the breeding farm may be the breeding density.
  • the calculation method of the total breeding amount of the target breeding farm may be: determining the area of each breeding farm according to the location information of each breeding farm in the breeding information, that is, the calculation process of the total area of the aforementioned breeding farm.
  • the corresponding relationship between the area of the breeding farm and the feeding amount can be set in advance, for example, 1.5 square meters can be set to produce one pig. According to the calculated area of each breeding farm and the corresponding relationship, the total breeding capacity of each breeding farm is determined.
  • the total breeding capacity of each breeding farm is 100 pigs; the total breeding capacity is divided by the total area, and the result obtained is used as the characteristic value of the density of the breeding farm.
  • Step S303 Input the characteristic value of the density of the waterway network and the characteristic value of the density of the breeding farm into the risk prediction model to obtain the risk prediction result of the farm to be predicted.
  • the characteristic value of the density of the waterway network and the characteristic value of the density of the farm are input into the risk prediction model to obtain the risk prediction result of the farm to be predicted.
  • the characteristic value of the density of the farm and the characteristic value of the waterway network density are input into the risk prediction model to calculate the The risk prediction result of the farm realizes the prediction of the risk of the farm based on the density characteristics of the preset range of the farm and the waterway network, combined with the actual infectious characteristics of animal infectious diseases, so that the predicted risk prediction results are more accurate.
  • the risk prediction model includes preset weight values respectively corresponding to the characteristic value of the waterway network density and the characteristic value of the density of the breeding farm, and the step S303 includes:
  • Step S3031 Input the characteristic value of the density of the waterway network and the characteristic value of the density of the breeding farm into the risk prediction model to call the risk prediction model based on the weight value, the characteristic value of the density of the waterway network and the breeding The farm density characteristic value is calculated to obtain the risk coefficient of the farm to be predicted;
  • the risk prediction model may be a linear model, and the weights corresponding to the density feature values of the farms, the weights corresponding to the water network density feature values, and the weights corresponding to the road network density feature value can be respectively set.
  • the characteristic value of the density of the waterway network and the characteristic value of the density of the breeding farm are input into the risk prediction model to call the risk prediction model based on each weight value, the characteristic value of the density of the waterway network and the breeding
  • the field density feature value is calculated to obtain the risk coefficient of the farm to be predicted.
  • the density feature value of the farm can be multiplied by the corresponding weight
  • the water network density feature value can be multiplied by the corresponding weight
  • the road network density feature value can be multiplied by the corresponding weight
  • each of the multiplications can be obtained.
  • the results are added together, and the result of the addition is used as the risk coefficient, that is, the nature of the risk prediction model can be a linear model. It should be noted that each weight value in the risk prediction model can be set according to specific experience.
  • Step S3032 using the risk coefficient as the risk prediction result of the farm to be predicted, or after determining the risk level according to the risk coefficient and the preset correspondence between the coefficient and the level, the risk level is taken as the risk level. Describe the risk prediction results of the farm to be predicted.
  • the risk coefficient can be directly used as the risk prediction result of the farm to be predicted. It can also be that the corresponding relationship between the risk coefficient and the level is preset, for example, the risk coefficient is sorted from low to high, and the lowest 1/4 is divided into the first level, and 1/4-1/2 is divided into the second Level, 1/2-3/4 is classified as the third level, and the highest 1/4 is classified as the fourth level. It should be noted that the number of grades and the dividing line of each grade can be selected based on specific experience. According to the obtained risk coefficient and the corresponding relationship between the coefficient and the level, the risk level corresponding to the risk coefficient is determined, and the determined risk level is used as the risk prediction result of the farm to be predicted. According to the actual needs of users, the risk prediction results can be different, and the risk level may be more intuitive for some users, which is more conducive to users to make targeted prevention and control programs.
  • an embodiment of the present application also proposes a risk prediction device for a farm.
  • the risk prediction device for a farm includes:
  • the input module 10 is used to input the map to be detected into the farm detection model to obtain the farm information in the map to be detected;
  • the determining module 20 is used to determine the farm to be predicted according to the farm information, and to determine the farming information within a preset range around the farm to be predicted, and to obtain other impacts within the preset range except the farm Factor information
  • the prediction module 30 is configured to input the breeding information and the other influencing factor information into a preset risk prediction model to obtain the risk prediction result of the farm to be predicted.
  • the input module 10 includes:
  • the first input unit is used to input the map to be detected into the detection model of the farm to obtain the classification category of each pixel in the map to be detected, wherein the classification category is used to indicate whether the corresponding pixel belongs to the farm ;
  • the first determining unit is configured to determine the farm information in the map to be detected according to the classification category.
  • the determining unit includes:
  • the first determining subunit is configured to determine the target pixels belonging to the breeding farm in the map to be detected according to the classification category;
  • the clustering subunit is used to cluster the target pixels by using a preset clustering algorithm to obtain each cluster group;
  • the second determining subunit is configured to determine the farm information in the map to be detected according to each of the cluster groups.
  • the device for predicting the risk of a farm further includes:
  • the preliminary training module is used for preliminary training of the model to be trained using positive training data, where the positive training data includes a plurality of remote sensing satellite maps containing farms collected in advance, and the farms corresponding to each remote sensing satellite map Label data;
  • the adjustment module adopts negative training data, or uses the negative training data and the positive training data to adjust the model to be trained after preliminary training, wherein the negative training data includes a plurality of pre-collected Does not include remote sensing satellite maps of farms;
  • the definition module is used for when it is detected that the adjusted model to be trained meets the preset model conditions, the adjusted model to be trained is used as the farm detection model; otherwise, the steps are executed again based on the adjusted model to be trained: Use positive training data for preliminary training of the training model. .
  • the preliminary training module includes:
  • An augmentation unit configured to perform a data augmentation operation on each remote sensing satellite map in the normal training data to obtain an augmented map, wherein the data augmentation operation includes at least a warping operation, a flipping operation, and a noise adding operation;
  • the preliminary training unit is configured to use the positive training data and the augmented map to perform preliminary training on the model to be trained.
  • a second determining unit configured to determine the waterway network density characteristic value of the preset range according to the waterway network information and the preset range;
  • the second determining unit is configured to determine the density characteristic value of the farm in the preset range according to the farming information and the preset range;
  • the second input unit is configured to input the characteristic value of the density of the waterway network and the characteristic value of the density of the breeding farm into the risk prediction model to obtain the risk prediction result of the farm to be predicted.
  • the risk prediction model includes preset weight values respectively corresponding to the characteristic value of the waterway network density and the characteristic value of the density of the breeding farm, and the second input unit includes:
  • the input subunit is used to input the waterway network density feature value and the breeding farm density feature value into the risk prediction model to call the risk prediction model based on the weight value, the waterway network density feature value and
  • the risk coefficient of the farm to be predicted is calculated by calculating the characteristic value of the density of the farm;
  • the third determining subunit is used to use the risk coefficient as the risk prediction result of the farm to be predicted, or after determining the risk level according to the risk coefficient and the preset correspondence between the coefficient and the level, the risk level is determined.
  • the risk level is used as the risk prediction result of the farm to be predicted.
  • an embodiment of the present application also proposes a computer-readable storage medium, the storage medium stores a farm risk prediction program, and the farm risk prediction program is executed by a processor to realize the following farm risk prediction Method steps.
  • the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, The optical disc) includes several instructions to enable a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the method described in each embodiment of the present application.
  • a terminal device which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • General Business, Economics & Management (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Software Systems (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computational Linguistics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Tourism & Hospitality (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Game Theory and Decision Science (AREA)
  • Educational Administration (AREA)
  • Development Economics (AREA)
  • Remote Sensing (AREA)
  • Agronomy & Crop Science (AREA)
  • Animal Husbandry (AREA)
  • Marine Sciences & Fisheries (AREA)

Abstract

一种养殖场风险预测方法、装置、设备及存储介质，该方法包括：将待检测地图输入养殖场检测模型得到所述待检测地图中的养殖场信息（S10）；根据养殖场信息确定待预测养殖场，以及确定所述待预测养殖场周围预设范围内的养殖信息，并获取所述预设范围内除养殖场外的其他影响因素信息(S20)；将所述养殖信息和所述其他影响因素信息输入预设的风险预测模型得到所述待预测养殖场的风险预测结果（S30）。

Description

养殖场风险预测方法、装置、设备及存储介质
本申请要求2020年2月28日申请的，申请号为202010130767.1，名称为“养殖场风险预测方法、装置、设备及存储介质”的中国专利申请的优先权，在此将其全文引入作为参考。
技术领域
本申请涉及人工智能技术领域，尤其涉及一种养殖场风险预测方法、装置、设备及存储介质。
 背景技术
非洲猪瘟是由非洲猪瘟病毒引起的猪的一种急性、热性、高度接触性动物传染病，相关致死率高达100%。从中国动物卫生与流行病学中心确诊中国第一例非洲猪瘟以来，中国部分省份陆续发生非洲猪瘟疫情，损失严重。类似于非洲猪瘟的这种动物传染病一旦发现，就需要快速地、准确地采取防控措施，以避免疫情持续扩大而造成社会各方面的不良影响。而为做出快速、准确地防控措施，就需要快速、准确地了解到各个养殖场的风险情况。但是目前的状况是对养殖场的风险评估，需要专家实地考察养殖场的情况后进行评估，或者依赖养殖户的上报信息。然而专家实地考察的方案显然效率低下，难以应对突发且蔓延迅速的疫情；养殖户的主动上报则会有养殖户瞒报养殖信息的情况，导致风险评估准确率低。无法快速、准确地获得养殖场的风险情况，将导致防控工作效果差。
 技术解决方案
本申请的主要目的在于提供一种养殖场风险预测方法、装置、设备及存储介质，旨在解决目前针对动物疫情，无法快速、准确地获得养殖场的风险情况，从而导致防控工作效果差的问题。
为实现上述目的，本申请提供一种养殖场风险预测方法，所述养殖场风险预测方法包括以下步骤：
将待检测地图输入养殖场检测模型得到所述待检测地图中的养殖场信息；
根据所述养殖场信息确定待预测养殖场，以及确定所述待预测养殖场周围预设范围内的养殖信息，并获取所述预设范围内除养殖场外的其他影响因素信息；
将所述养殖信息和所述其他影响因素信息输入预设的风险预测模型得到所述待预测养殖场的风险预测结果。
为实现上述目的，本申请还提供一种养殖场风险预测装置，所述养殖场风险预测装置包括：
输入模块，用于将待检测地图输入养殖场检测模型得到所述待检测地图中的养殖场信息；
确定模块，用于根据所述养殖场信息确定待预测养殖场，以及确定所述待预测养殖场周围预设范围内的养殖信息，并获取所述预设范围内除养殖场外的其他影响因素信息；
预测模块，用于将所述养殖信息和所述其他影响因素信息输入预设的风险预测模型得到所述待预测养殖场的风险预测结果。
为实现上述目的，本申请还提供一种养殖场风险预测设备，所述养殖场风险预测设备包括：存储器、处理器及存储在所述存储器上并可在所述处理器上运行的养殖场风险预测程序，所述养殖场风险预测程序被所述处理器执行时实现如上所述的养殖场风险预测方法的步骤。
此外，为实现上述目的，本申请还提出一种计算机可读存储介质，所述计算机可读存储介质上存储有养殖场风险预测程序，所述养殖场风险预测程序被处理器执行时实现如上所述的养殖场风险预测方法的步骤。
本申请中，通过将待检测地图输入预先训练得到的养殖场检测模型，得到待检测地图中的养殖场信息；根据养殖场信息确定待预测养殖场以及确定待预测养殖场周围预设范围内的养殖信息，并获取预设范围内的其他影响因素信息；将养殖信息和其他影响因素信息输入预设的风险检测模型，得到待预测养殖场的风险预测结果。基于快速、准确的风险预测结果，使得疫情防控工作能够更加快速、有针对性地展开，避免疫情失控，提高了农产品的质量安全，改善了生态环境。
 附图说明
图1是本申请实施例方案涉及的硬件运行环境的结构示意图；
图2为本申请养殖场风险预测方法第一实施例的流程示意图；
图3为本申请养殖场风险预测装置较佳实施例的功能示意图模块图；
图4为本申请实施例涉及的一种风险预测流程示意图；。
本申请目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
 本发明的实施方式
应当理解，此处所描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
如图1所示，图1是本申请实施例方案涉及的硬件运行环境的设备结构示意图。
需要说明的是，本申请实施例养殖场风险预测设备可以是智能手机、个人计算机和服务器等设备，在此不做具体限制。
如图1所示，该养殖场风险预测设备可以包括：处理器1001，例如CPU，网络接口1004，用户接口1003，存储器1005，通信总线1002。其中，通信总线1002用于实现这些组件之间的连接通信。用户接口1003可以包括显示屏（Display）、输入单元比如键盘（Keyboard），可选用户接口1003还可以包括标准的有线接口、无线接口。网络接口1004可选的可以包括标准的有线接口、无线接口（如WI-FI接口）。存储器1005可以是高速RAM存储器，也可以是稳定的存储器（non-volatile memory），例如磁盘存储器。存储器1005可选的还可以是独立于前述处理器1001的存储装置。
本领域技术人员可以理解，图1中示出的设备结构并不构成对养殖场风险预测设备的限定。
如图1所示，作为一种计算机存储介质的存储器1005中可以包括操作系统、网络通信模块、用户接口模块以及养殖场风险预测程序。
在图1所示的设备中，用户接口1003主要用于与客户端进行数据通信；网络接口1004主要用于与服务器建立通信连接；而处理器1001可以用于调用存储器1005中存储的养殖场风险预测程序，并执行以下操作：
将待检测地图输入养殖场检测模型得到所述待检测地图中的养殖场信息；
根据所述养殖场信息确定待预测养殖场，以及确定所述待预测养殖场周围预设范围内的养殖信息，并获取所述预设范围内除养殖场外的其他影响因素信息；
将所述养殖信息和所述其他影响因素信息输入预设的风险预测模型得到所述待预测养殖场的风险预测结果。
基于上述的结构，提出养殖场风险预测方法的各个实施例。
参照图2，图2为本申请养殖场风险预测方法第一实施例的流程示意图。
本申请实施例提供了养殖场风险预测方法的实施例，需要说明的是，虽然在流程图中示出了逻辑顺序，但是在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤。本申请养殖场风险预测方法各个实施例的执行主体可以是智能手机、个人计算机和服务器等设备，为便于描述，以下各实施例中省略执行主体进行阐述。在本实施例中，养殖场风险预测方法包括：
步骤S10，将待检测地图输入养殖场检测模型得到所述待检测地图中的养殖场信息，其中，所述养殖场检测模型是预先训练得到的；
在本实施例中，对于某个区域（待检测区域），需要预测该区域内的各个养殖场的风险情况，可以获取该区域的地图，作为待检测地图。其中，风险可以指该养殖场在猪瘟等动物传染病传播的情况下，可能受到的影响的大小程度。获取地图的方式有多种，例如可以是从地图软件中下载到的地图，也可以是通过遥感卫星获取遥感卫星地图。预先可以训练一个养殖场检测模型，用于检测图片中的养殖场。养殖场检测模型可以采用常用的图像目标检测模型，如可采用语义图像分割模型DeepLab-v3+。养殖场检测模型的训练方法可以采用现有的机器学习模型的训练方法。
将待检测地图输入养殖场检测模型，根据养殖场检测模型的输出结果得到待检测地图中的养殖场信息。具体地，养殖场检测模型的输出结果可以是待检测地图中养殖场的位置，再根据待检测地图的比例尺，以及待检测地图所对应的实际地理位置，可以计算得到待检测区域内的养殖场的实际地理位置。那么获取到的养殖场信息可以至少包括养殖场的实际地理位置。其中，可以是将养殖场以一个点（养殖场的中心点）来看待，养殖场的实际地理位置即该点的位置；也可以是将养殖场以一个区域来看待，养殖场的实际地理位置即该区域的位置。实际地理位置可以是用经纬度表示，或者用预先制定的特定坐标体系下的坐标表示，等等。
步骤S20，根据所述养殖场信息确定待预测养殖场，以及确定所述待预测养殖场周围预设范围内的养殖信息，并获取所述预设范围内除养殖场外的其他影响因素信息；
在获取到待检测地图中的养殖场信息后，可以根据养殖场信息确定待预测养殖场。具体地，根据养殖场信息可以确定待检测区域中包括多少个养殖场，以及各个养殖场的实际地理位置等。可以从检测到的待检测区域中的多个养殖场中选择一个作为待预测养殖场，具体可以是将各个养殖场依次作为待预测养殖场，也可以是基于用户的选择确定待预测养殖场。
对于确定的待预测养殖场，要预测该养殖场的风险，本实施例中采用的原理是可以以该养殖场周围的区域的相关数据来作为评判该养殖场风险的依据。具体地，可以先确定待预测养殖场周围的一个预设范围。预设范围的确定方法可以有多种；例如，可以预先设置一个半径值，以该养殖场作为圆心，以该半径值为半径，划分一个圆形区域，即可作为该养殖场的预设范围，根据该养殖场的实际地理位置，以及该半径值，就可以确定预设范围的实际地理位置；还可以是以该养殖场作为矩形的中心，预先设置矩阵边长值，以此来划分出一个矩阵区域作为该养殖场的预设范围；还有其他的可行方法，根据实际的应用场景不同，可以采取不同的方法来确定预设范围，在此不做一一列举。
在确定了待预测养殖场的预设范围后，可以根据待检测地图中的养殖场信息，确定该预设范围内的养殖信息。具体地，根据预设范围的实际地理位置，以及养殖场信息中各个养殖场的实际地理位置，可以检测各个养殖场是否落入了预设范围，将落入预设范围的养殖场确定为目标养殖场，那么可以从养殖场信息中提取出目标养殖场的信息作为预设范围内的养殖信息。例如，当实际地理位置用经纬度表示时，预设范围的位置可以是由经度区间和纬度区间构成，那么需要判断各个养殖场的经度是否落入该经度区间，纬度是否落入该维度区间，若均落入，则确定该养殖场落入预设范围。
预先可以规定各种风险影响因素类型，如，养殖场、水网、路网和村庄等等，这些是不同类型的影响因素。其中，水网可以是指各种大小的河流、湖泊等，路网可以是指各种类型的道路，如国道、高速等。以下将水网和路网统称为“水路网”，可以表示将水网和路网统一看待，也可以表示将其分开看待。
在获取到预设范围内的养殖信息后，可以获取预设范围内除养殖场以外的其他风险影响因素的信息，即其他影响因素信息。具体地，可以预先采集待检测区域内的其他影响因素信息，其中，其他影响因素信息的获取方式可以是从电子地图数据中进行提取，也可以是人工采集。其他影响因素信息具体可以包括其他的风险影响因素的位置、大小等等，根据风险影响因素的类型不同数据不同，例如，采集的村庄信息可以包括村庄的位置信息、面积等，采集的水路网信息可以包括水路网的位置信息和长度等。
根据待检测区域内其他影响因素信息中各个其他影响因素的实际地理位置，和预设范围的实际地理位置，确定预设范围内的其他影响因素信息。
步骤S30，将所述养殖信息和所述其他影响因素信息输入预设的风险预测模型得到所述待预测养殖场的风险预测结果。
将预设范围内的养殖信息和其他影响因素信息输入预设的风险预测模型，得到待预测养殖场的风险预测结果。具体地，预先可以设置一个风险预测模型，风险预测模型可以是采用线性模型或神经网络模型等模型结构，如果是神经网络模型，则需要预先对神经网络模型进行训练，采取训练后得到的模型来进行风险预测，而神经网络模型的训练过程可参考现有的神经网络模型训练过程，在此不作详细赘述。风险预测模型的设置原理是：当输入数据表征待预测养殖场的预设范围内的目标影响因素越多，越密集时，输出的风险预测结果表征待预测养殖场的风险越高。风险预测模型的输入数据可以根据情况进行设置，例如，输入数据可以设置为预设范围内的养殖场总面积、水路网总面积。可以将养殖信息和其他影响因素信息进行处理，处理为与风险预测模型输入数据匹配的形式，再将处理后的数据输入风险预测模型，经过风险预测模型对数据的处理，输出待预测养殖场的风险预测结果。风险预测结果的形式可以是多种，例如，可以是一个分数，也可以是等级。可以预先规定各种分数或等级所对应的风险程度，例如，分数是0到10分，从0到10风险程度逐渐变大。
在本实施例中，通过将待检测地图输入预先训练得到的养殖场检测模型，得到待检测地图中的养殖场信息；根据养殖场信息确定待预测养殖场以及确定待预测养殖场周围预设范围内的养殖信息，并获取预设范围内的其他影响因素信息；将养殖信息和其他影响因素信息输入预设的风险检测模型，得到待预测养殖场的风险预测结果。在本实施例中，通过训练一个养殖场检测模型来从地图中检测出各个养殖场，获取养殖场信息，作为风险预测的数据来源，减少了对养殖户上报信息的依赖，提高了养殖场的管理效率，也提高了风险预测效率；精准及时统计养殖场信息，以便为养殖主提供生产、管理等方案，提高畜牧业养殖经营效率。并通过设置一个风险预测模型，来对待预测养殖场预设范围内的养殖信息和其他影响因素信息进行处理，得到待预测养殖场的风险预测结果，无需专家实地考察每个养殖场的情况，从而提高了养殖场风险预测的效率，更能够应对疫情蔓延迅速的情况；相比于养殖户主动上报风险情况的方案，本实施例中采用统一的风险预测模型，能够更加准确地预测各个养殖场的风险情况；基于快速、准确的风险预测结果，使得疫情防控工作能够更加快速、有针对性地展开，避免疫情失控，提高了农产品的质量安全，改善了生态环境。
进一步地，可以预先针对不同风险预测结果设置不同的防控方案，当得到待预测养殖场的风险预测结果后，根据风险预测结果从各个防控方案中匹配合适的防控方案，将匹配到的防控方案确定为待预测养殖场的目标防控方案，并可将目标防控方案输出。输出方式可以是输出至显示屏中显示，也可以是以语音方式输出，输出方式可以根据具体应用场景进行不同的设置。通过直接将待预测养殖场的防控方案输出显示，可以使得用户直观地获知防控方案，从而根据防控方案进行防控工作。特别地，对于需要进行防控措施，但是又不了解有哪些防控手段，或者不解养殖场的风险情况的用户，例如养殖户，或养殖场周边的居民，则也可以很方便地直接获取与养殖场的风险情况匹配的防控手段，从而更加利于防控工作的开展。
基于本实施例中的风险预测方法，可以以手机或电脑应用软件的形式进行实际应用，从而使得用户能够基于安装的应用程序，输入待预测养殖场的相关信息，例如，位置信息，即可直接得到待预测养殖场的风险预测结果，甚至是防控方案。
进一步地，基于上述第一实施例，提出本申请养殖场风险预测方法第二实施例，在本实施例中，所述步骤S10包括：
步骤S102，将待检测地图输入所述养殖场检测模型，得到所述待检测地图中各像素点的分类类别，其中，所述分类类别用于表示对应的像素点是否属于养殖场；
养殖场检测模型可以是一个二分类模型，将待检测地图输入养殖场检测模型后，养殖场检测模型可以输出待检测地图中各个像素点的分类结果，分类结果用于表示对应的像素点是属于养殖场还是不属于养殖场。例如，分类结果可以是0和1，1表示对应的像素点是属于养殖场，0表示对应的像素点不属于养殖场。
步骤S103，根据所述分类类别确定所述待检测地图中的养殖场信息。
根据各个像素点的分类类别，可以确定待检测地图中的养殖场信息。具体的，可以是根据各个像素点的分类类别，从各个像素点中确定属于养殖场的目标像素点；然后根据待检测地图的比例尺，可以确定每个像素点代表实际的多少区域，再根据待检测地图的经纬度区间，即可确定每个目标像素点的经纬度区间，由于每个目标像素点对应的是养殖场区域，所以目标像素点的经维度区间即可作为其对应的养殖场区域的经纬度区间。
进一步地，所述步骤S103包括：
步骤S1031，根据所述分类类别确定所述待检测地图中属于养殖场的目标像素点；
根据各个像素点的分类类别，从各个像素点中确定属于养殖场的目标像素点，例如，当分类类别有0和1两种时，将所有分类类别为1的像素点作为目标像素点。
步骤S1032，采用预设聚类算法对所述目标像素点进行聚类得到各个聚类群；
可以预先设置一个聚类算法，采用聚类算法对目标像素点进行聚类，得到各个聚类群。需要说明的是，当待检测地图中一个像素点实际对应的区域比较小时，可能一个像素点表示的是一个养殖舍（即比养殖场更小的单位），可能一个养殖场的区域对应多个目标像素点，此时，可以采用聚类算法，对识别出的多个像素点进行聚类，或者对多个像素点所对应的经纬度区间进行聚类，将多个距离较近的养殖舍聚类为一个聚类群。聚类算法可以采用常用的聚类算法，例如近邻聚类法。
步骤S1033，根据各所述聚类群确定所述待检测地图中的养殖场信息。
将聚类得到的各个聚类群，分别作为一个养殖场，养殖场的经纬度可以是该聚类群的中心，养殖场的面积可以是该聚类群范围内所有养殖舍的面积的总和，那么待检测养殖场中的养殖场信息可以至少包括各个养殖场的位置和面积。
进一步地，所述步骤S10之前，还包括：
步骤S40，采用正例训练数据对待训练模型进行初步训练，其中，所述正例训练数据包括预先采集到的多张包含养殖场的遥感卫星地图，以及各遥感卫星地图对应的养殖场标注数据；
进一步地，在本实施例中，养殖场检测模型的训练方式可如下：
预先采集多张包含养殖场的遥感卫星地图，以及采集遥感卫星地图中的养殖场的标注数据，即养殖场标注数据。养殖场标注数据可以是与遥感卫星图片对应的掩膜图，掩膜图中有遥感卫星图片的每个像素点对应的分类类别，如0表示该像素点不属于养殖场，1表示该像素点属于养殖场，掩膜图可以采用不同的颜色表示不同的类别。将多张包含养殖场的遥感卫星地图以及养殖场数据作为正例训练数据。
先采用正例训练数据来对待训练模型进行初步训练。其中，待训练模型的结构可采用常用的语义分割模型，例如语义图像分割模型DeepLab-v3+。初步训练的过程可采用现有的机器学习模型训练过程。
步骤S50，采用负例训练数据、或者采用所述负例训练数据和所述正例训练数据对初步训练后的待训练模型进行调整，其中，所述负例训练数据包括预先采集到的多张不包含养殖场的遥感卫星地图；
预先采集多张不包含养殖场的遥感卫星地图，作为负例训练数据，负例训练数据与正例训练数据的比例可以根据实际经验设置，例如负例训练数据可以是正例训练数据的十分之一。在初步训练后，采用负例训练数据和上述正例训练数据一起来对初步训练后的语义分割模型进行调整，或者单独采用负例训练数据来对初步训练后的语义分割模型进行调整，此时的调整可以是微调。具体地，微调即采用正例训练数据和负例训练数据一起来训练，训练过程也可以采用现有的机器学习模型的训练过程。需要说明的是，在微调时，可以根据正例训练数据与负例训练数据的比例，对于模型的超参数进行调整，调整后再进行微调训练，例如，负例训练数据可以是正例训练数据的十分之一时，微调阶段的学习率也可以设置为初步训练阶段的十分之一。
步骤S60，当检测到调整后的待训练模型符合预设模型条件时，将调整后的待训练模型作为所述养殖场检测模型，否则基于调整后的待训练模型再执行所述步骤：采用正例训练数据对待训练模型进行初步训练。
检测调整后的待训练模型是否符合预设模型条件，若符合预设模型条件，则将调整后的待训练模型作为养殖场检测模型，以便后续采用养殖场检测模型进行养殖场的检测。否则，若不符合预设模型条件，则再采用正例训练数据对调整后的待训练模型进行初步训练，之后再进行调整训练，直到检测到符合预设模型条件时为止。其中，预设模型条件可以是预先根据对模型的性能需求设置的条件，如可以将模型的损失函数收敛作为一个条件，还可以是将常用的检测模型性能的客观指标作为条件，如准确率、召回率和IOU（Intersection over Union,交并比）等客观指标。其中，各个客观指标的计算方式可参考现有的指标计算方式，在此不做详细赘述。
在本实施例中，通过预先训练一个养殖场检测模型，对待检测地图进行检测识别，得到待检测地图中的养殖场信息，从而可以实现智能地、快速地获取各个地区的养殖场信息，进而提高了待预测养殖场的风险预测效率；并且相比于依赖养殖户主动上报的方式，本实施例中的方案能够更加全面地获取养殖场信息，从而使得风险预测所依据的数据更加准确、全面，使得风险预测结果更加准确。
进一步地，所述步骤S40包括：
步骤S401，对所述正例训练数据中的各遥感卫星地图进行数据增广操作得到增广地图，其中，所述数据增广操作至少包括扭曲操作、翻转操作和加噪操作；
进一步地，对于正例训练数据中的各个遥感卫星地图，可以将各个遥感卫星地图进行数据增广操作，数据增广操作可以是对遥感卫星图像进行扭曲操作、翻转操作和加噪操作等操作，这些操作与现有的数据增广操作类似，在此不做详细赘述。通过数据增广操作一方面可以使得在遥感卫星图像数量较少时，增加遥感卫星图像的数量，从而使得训练数据变多，使得模型得到充分的训练，另一方面，可以使得训练得到的模型能够识别各种不同的养殖场，也即提高了模型的通用性。
步骤S402，采用所述正例训练数据和所述增广地图对待训练模型进行初步训练。
由一个遥感卫星地图进行数据增广得到的增广地图，增广地图的标注数据与该遥感卫星的标注数据相同。将正例训练数据中的遥感卫星地图和增广地图均作为待训练模型的输入数据，采用标注数据来纠正待训练模型的输出，以对待训练模型进行初步训练。
进一步地，对于负例训练数据同样可采用该增广操作进行增广。
如图4所示，为本实施例中一种可行的风险预测流程，包括数据准备、目标检测和风险预测三个阶段。
1、数据准备。
数据准备阶段即准备用于训练养殖场检测模型的训练数据，以及其他影响因素信息，如水路网信息和村庄信息等。
具体地，如图4所示，可以下载常用地图软件上广东省18级1:50000卫星地图，其中32个作为训练数据，2个作为验证数据，其他作为测试数据；同时，可解析地图上广东省内村庄、路网（包括高速公路、国道等）和水网的经纬度，用于转换后辅助疫情风险预测；利用ArcGIS地理卫星标注软件，对34张遥感卫星地图进行数据标注，标明具体的养殖舍，生成带有标注的地理卫星图片作为实验比照，生成背景全黑的标注掩膜作为实验输入；裁剪标注后的地图，32个训练地图有重叠地裁成多个大小为1024*1024的小块，其他地图无重叠地裁剪成多个大小为8192*8192的小块；挑选有养殖场的小块，以及小块的标注掩膜作为正例训练数据，挑选没有养殖场的小块作为负例训练数据。
采用正例训练数据进行初步训练，再采用负例训练数据进行微调，再计算各个客观指标，对模型进行评估，选取最佳的模型作为最终的养殖场检测模型。
2、目标检测。
目标检测阶段主要是采用养殖场检测模型来检测各个区域的地图中的养殖场，得到养殖舍掩膜图。
3、风险预测。
对于待预测养殖场，先确定该待预测养殖场周围的一个范围，再采用聚类算法对目标检测阶段获取到的养殖信息以及其他风险影响因素信息进行聚类计算，得到该范围内的各种密度值，具体可以包括饲养密度、高速密度、村庄密度、水网密度和国道密度。再将这些密度值输入风险预测模型，最终输出待预测养殖场的风险预测结果。
进一步地，基于上述第一和第二实施例，提出本申请养殖场风险预测方法第三实施例，在本实施例中，所述其他影响因素信息包括水路网信息，所述步骤S30包括：
步骤S301，根据所述水路网信息和所述预设范围确定所述预设范围的水路网密度特征值；
进一步地，在本实施例中，其他影响因素信息可以是水路网信息。那么，根据预设范围内的养殖信息和水路网信息来预测待预测养殖场的风险情况。
具体地，根据水路网信息和预设范围确定预设范围的水路网密度特征值。其中，水路网密度特征值可以是预设范围内水路网的总长度除以预设范围的总面积得到的结果，也可以是预设范围内水路网的总面积除以预设范围的总面积得到的结果。面积和长度的计算过程可如下：可以根据预设范围的位置信息确定预设范围的总面积，如可以根据预设范围的经纬度区间，计算得到预设范围的总面积；根据水路网信息中各水路网的位置信息，确定各水路网的长度，例如，一条河流或一条道路可以是由两个点的经纬度来表示，此时根据两个点的经纬度计算河流或道路的长度，将所有河流或道路的长度相加，得到目标水路网的长度；也可以根据水路网信息中各水路网的位置信息，确定水路网的总面积，例如，一条河流或一条道路可以是由经纬度区间来表示，此时根据经纬度区间来计算河流或道路的面积，将所有河流或道路的面积相加，得到水路网的总面积。
步骤S302，根据所述养殖信息和所述预设范围确定所述预设范围的养殖场密度特征值；
根据养殖信息和预设范围确定预设范围的养殖场密度特征值。具体地，养殖场密度特征值可以是预设范围内养殖场的总面积除以预设范围的总面积得到的结果。预设范围内养殖场的总面积计算方法可以是：各个养殖场可以是分别由经纬度区间来表示，根据经纬度区间计算得到各个养殖场的面积，然后将各个面积相加，即得到了目标养殖场的总面积。
养殖场密度特征值还可以是预设范围养殖场的总饲养量除以预设范围的总面积得到的结果，也即养殖场密度特征值可以是饲养密度。目标养殖场的总饲养量计算方式可以是：根据养殖信息中各个养殖场的位置信息确定各个养殖场的面积，也即上述养殖场的总面积计算过程。预先可以设置养殖场的面积与饲养量之间的对应关系，例如，设置1.5平方米产一头猪。根据计算得到各个养殖场的面积，以及该对应关系，确定各个养殖场的总饲养量。例如，若养殖场面积是150平方米，按照上述的例子，则各个养殖场的总饲养量是100头猪；将总饲养量除以总面积，得到的结果就作为养殖场密度特征值。
步骤S303，将所述水路网密度特征值和所述养殖场密度特征值输入所述风险预测模型得到所述待预测养殖场的风险预测结果。
将水路网密度特征值和养殖场密度特征值输入风险预测模型得到待预测养殖场的风险预测结果。在本实施例中，通过计算待预测养殖场周围预设范围的养殖场密度特征值和水路网密度特征值，将养殖场密度特征值和水路网密度特征值输入风险预测模型，计算待预测养殖场的风险预测结果，实现根据预设范围的养殖场和水路网的密度特征来预测养殖场的风险，结合了实际的动物传染病的传染特征，使得预测得到的风险预测结果更加准确。
进一步地，所述风险预测模型中包括与所述水路网密度特征值和所述养殖场密度特征值分别对应的预设权重值，所述步骤S303包括：
步骤S3031，将所述水路网密度特征值和所述养殖场密度特征值输入所述风险预测模型，以调用所述风险预测模型基于所述权重值、所述水路网密度特征值和所述养殖场密度特征值计算得到所述待预测养殖场的风险系数；
进一步地，风险预测模型可以是一个线性模型，可以分别设置养殖场密度特征值对应的权重，水网密度特征值对应的权重和路网密度特征值对应的权重。
在获取到水路网密度特征值和养殖场密度特征值后，将水路网密度特征值和养殖场密度特征值输入风险预测模型，以调用风险预测模型基于各个权重值、水路网密度特征值和养殖场密度特征值计算得到待预测养殖场的风险系数。具体地，可以将养殖场密度特征值与对应的权重相乘，将水网密度特征值与对应的权重相乘，将路网密度特征值与对应的权重相乘，再将相乘得到的各个结果进行相加，相加的结果作为风险系数，也即风险预测模型的本质可以是一个线性模型。需要说明的是，风险预测模型中的各个权重值可以是根据具体经验设置的。
步骤S3032，将所述风险系数作为所述待预测养殖场的风险预测结果，或根据所述风险系数，以及系数与等级之间的预设对应关系确定风险等级后，将所述风险等级作为所述待预测养殖场的风险预测结果。
在得到风险系数后，可以直接将风险系数作为待预测养殖场的风险预测结果。也可以是，预先设置了风险系数与等级之间的对应关系，例如，将风险系数从低到高排序，最低的1/4划分为第一级，1/4-1/2划分为第二级，1/2-3/4划分为第三级，最高的1/4划为第四级。需要说明的是，划分等级数目和每一个等级的分界线可以根据具体经验选取。根据得到的风险系数，以及该系数与等级间的对应关系，确定该风险系数对应哪一个风险等级，将确定的风险等级作为待预测养殖场的风险预测结果。根据用户的实际需求不同，风险预测结果可以不同，风险等级可能对于部分用户来说更直观，从而更利于用户来做出针对性的防控方案。
此外，此外本申请实施例还提出一种养殖场风险预测装置，参照图3，所述养殖场风险预测装置包括：
输入模块10，用于将待检测地图输入养殖场检测模型得到所述待检测地图中的养殖场信息；
确定模块20，用于根据所述养殖场信息确定待预测养殖场，以及确定所述待预测养殖场周围预设范围内的养殖信息，并获取所述预设范围内除养殖场外的其他影响因素信息；
预测模块30，用于将所述养殖信息和所述其他影响因素信息输入预设的风险预测模型得到所述待预测养殖场的风险预测结果。
进一步地，所述输入模块10包括：
第一输入单元，用于将待检测地图输入所述养殖场检测模型，得到所述待检测地图中各像素点的分类类别，其中，所述分类类别用于表示对应的像素点是否属于养殖场；
第一确定单元，用于根据所述分类类别确定所述待检测地图中的养殖场信息。
进一步地，所述确定单元包括：
第一确定子单元，用于根据所述分类类别确定所述待检测地图中属于养殖场的目标像素点；
聚类子单元，用于采用预设聚类算法对所述目标像素点进行聚类得到各个聚类群；
第二确定子单元，用于根据各所述聚类群确定所述待检测地图中的养殖场信息。
进一步地，所述养殖场风险预测装置还包括：
初步训练模块，用于采用正例训练数据对待训练模型进行初步训练，其中，所述正例训练数据包括预先采集到的多张包含养殖场的遥感卫星地图，以及各遥感卫星地图对应的养殖场标注数据；
调整模块，采用负例训练数据、或者采用所述负例训练数据和所述正例训练数据对初步训练后的待训练模型进行调整，其中，所述负例训练数据包括预先采集到的多张不包含养殖场的遥感卫星地图；
定义模块，用于当检测到调整后的待训练模型符合预设模型条件时，将调整后的待训练模型作为所述养殖场检测模型，否则基于调整后的待训练模型再执行所述步骤：采用正例训练数据对待训练模型进行初步训练。。
进一步地，所述初步训练模块包括：
增广单元，用于对所述正例训练数据中的各遥感卫星地图进行数据增广操作得到增广地图，其中，所述数据增广操作至少包括扭曲操作、翻转操作和加噪操作；
初步训练单元，用于采用所述正例训练数据和所述增广地图对待训练模型进行初步训练。
进一步地，所述其他影响因素信息包括水路网信息，所述预测模块30包括：
第二确定单元，用于根据所述水路网信息和所述预设范围确定所述预设范围的水路网密度特征值；
第二确定单元，用于根据所述养殖信息和所述预设范围确定所述预设范围的养殖场密度特征值；
第二输入单元，用于将所述水路网密度特征值和所述养殖场密度特征值输入所述风险预测模型得到所述待预测养殖场的风险预测结果。
进一步地，所述风险预测模型中包括与所述水路网密度特征值和所述养殖场密度特征值分别对应的预设权重值，所述第二输入单元包括：
输入子单元，用于将所述水路网密度特征值和所述养殖场密度特征值输入所述风险预测模型，以调用所述风险预测模型基于所述权重值、所述水路网密度特征值和所述养殖场密度特征值计算得到所述待预测养殖场的风险系数；
第三确定子单元，用于将所述风险系数作为所述待预测养殖场的风险预测结果，或根据所述风险系数，以及系数与等级之间的预设对应关系确定风险等级后，将所述风险等级作为所述待预测养殖场的风险预测结果。
本申请养殖场风险预测装置的具体实施方式的拓展内容与上述养殖场风险预测方法各实施例基本相同，在此不做赘述。
此外，本申请实施例还提出一种计算机可读存储介质，所述存储介质上存储有养殖场风险预测程序，所述养殖场风险预测程序被处理器执行时实现如下所述的养殖场风险预测方法的步骤。
本申请养殖场风险预测设备和计算机可读存储介质的各实施例，均可参照本申请养殖场风险预测方法各个实施例，此处不再赘述。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质（如ROM/RAM、磁碟、光盘）中，包括若干指令用以使得一台终端设备（可以是手机，计算机，服务器，空调器，或者网络设备等）执行本申请各个实施例所述的方法。
以上仅为本申请的优选实施例，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种养殖场风险预测方法，其中，所述养殖场风险预测方法包括以下步骤：
    将待检测地图输入养殖场检测模型得到所述待检测地图中的养殖场信息；
    根据所述养殖场信息确定待预测养殖场，以及确定所述待预测养殖场周围预设范围内的养殖信息，并获取所述预设范围内除养殖场外的其他影响因素信息；以及
    将所述养殖信息和所述其他影响因素信息输入预设的风险预测模型得到所述待预测养殖场的风险预测结果。
  2. 如权利要求1所述的养殖场风险预测方法，其中，所述将待检测地图输入养殖场检测模型得到所述待检测地图中的养殖场信息的步骤包括：
    将待检测地图输入养殖场检测模型，根据养殖场检测模型的输出结果得到待检测地图中的养殖场信息，其中，养殖场检测模型的输出结果是待检测地图中养殖场的位置，根据待检测地图的比例尺以及待检测地图所对应的实际地理位置，计算得到待检测区域内的养殖场的实际地理位置。
  3. 如权利要求1所述的养殖场风险预测方法，其中，所述根据所述养殖场信息确定待预测养殖场，以及确定所述待预测养殖场周围预设范围内的养殖信息，并获取所述预设范围内除养殖场外的其他影响因素信息包括：
    在获取到待检测地图中的养殖场信息后，根据养殖场信息确定待预测养殖场；确定待预测养殖场周围的一个预设范围；
    根据待检测地图中的养殖场信息，确定该预设范围内的养殖信息；
    获取到预设范围内的养殖信息后，获取预设范围内除养殖场以外的其他风险影响因素的信息，即其他影响因素信息；以及
    根据待检测区域内其他影响因素信息中各个其他影响因素的实际地理位置和预设范围的实际地理位置，确定预设范围内的其他影响因素信息。
  4. 如权利要求1所述的养殖场风险预测方法，其中，所述将所述养殖信息和所述其他影响因素信息输入预设的风险预测模型得到所述待预测养殖场的风险预测结果的步骤之后，还包括：
    根据所述待预测养殖场的风险预测结果从各个防控方案中匹配合适的防控方案，将匹配到的防控方案确定为待预测养殖场的目标防控方案，并可将目标防控方案输出。
  5. 如权利要求1所述的养殖场风险预测方法，其中，所述将待检测地图输入养殖场检测模型得到所述待检测地图中的养殖场信息的步骤包括：
    将待检测地图输入所述养殖场检测模型，得到所述待检测地图中各像素点的分类类别，其中，所述分类类别用于表示对应的像素点是否属于养殖场；以及根据所述分类类别确定所述待检测地图中的养殖场信息。
  6. 如权利要求5所述的养殖场风险预测方法，其中，所述养殖场检测模型是二分类模型，将所述待检测地图输入养殖场检测模型后，所述养殖场检测模型输出待检测地图中各个像素点的分类结果，所述分类结果用于表示对应的像素点是属于养殖场还是不属于养殖场。
  7. 如权利要求5所述的养殖场风险预测方法，其中，所述根据所述分类类别确定所述待检测地图中的养殖场信息的步骤包括：
    根据所述分类类别确定所述待检测地图中属于养殖场的目标像素点；
    采用预设聚类算法对所述目标像素点进行聚类得到各个聚类群；以及
    根据各所述聚类群确定所述待检测地图中的养殖场信息。
  8. 如权利要求1所述的养殖场风险预测方法，其中，所述将待检测地图输入养殖场检测模型得到所述待检测地图中的养殖场信息的步骤之前，还包括：
    采用正例训练数据对待训练模型进行初步训练，其中，所述正例训练数据包括预先采集到的多张包含养殖场的遥感卫星地图，以及各遥感卫星地图对应的养殖场标注数据；
    采用负例训练数据、或者采用所述负例训练数据和所述正例训练数据对初步训练后的待训练模型进行调整，其中，所述负例训练数据包括预先采集到的多张不包含养殖场的遥感卫星地图；以及
    当检测到调整后的待训练模型符合预设模型条件时，将调整后的待训练模型作为所述养殖场检测模型，否则基于调整后的待训练模型再执行所述步骤：采用正例训练数据对待训练模型进行初步训练。
  9. 如权利要求8所述的养殖场风险预测方法，其中，所述养殖场检测模型的训练方式为：
    预先采集多张包含养殖场的遥感卫星地图，以及采集遥感卫星地图中的养殖场的标注数据；以及将多张包含养殖场的遥感卫星地图以及养殖场数据作为正例训练数据。
  10. 如权利要求8所述的养殖场风险预测方法，其中，所述负例训练数据是正例训练数据的十分之一。
  11. 如权利要求8所述的养殖场风险预测方法，其中，所述预设模型条件是模型的损失函数收敛、准确率、召回率和IOU中的至少一个。
  12. 如权利要求8所述的养殖场风险预测方法，其中，所述采用正例训练数据对待训练模型进行初步训练的步骤包括：
    对所述正例训练数据中的各遥感卫星地图进行数据增广操作得到增广地图，其中，所述数据增广操作至少包括扭曲操作、翻转操作和加噪操作；以及采用所述正例训练数据和所述增广地图对待训练模型进行初步训练。
  13. 如权利要求1至12任一项所述的养殖场风险预测方法，其中，所述其他影响因素信息包括水路网信息，所述将所述养殖信息和所述其他影响因素信息输入预设的风险预测模型得到所述待预测养殖场的风险预测结果的步骤包括：
    根据所述水路网信息和所述预设范围确定所述预设范围的水路网密度特征值；根据所述养殖信息和所述预设范围确定所述预设范围的养殖场密度特征值；以及将所述水路网密度特征值和所述养殖场密度特征值输入所述风险预测模型得到所述待预测养殖场的风险预测结果。
  14. 如权利要求13所述的养殖场风险预测方法，其中，所述风险预测模型中包括与所述水路网密度特征值和所述养殖场密度特征值分别对应的预设权重值，所述将所述水路网密度特征值和所述养殖场密度特征值输入所述风险预测模型得到所述待预测养殖场的风险预测结果的步骤包括：
    将所述水路网密度特征值和所述养殖场密度特征值输入所述风险预测模型，以调用所述风险预测模型基于所述权重值、所述水路网密度特征值和所述养殖场密度特征值计算得到所述待预测养殖场的风险系数；以及
    将所述风险系数作为所述待预测养殖场的风险预测结果，或根据所述风险系数，以及系数与等级之间的预设对应关系确定风险等级后，将所述风险等级作为所述待预测养殖场的风险预测结果。
  15. 一种养殖场风险预测装置，其中，所述养殖场风险预测装置包括：
    输入模块，用于将待检测地图输入养殖场检测模型得到所述待检测地图中的养殖场信息；
    确定模块，用于根据所述养殖场信息确定待预测养殖场，以及确定所述待预测养殖场周围预设范围内的养殖信息，并获取所述预设范围内除养殖场外的其他影响因素信息；以及
    预测模块，用于将所述养殖信息和所述其他影响因素信息输入预设的风险预测模型得到所述待预测养殖场的风险预测结果。
  16. 如权利要求15所述的养殖场风险预测装置，其中，所述输入模块包括：
    第一输入单元，用于将待检测地图输入所述养殖场检测模型，得到所述待检测地图中各像素点的分类类别，其中，所述分类类别用于表示对应的像素点是否属于养殖场；
    第一确定单元，用于根据所述分类类别确定所述待检测地图中的养殖场信息。
  17. 一种养殖场风险预测设备，其中，所述养殖场风险预测设备包括：存储器、处理器及存储在所述存储器上并可在所述处理器上运行的养殖场风险预测程序，所述养殖场风险预测程序被所述处理器执行时实现如权利要求1所述的养殖场风险预测方法的步骤。
  18. 一种养殖场风险预测设备，其中，所述养殖场风险预测设备包括：存储器、处理器及存储在所述存储器上并可在所述处理器上运行的养殖场风险预测程序，所述养殖场风险预测程序被所述处理器执行时实现如权利要求5所述的养殖场风险预测方法的步骤。
  19. 一种计算机可读存储介质，其中，所述计算机可读存储介质上存储有养殖场风险预测程序，所述养殖场风险预测程序被处理器执行时实现如权利要求1所述的养殖场风险预测方法的步骤。
  20. 一种计算机可读存储介质，其中，所述计算机可读存储介质上存储有养殖场风险预测程序，所述养殖场风险预测程序被处理器执行时实现如权利要求1所述的养殖场风险预测方法的步骤。
PCT/CN2020/134032 2020-02-28 2020-12-04 养殖场风险预测方法、装置、设备及存储介质 WO2021169464A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010130767.1A CN111339237A (zh) 2020-02-28 2020-02-28 养殖场风险预测方法、装置、设备及存储介质
CN202010130767.1 2020-02-28

Publications (1)

Publication Number Publication Date
WO2021169464A1 true WO2021169464A1 (zh) 2021-09-02

Family

ID=71185750

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/134032 WO2021169464A1 (zh) 2020-02-28 2020-12-04 养殖场风险预测方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN111339237A (zh)
WO (1) WO2021169464A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114545765A (zh) * 2022-01-24 2022-05-27 南通世创机械有限公司 一种基于pid控制的变压器参数确定方法及系统
CN116313145A (zh) * 2023-05-10 2023-06-23 江西省农业科学院畜牧兽医研究所 一种智能化动物流行病学调查系统
CN116484226A (zh) * 2023-05-04 2023-07-25 乳山市海洋经济发展中心 一种通过神经网络建立渔业养殖模型的方法及装置
CN116957342A (zh) * 2023-09-18 2023-10-27 深圳市博安智控科技有限公司 基于物联网的养殖场智能管理方法、系统及介质

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111339237A (zh) * 2020-02-28 2020-06-26 深圳前海微众银行股份有限公司 养殖场风险预测方法、装置、设备及存储介质
CN117787736A (zh) * 2023-11-17 2024-03-29 新光瑞合(青岛)生物科技有限公司 一种健康养殖安全体系的构建方法及系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101763095A (zh) * 2009-12-09 2010-06-30 中国人民解放军防化指挥工程学院 一种移动化学风险源预测预警和应急控制方法
CN105205587A (zh) * 2015-08-15 2015-12-30 东北师范大学 一种北方草原火灾风险等级快速评估方法
CN109993432A (zh) * 2019-03-29 2019-07-09 柳州市水产畜牧兽医局 一种养猪场养殖质量安全风险评估方法
CN111339237A (zh) * 2020-02-28 2020-06-26 深圳前海微众银行股份有限公司 养殖场风险预测方法、装置、设备及存储介质

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103679558B (zh) * 2013-12-20 2017-01-04 国家电网公司 电动汽车充换电站火灾风险数据评估方法
CN107705047B (zh) * 2017-11-08 2021-07-13 泰康保险集团股份有限公司 一种保险核保风险预测方法及装置
CN110059758B (zh) * 2019-04-24 2020-07-10 海南长光卫星信息技术有限公司 一种基于语义分割的遥感影像养殖塘检测方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101763095A (zh) * 2009-12-09 2010-06-30 中国人民解放军防化指挥工程学院 一种移动化学风险源预测预警和应急控制方法
CN105205587A (zh) * 2015-08-15 2015-12-30 东北师范大学 一种北方草原火灾风险等级快速评估方法
CN109993432A (zh) * 2019-03-29 2019-07-09 柳州市水产畜牧兽医局 一种养猪场养殖质量安全风险评估方法
CN111339237A (zh) * 2020-02-28 2020-06-26 深圳前海微众银行股份有限公司 养殖场风险预测方法、装置、设备及存储介质

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
DU WEI, LI WEN-JING, FAN JIA-HUI, TIAN JIA-RONG, LIAO XIAO-WEN, GUO PENG, YAO SHI-HAO, LI MINGSHI, WANG WEN-LIN: "Extracting Large-scale Pig Farms in Plain River Network Area From GF-2 Image", JOURNAL OF ECOLOGY AND RURAL ENVIRONMENT, NANJING INSTITUTE OF ENVIRONMENTAL SCIENCES, MINISTRY OF ECOLOGY AND ENVIRONMENT, CN, vol. 36, no. 11, 1 November 2020 (2020-11-01), CN, pages 1485 - 1494, XP055841659, ISSN: 1673-4831, DOI: 10.19741/j.issn.1673-4831.2019.0764 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114545765A (zh) * 2022-01-24 2022-05-27 南通世创机械有限公司 一种基于pid控制的变压器参数确定方法及系统
CN116484226A (zh) * 2023-05-04 2023-07-25 乳山市海洋经济发展中心 一种通过神经网络建立渔业养殖模型的方法及装置
CN116484226B (zh) * 2023-05-04 2023-10-03 乳山市海洋经济发展中心 一种通过神经网络建立渔业养殖模型的方法及装置
CN116313145A (zh) * 2023-05-10 2023-06-23 江西省农业科学院畜牧兽医研究所 一种智能化动物流行病学调查系统
CN116957342A (zh) * 2023-09-18 2023-10-27 深圳市博安智控科技有限公司 基于物联网的养殖场智能管理方法、系统及介质
CN116957342B (zh) * 2023-09-18 2023-12-08 深圳市博安智控科技有限公司 基于物联网的养殖场智能管理方法、系统及介质

Also Published As

Publication number Publication date
CN111339237A (zh) 2020-06-26

Similar Documents

Publication Publication Date Title
WO2021169464A1 (zh) 养殖场风险预测方法、装置、设备及存储介质
US20200293892A1 (en) Model test methods and apparatuses
WO2021051609A1 (zh) 细颗粒物污染等级的预测方法、装置及计算机设备
WO2021218314A1 (zh) 基于位置定位的事件识别方法、装置、设备及存储介质
CN108830145A (zh) 一种基于深度神经网络的人数统计方法及存储介质
CN111401419A (zh) 基于改进RetinaNet的员工着装规范检测方法
CN112364439A (zh) 自动驾驶系统的仿真测试方法、装置以及存储介质
WO2021051917A1 (zh) 人工智能ai模型的评估方法、系统及设备
TW200849917A (en) Detecting method of network invasion
CN111339965A (zh) 养殖场选址方法、装置、设备及计算机可读存储介质
CN110020650B (zh) 一种基于深度学习识别模型的倾斜车牌的识别方法及装置
CN111523088A (zh) 一种基于dpsir模型的生态环境评价方法
CN110909804B (zh) 基站异常数据的检测方法、装置、服务器和存储介质
CN111126720A (zh) 养殖场风险预测方法、装置、设备及存储介质
CN116384844B (zh) 基于地理信息云平台的决策方法及装置
CN114359654A (zh) 一种基于位置关联性特征融合的YOLOv4混凝土表观病害检测方法
CN112529836A (zh) 高压线路缺陷检测方法、装置、存储介质及电子设备
CN111339950B (zh) 一种遥感图像目标检测方法
CN109934151B (zh) 一种基于movidius计算芯片和Yolo face的人脸检测方法
WO2020119533A1 (zh) 基于循环神经网络算法的舆情预警方法、装置、终端及介质
CN116721781A (zh) 虫媒传染病传播风险的预测方法、装置、电子设备及介质
CN114332570B (zh) 基于cdl深度学习进行作物迁移分类的方法和系统
CN110866051A (zh) 一种表征区域物流企业时空演变的系统
CN105469116A (zh) 一种基于人机交互的幼儿识图及数据扩充方法
KR101401801B1 (ko) 가축질병 예찰 시스템 및 방법

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20922305

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20922305

Country of ref document: EP

Kind code of ref document: A1