Method and system for automatically detecting small-size pests

Abstract

The invention discloses an automatic detection method and system for small-size pests, and relates to the technical field of automatic detection of pests, environmental condition data in a pest monitoring area are collected, and pest data are collected by a sensor network; identifying insect types and position information in the insect pest image by using the trained insect pest target detection model; the insect types are verified according to the insect pest characteristics of the small-size insects, the insect intensity in the area is calculated according to the position information of the target insect pests and the time nodes, when the insect intensity exceeds the expectation, the trained insect pest outbreak prediction model is used for predicting the number of the insect pests, and if the prediction result exceeds the expectation, an alarm value is constructed according to the alarm data. And when the warning value exceeds a warning threshold value, outputting an insect pest emergency treatment scheme by the small-size insect pest treatment knowledge graph. On the basis of mutual cooperation and verification of a plurality of different characteristics, the detection reliability is improved when small-size insects are detected and identified.

Description

Technical Field

[0001] The invention relates to the technical field of automatic pest detection, and in particular to a method and system for automatic detection of small-volume pests. Background Art

[0002] The tip-cutting beetle, a member of the family of the order Coleoptera, is a natural enemy of pine species, especially precious species such as Pinus armandii and Pinus yunnanensis. Although small in size, they are incredibly destructive, quietly weakening the vitality of trees by eating the tips and bast tissue of trees. The larvae of these insects build nests under the bark, while the adults constantly gnaw on the bark and treetops, cutting off the water and nutrient transport system inside the tree, causing the tree to gradually weaken, the leaves to turn yellow, and eventually the entire tree may die.

[0003] The harmfulness of the shoot beetle should not be underestimated. A large-scale outbreak of the beetle can quickly destroy an entire pine forest, causing huge losses of forest resources. In addition, since pine forests play an important role in the ecosystem, such as maintaining water and soil and regulating the climate, the rampant spread of the beetle may also have a profound impact on the ecological environment. In order to effectively control the harm of the shoot beetle, comprehensive prevention and control measures need to be taken, including forest management, biological control, chemical control, and enhanced quarantine.

[0004] In the Chinese invention patent with the authorization announcement number CN105243390B, an insect image detection method and an insect classification method are disclosed, which include the following steps: converting each frame image in the collected video data from color to grayscale, and performing a differential operation on the grayscale image and the corresponding modeling background before segmentation; calculating the eigenvalues ​​of each image block, normalizing the eigenvalues, and fusion processing; performing adaptive threshold segmentation on the obtained superpixel precision saliency map according to the maximum inter-class variance method to obtain a binary map; generating an image description of the image, and performing image matching on it with each image in the knowledge database, and judging whether the fruit fly to be classified belongs to a known species in the knowledge database according to the matching results. The present invention realizes the automatic identification of insect species through computer pattern recognition technology, combines visual saliency with phrases well, can more accurately reflect the salient areas in the query image, and improves the accuracy of recognition.

[0005] Combining the above content with the content in the prior art:

[0006] Considering that insect pest outbreaks can cause serious damage to trees and cash crops in the same area, resulting in a certain degree of economic losses and ecological imbalance, in order to reduce insect pest outbreaks and reduce the impact of insect pests, it is necessary to detect and identify harmful insects in the pest monitoring area in real time, especially some small insects, and to give early warning of pest outbreaks;

[0007] Existing automatic pest detection methods are mainly based on image recognition. After real-time collection of insect image data in the area that needs to be monitored, the type of insect is confirmed through image recognition and detection. However, when an insect outbreak occurs, there may be more than one type of pest that grows rapidly in number. In particular, when the environmental conditions in the pest detection area are highly consistent with the pest outbreak conditions and the possibility of an insect outbreak is high, in this scenario, since the acquired insect image data may be large, the risk of errors in the detection and identification of small insects is also high. On this basis, false alarms may also occur when warning of pest outbreaks, resulting in the inability to deal with pests in a timely and targeted manner when they occur.

[0008] To this end, the present invention provides a method and system for automatically detecting small-volume pests. Summary of the invention

[0009] 1. Technical issues to be resolved

[0010] In view of the deficiencies of the prior art, the present invention provides a method and system for automatically detecting small-volume pests, which collects pest data through a sensor network; uses a trained pest target detection model to identify the insect species and location information in the pest image; verifies the insect species based on the pest characteristics of small-volume insects, and calculates the insect density in the area based on the location information and time nodes of the target pests. When the insect density exceeds expectations, the trained pest outbreak prediction model is used to predict the number of pests. If the prediction result exceeds expectations, an alert value is constructed from the alarm data. When the alert value exceeds the alert threshold, the pest emergency treatment plan is output from the knowledge graph of small-volume pest management. On the basis of the mutual cooperation and verification of several different features, the reliability of detection is improved when detecting and identifying small-volume insects; thereby solving the technical problems raised in the background technology.

[0011] (II) Technical solution

[0012] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0013] The automatic detection method of small-volume pests includes collecting environmental condition data in the pest monitoring area, verifying the similarity between the environmental condition data and the target value, and if the similarity exceeds expectations, collecting pest data by the sensor network;

[0014] After targeted optimization of the collected low-quality pest images, the trained pest target detection model is used to identify the insect species and location information in the pest images;

[0015] The insect species are verified by the pest characteristics of small insects, and the insect density in the area is calculated based on the location information and time nodes of the target pests. When the insect density exceeds the expected level, an external pest warning instruction is issued;

[0016] Use the trained pest outbreak prediction model to predict the number of pests. If the prediction result exceeds expectations, a first-level alarm command will be issued to the outside world.

[0017] If the first-level alarm command does not obtain an effective response for multiple consecutive times, an alert value is constructed based on the alarm data. When the alert value exceeds the alert threshold, a second-level alarm command is issued, and a pest emergency treatment plan is output from the small-volume pest control knowledge graph.

[0018] Furthermore, data is collected in the selected pest monitoring area, and the environmental condition data is obtained and summarized to generate an environmental condition data set; the environmental condition data is analyzed for similarity with the target environmental condition data to obtain the corresponding condition similarity. If the condition similarity exceeds the expected value, a pest monitoring instruction is issued to the outside.

[0019] Further, after receiving the pest monitoring instruction, a sensor network and an insect capture device are arranged in the pest monitoring area;

[0020] The data collection frequency of the sensor network is constrained according to the conditional similarity. Based on the data collection frequency that meets the constraint conditions, the sensor network continuously collects pest images and related data in the pest monitoring area, and aggregates them to generate pest-related data sets.

[0021] Furthermore, the collected pest image quality data is used as input, and the trained image quality assessment model is used to evaluate the image quality of the pest image. If the pest image quality is lower than expected, feature extraction is performed on the low-quality image to obtain the corresponding optimization features. Based on the correspondence between the optimization features and the image optimization strategy, the corresponding image optimization strategy is given by the pre-built optimization strategy library.

[0022] Furthermore, the pest image data is used as input, and the trained pest target detection model is used to identify the insect species in the pest image, and the location information of the pest is obtained according to the imaging position.

[0023] Furthermore, different types of pest-related data are aligned according to timestamps and spatial locations;

[0024] After extracting the insect feature information in the pest image, the location of the insect is taken as the target location, the environmental features within the target location are extracted, and the sound features of the insect are extracted, which are summarized as a pest feature data set.

[0025] Furthermore, using pest features as input, the trained pest recognition model identifies the insect species and marks the pests according to the time nodes and location information of the target pests in the target area;

[0026] If the number of target pests that appear during the collection period exceeds expectations, the insect density in the area is calculated based on the location information and time nodes of the target pests.

[0027] Furthermore, after receiving the pest warning instruction, the sensor network collects environmental condition data and pest data in each sub-area within the pest detection area, performs correlation analysis on the two and obtains the correlation coefficient;

[0028] If the correlation coefficient exceeds the expected value, the trained pest outbreak prediction model is used to predict the number of pests in each sub-region and obtain the pest prediction value.

[0029] Further, an alarm threshold of the number of insects is set after constructing an insect pest value based on the number of insect pests in the sub-area;

[0030] When the number of first-level alarm instructions received within the preset alarm period exceeds expectations, an alarm value is generated based on the time node when the first-level alarm instructions are received and the degree to which the number of predicted pests exceeds the alarm threshold.

[0031] Furthermore, after receiving the second-level alarm command, the changing trend data of the number of insects and environmental data are summarized, a detection report is generated, and the distribution and diffusion path of the insects are displayed through a dynamic map.

[0032] Furthermore, taking pest control of small insects as the target word, a knowledge graph for small pest control is pre-constructed; taking insect quantity, species data, environmental data and tree-related data as input, the knowledge graph for small pest control provides corresponding emergency treatment plans for pest outbreaks in the pest monitoring area.

[0033] The small-volume automatic pest detection system includes an environmental analysis unit for collecting environmental condition data in the pest monitoring area, verifying the similarity between the environmental condition data and the target value, and if the similarity exceeds expectations, collecting pest data by the sensor network;

[0034] The image recognition unit optimizes the collected low-quality pest images in a targeted manner and uses the trained pest target detection model to identify the insect species and location information in the pest images;

[0035] The pest verification unit verifies the insect species based on the pest characteristics of small insects, calculates the insect density in the area based on the location information and time nodes of the target pests, and issues pest warning instructions to the outside when the insect density exceeds expectations;

[0036] The alarm unit uses the trained pest outbreak prediction model to predict the number of pests. If the prediction result exceeds expectations, a first-level alarm command is issued to the outside world.

[0037] The pest response unit, if the first-level alarm command does not obtain an effective response for multiple consecutive times, constructs an alert value based on the alarm data, issues a second-level alarm command when the alert value exceeds the alert threshold, and outputs a pest emergency treatment plan based on the small-volume pest control knowledge graph.

[0038] (III) Beneficial effects

[0039] The present invention provides a method and system for automatically detecting small-volume pests, which have the following beneficial effects:

[0040] 1. Constrain the data collection frequency of the sensor network based on conditional similarity. When collecting small-volume insects and related data, it can avoid excessive data collection and omissions in data collection.

[0041] 2. By matching low-quality images with corresponding image optimization strategies, the subsequent detection effect is guaranteed; through image recognition and image detection, the type and location of small-volume insects can be preliminarily determined, and preliminary automated detection can be completed on this basis, which can improve the efficiency of small-volume pest detection.

[0042] 3. Other pest characteristics obtained through sensor network detection can improve the reliability of small-volume insect detection when detecting and identifying small-volume insects based on the mutual coordination and verification of several different characteristics.

[0043] 4. Calculate the insect density in the area based on the location information and time nodes of the target pests. The insect density can be used to evaluate and judge the pest situation in the pest detection area. When the number of insects is large and there may be an outbreak of pests, timely treatment can be carried out.

[0044] 5. By predicting the pest situation, determining the pest data and location in each sub-area, it is possible to timely understand and deal with the pests before they occur, and take appropriate measures to delay or avoid the outbreak of pests; by setting dynamic thresholds and dynamic alarms, the insect pest alarms are more consistent with the actual environmental conditions and actual scenarios, and the accuracy of the early warning is more reasonable;

[0045] 6. The severity of the current pest can be judged based on the warning value. If the severity exceeds expectations, the corresponding treatment measures can be further collected. By building a multi-level alarm and early warning system, when pests occur, they can be treated and paid attention to in real time.

[0046] 7. Provide emergency treatment plans through the knowledge map of small-volume pest control. On the basis of completing automatic detection of small-volume insects and prediction of pest outbreaks, quickly provide reference plans so that pest outbreaks can be quickly dealt with to avoid further spread of pests and cause greater losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a schematic diagram of the process of the automatic detection method of small-volume pests of the present invention;

[0048] Figure 2 It is a schematic diagram of the structure of the automatic detection system for small-volume pests of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] See also Figure 1 The present invention provides a method for automatically detecting small-volume pests, comprising:

[0051] Step 1: Collect environmental condition data in the pest monitoring area, verify the similarity between the environmental condition data and the target value, and if the similarity exceeds expectations, collect pest data through the sensor network;

[0052] The step 1 includes the following contents:

[0053] Step 101: When it is necessary to detect pests, data is collected in the selected pest monitoring area, including abnormal climate data, tree damage or disease data, soil moisture data and temperature data, etc., and corresponding environmental condition data are obtained and summarized to generate an environmental condition data set;

[0054] Perform similarity analysis on the real-time environmental condition data and the target environmental condition data to obtain the corresponding condition similarity, and verify whether the current conditions for insect pest outbreaks are met based on the similarity. If the condition similarity exceeds the expected value, it means that under the current environmental conditions, insect pests may occur in the pest monitoring area, and a pest monitoring instruction is issued to the outside.

[0055] When using it, considering that the outbreak of insect pests is closely related to the changes in the actual environmental conditions, after determining the small-sized insects to be concerned, the environmental conditions that may cause the outbreak of pests are determined on the basis of obtaining the target pest type. Therefore, when starting the monitoring of small-sized pests, the environmental conditions are used as the starting conditions, and the detection is carried out when the starting conditions are met, which can improve the efficiency of insect detection;

[0056] Step 102: after receiving the pest monitoring instruction, a sensor network is arranged in the pest monitoring area, specifically including arranging high-resolution cameras, infrared sensors, temperature and humidity sensors, sound sensors, etc. at different heights and positions, and setting up insect catching devices, etc.;

[0057] The data collection frequency Cp of the sensor network is constrained according to the conditional similarity Xs in the following way:

[0058]

[0059] Where n is the number of sub-areas in the insect detection area, Xs ij is the difference between the similarity of the i-th condition and the similarity of the j-th condition, Xs a is the average value of the difference; weight coefficient: 0≤F1≤1, 0≤F2≤1, and F1+F2=1; the weight coefficient can be obtained by referring to the hierarchical analysis method;

[0060] According to the data collection frequency that meets the constraint conditions, the sensor network continuously collects pest images and related data in the pest monitoring area, and collects and aggregates the acquired pest data to generate pest-related data sets;

[0061] When using, combine the contents in steps 101 and 102:

[0062] When conditional similarity is used as an indicator of the possibility of insect outbreaks, when detecting small-volume insects, the data collection frequency of the sensor network is constrained based on conditional similarity. When collecting small-volume insects and related data, energy waste caused by excessive data collection can be avoided, and data collection omissions can also be avoided.

[0063] Existing automatic pest detection methods are mainly based on image recognition. After real-time collection of insect image data in the area that needs to be monitored, the type of insect is confirmed through image recognition and detection. However, when an insect outbreak occurs, there may be more than one type of pest that grows rapidly in number. In particular, when the environmental conditions in the pest detection area are highly consistent with the pest outbreak conditions and the possibility of an insect outbreak is high, in this scenario, since the acquired insect image data may be large, the risk of errors in the detection and identification of small insects is also high. On this basis, false alarms may also occur when warning of pest outbreaks, resulting in the inability to deal with pests in a timely and targeted manner when they occur.

[0064] Step 2: After optimizing the collected low-quality pest images, the trained pest target detection model is used to identify the insect species and location information in the pest images;

[0065] The step 2 includes the following contents:

[0066] Step 201: train a convolutional neural network with the labeled sample data to obtain a trained image quality assessment model, use the collected pest image quality data as input, and use the trained image quality assessment model to assess the image quality of the pest image. If the pest image quality is lower than expected, issue an image optimization instruction to the outside.

[0067] Step 202: after receiving the image optimization instruction, extract features of the low-quality image to obtain corresponding optimization features; collect and summarize several image optimization strategies through online retrieval or offline collection, and pre-build an optimization strategy library;

[0068] According to the correspondence between the optimization features and the image optimization strategy, the corresponding image optimization strategy is given by the pre-built optimization strategy library. After the image optimization strategy is executed, the low-quality pest images are optimized in a targeted manner.

[0069] When using it, after completing the collection of image data of small-volume insects, before using it for image and insect detection, considering that image quality will directly affect the effect of subsequent recognition and detection, therefore, by matching the corresponding image optimization strategy for low-quality images and performing targeted optimization, the subsequent detection effect can be guaranteed;

[0070] Step 203: Use the labeled insect image data to train the deep learning algorithm to obtain a trained pest target detection model; use the pest image data with image quality that meets the requirements as input, use the trained pest target detection model to identify the insect species in the pest image, and obtain the location information of the pest according to the imaging position;

[0071] When using, combine the contents in steps 201 to 203:

[0072] After optimizing low-quality images, the type and location of small insects can be preliminarily determined through image recognition and image detection. On this basis, preliminary automated detection can be completed to improve the efficiency of small insect pest detection.

[0073] Step 3: Verify the insect species based on the pest characteristics of small-volume insects, calculate the insect density in the area based on the location information and time nodes of the target pests, and issue pest warning instructions to the outside when the insect density exceeds expectations;

[0074] The step three includes the following contents:

[0075] Step 301: Convert different types of pest-related data, such as pest images, temperature and humidity, and sound, into a unified format and align them according to timestamps and spatial positions;

[0076] After extracting the characteristic information such as the insect species, quantity, and location information in the pest image, the location of the insect is taken as the target location; the environmental characteristics such as temperature, humidity, and light in the target location are extracted, and the sound characteristics such as the sound frequency and intensity of the insect are extracted through sound recognition. The above several characteristics are summarized as a pest characteristic data set;

[0077] When in use, on the basis of the preliminary detection of small-volume insect pests, other features of the insect pests obtained through the sensor network detection, on the basis of the mutual coordination and verification of several different features, when detecting and identifying small-volume insects, the reliability of the detection of small-volume insects can be improved;

[0078] Step 302: After training the neural convolution network with the labeled sample data, a trained pest recognition model is obtained;

[0079] Using pest features as input, the trained pest recognition model identifies the insect species and marks the pests according to the time nodes and location information of the target pests in the target area.

[0080] If the number of target pests exceeds expectations during the preset collection period, the insect density Jtp in the area is calculated based on the location information and time nodes of the target pests as follows:

[0081]

[0082] Where: Aρ(x, y, t) is the nonlinear insect density formula, w(x, y, t) is the weight function, and A is the area of ​​the pest detection area;

[0083] ρ(x,y,t)=a0+a1x+a2y+a3t+a4x 2 +a4x 2 +a5y 2 +a6t 2 +a7xy+a8xt+a9yt

[0084] In the formula, a0, a1...a9 are specific values ​​obtained by data fitting;

[0085]

[0086] Where: k1 and k2 are weight coefficients, with values ​​falling between [0,1], reflecting the influence of space and time on insect density;

[0087] σ is the spatial diffusion coefficient, which determines the spatial expansion range of the weight function and can be set to 3.5; (x i ,y i ) is the position of the insect, T is the length of the detection period;

[0088] Pre-set density thresholds based on historical data and the acceptable level of target pests;

[0089] On the basis of the automatic detection of small insects, if the insect density exceeds the density threshold, it means that the target pests in the pest monitoring area have already affected the growth status and growth safety of trees, and the pests need to be dealt with in time. At this time, a pest warning instruction is issued to the outside;

[0090] Combine the contents in steps 301 and 302:

[0091] When in use, the insect density in the area is calculated based on the location information and time nodes of the target pests. The insect density can be used to evaluate and judge the pest situation in the pest detection area. When the number of insects is large and there may be a pest outbreak, it can be dealt with in a timely manner.

[0092] Step 4: Use the trained pest outbreak prediction model to predict the number of pests. If the prediction result exceeds expectations, issue a first-level alarm command to the outside world.

[0093] The step 4 includes the following contents:

[0094] Step 401: After receiving the pest warning instruction, the sensor network collects environmental condition data and pest data in each sub-area of ​​the pest detection area, performs correlation analysis on the two and obtains a correlation coefficient;

[0095] If the obtained correlation coefficient exceeds the expected value, the trend data is obtained after analyzing the changing trends of insect, location and environmental data; the long short-term memory network is trained with the labeled sample data to obtain the pest outbreak prediction model, and the trained pest outbreak prediction model is used to predict the number of pests in each sub-area and obtain the pest prediction value;

[0096] When using it, to determine the impact of environmental condition data on pest outbreaks, the pest situation is predicted to determine the pest data and location of each sub-area, so that it can be understood and handled in time before the pest occurs, and appropriate treatment measures can be taken to delay or avoid pest outbreaks;

[0097] Step 402: Set an alarm threshold for the number of insects based on the number of insect pests in the sub-area and its historical data, in the following manner:

[0098] Perform linear normalization on the number of pests Et, map the corresponding data values ​​to the interval [0,1], and construct the pest value Hts in the following way:

[0099]

[0100] Where i = 1, 2, ..., n, n is the number of sub-regions, weight coefficient: 0 ≤ F1 ≤ 1, 0 ≤ F2 ≤ 1, and F1 + F2 = 1; the value of the weight coefficient is consistent with the previous value; Et i is the number of pests in the ith sub-area, Et a is the mean number of pests, Et b is the acceptable value of the number of pests; the method of constructing the alarm threshold [Ya, Yb] from the pest value Hts is as follows:

[0101]

[0102] Among them, Hts i is the pest value in the ith sub-period, Hts min 、Hts max 、Hts a are the minimum, maximum and mean values ​​of pest value respectively;

[0103] If the pest value in the pest monitoring area exceeds the alarm threshold, it means that the pests in the pest monitoring area may break out and need to be dealt with in time. A first-level alarm instruction will be issued to the outside world, and the predicted pest data will be sent to the relevant personnel through multiple channels such as SMS, email, and mobile phone applications.

[0104] When using, combine the contents in steps 401 and 402:

[0105] After obtaining the number of pests in each sub-area, the pest value is constructed from the pest number, and the current pest situation is comprehensively described by the pest value. According to the relationship between the pest value and the alarm threshold, when there are many pests in the pest detection area, they can be dealt with in time. At the same time, by setting dynamic thresholds and dynamic alarms, the insect pest alarm is more consistent with the actual environmental conditions and actual scenes, and the accuracy of the early warning is more reasonable.

[0106] Step 5: If the first-level alarm command fails to obtain an effective response for multiple consecutive times, an alarm value is constructed based on the alarm data. When the alarm value exceeds the alarm threshold, a second-level alarm command is issued, and an emergency treatment plan for pests is output from the small-volume pest control knowledge graph;

[0107] The step five includes the following contents:

[0108] Step 501: When the number of first-level alarm instructions received within the preset alarm period exceeds the expected number, under dimensionless conditions, the alarm value Wttp is generated in the following manner in combination with the time node of receiving the first-level alarm instruction and the degree to which the number of predicted pests exceeds the alarm threshold, and the severity of the pest outbreak is evaluated based on the alarm value:

[0109]

[0110] Where: α is the adjustment coefficient, which can be between [0,1], N(t) is the predicted number of pests at time t, and Y(t) is the corresponding expected value; m is the number of convolution operations, and λ is the time decay coefficient, which controls the influence of time on the risk value, so that the risk value closer to the current time has a higher weight; t o and t are the initial time and current time respectively;

[0111] Pre-set warning thresholds based on historical data and pest management expectations within the pest monitoring area;

[0112] On the basis of automatic detection of small-volume insect pests, if the obtained warning value exceeds the warning threshold, it means that the pest situation in the pest monitoring area is already serious. At this time, a secondary alarm command is issued to the outside;

[0113] When in use, if the current insect pests are not dealt with in time after receiving the first-level alarm command, the severity of the current pests can be judged by building an alert value. If the severity exceeds expectations, the corresponding treatment measures can be further collected;

[0114] Based on the deep automatic detection of small-volume pests, by building a multi-level alarm and early warning system, when pests occur, it can be ensured that they are dealt with and paid attention to in real time;

[0115] Step 502: After receiving the secondary alarm instruction, summarize the change trend data of the number of insects and environmental data, generate a detection report containing information such as the number of insects, types, and environmental conditions, and display the distribution and diffusion path of insects through a dynamic map;

[0116] Taking the pest control of small insects as the target word, after deep retrieval and entity relationship building, a knowledge graph of small insect pest control is pre-built; using the insect quantity, species data, environmental data and tree-related data as input, the knowledge graph of small insect pest control provides corresponding emergency treatment plans for insect pest outbreaks in the pest monitoring area;

[0117] When using, combine the contents in steps 501 and 502:

[0118] When it is necessary to deal with the current pest outbreak, an emergency treatment plan is given through the small-volume pest control knowledge map based on the obtained prediction data. This is used as a reference. After completing the automatic detection of small-volume insects and the prediction of pest outbreaks, a reference plan is quickly given to enable rapid handling when there is an insect pest outbreak, avoiding further spread of the pest and causing greater losses.

[0119] The construction method of the knowledge graph of small-volume pest control can refer to the following:

[0120] First, it is necessary to clarify the construction goal of the small-volume pest control knowledge graph, including the types of pests covered, control methods, related drugs and precautions for their use, etc. This will help maintain consistency in direction during subsequent data collection and processing.

[0121] Data sources: Collect various data related to small-volume pest control, including structured data (such as pest information and drug information in databases), semi-structured data (such as pest descriptions and control plans on web pages), and unstructured data (such as academic papers and research reports). These data can come from open data sets, literature databases, professional websites, and social media. Data content: Focus on collecting information on pest types, characteristics, hazards, life cycles, control methods, recommended drugs, and instructions for use.

[0122] Data deduplication: remove duplicate data entries to ensure data uniqueness. Data format conversion: convert data from different sources into a unified format for subsequent processing. Data unification: unify the naming and units of data to ensure data consistency and comparability. Natural language processing: use natural language processing technologies such as word segmentation, part-of-speech tagging, entity recognition, and relationship extraction to extract useful information from unstructured text.

[0123] Entity extraction: Automatically discover specific entities such as pest names, drug names, and control methods from the data. Relationship extraction: Using linguistics, statistics, and other methods, discover semantic relationships between entities from the text, such as the correspondence between pests and drugs, the association between pest characteristics and control methods, etc. Attribute extraction: For each entity, extract its related attribute information, such as the morphological characteristics of the pest, the dosage of the drug, etc.

[0124] Since data comes from multiple channels, there may be problems such as knowledge duplication and unclear relationships between knowledge. Therefore, knowledge fusion is needed to eliminate the ambiguity between referents such as entities, relationships, attributes and factual objects, so that knowledge from different sources can be standardized and integrated.

[0125] Model layer construction: Start with the top-level concepts to build the top-level ontology, and then refine the concepts and relationships to form a well-structured concept hierarchy tree. This helps to clearly express the complex relationships between entities such as pests, drugs, and control methods in the knowledge graph.

[0126] Data layer filling: Fill the entities, relationships, attributes and other data obtained from knowledge extraction into the model layer to form a complete knowledge graph.

[0127] Quality assessment: Conduct quality assessment on the constructed knowledge graph to check the accuracy, completeness and consistency of the data. This helps to identify potential problems and make corrections.

[0128] Iterative updates: As new knowledge continues to emerge and old knowledge becomes obsolete, the knowledge graph needs to be updated regularly. This includes adding new pest types, control methods, and drug information, as well as deleting outdated or erroneous information.

[0129] The Analytic Hierarchy Process (AHP) is a decision-making method that decomposes decision-related elements into levels such as goals, criteria, and plans, and conducts qualitative and quantitative analysis on this basis. It is particularly suitable for dealing with target systems with hierarchical and staggered evaluation indicators, and when the target value is difficult to describe quantitatively, the AHP is an effective decision-making tool.

[0130] The core of the hierarchical analysis method is to decompose the decision problem into multiple levels to form a hierarchical structure, which usually includes the target level, the criterion level, the sub-criterion level and the solution level. By solving the eigenvector of the judgment matrix, the priority weight of each element of each level to an element of the previous level is obtained, and the final weight of each alternative solution to the total goal is recursively merged by the weighted sum method, so as to find the optimal solution.

[0131] See also Figure 2 The present invention provides a small-volume pest automatic detection system, comprising:

[0132] The environmental analysis unit collects environmental condition data in the pest monitoring area and verifies the similarity between the environmental condition data and the target value. If the similarity exceeds expectations, the sensor network collects pest data;

[0133] The image recognition unit optimizes the collected low-quality pest images in a targeted manner and uses the trained pest target detection model to identify the insect species and location information in the pest images;

[0134] The pest verification unit verifies the insect species based on the pest characteristics of small insects, calculates the insect density in the area based on the location information and time nodes of the target pests, and issues pest warning instructions to the outside when the insect density exceeds expectations;

[0135] The alarm unit uses the trained pest outbreak prediction model to predict the number of pests. If the prediction result exceeds the expectation, a first-level alarm command is issued to the outside.

[0136] The pest response unit, if the first-level alarm command does not obtain an effective response for multiple consecutive times, constructs an alert value based on the alarm data, issues a second-level alarm command when the alert value exceeds the alert threshold, and outputs a pest emergency treatment plan based on the small-volume pest control knowledge graph.

[0137] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0138] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0139] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only some logical function divisions. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0140] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0141] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for automatically detecting small-volume pests, characterized in that: include, Collect environmental condition data in the pest monitoring area and verify the similarity between the environmental condition data and the target value. If the similarity exceeds expectations, the sensor network will collect pest data. After targeted optimization of the collected low-quality pest images, the trained pest target detection model is used to identify the insect species and location information in the pest images; The insect species are verified by the pest characteristics of small insects, and the insect density in the area is calculated based on the location information and time nodes of the target pests. When the insect density exceeds the expected level, an external pest warning instruction is issued; Use the trained pest outbreak prediction model to predict the number of pests. If the prediction result exceeds expectations, a first-level alarm command will be issued to the outside world. If the first-level alarm command does not obtain an effective response for multiple consecutive times, an alert value is constructed based on the alarm data. When the alert value exceeds the alert threshold, a second-level alarm command is issued, and a pest emergency treatment plan is output from the small-volume pest control knowledge graph.

2. The automatic detection method for small-volume pests according to claim 1, characterized in that: Collect data in the selected pest monitoring area, obtain environmental condition data and generate an environmental condition data set after aggregating them; Perform similarity analysis on the environmental condition data and the target environmental condition data to obtain the corresponding condition similarity. If the condition similarity exceeds the expected value, a pest monitoring instruction is issued to the outside.

3. The automatic detection method for small-volume pests according to claim 2, characterized in that: After receiving the pest monitoring instruction, a sensor network and an insect catching device are arranged in the pest monitoring area; The data collection frequency of the sensor network is constrained according to the conditional similarity. Based on the data collection frequency that meets the constraint conditions, the sensor network continuously collects pest images and related data in the pest monitoring area, and aggregates them to generate pest-related data sets.

4. The automatic detection method for small-volume pests according to claim 3, characterized in that: The collected pest image quality data is used as input, and the trained image quality assessment model is used to assess the image quality of the pest image. If the pest image quality is lower than expected, feature extraction is performed on the low-quality image to obtain the corresponding optimized features. According to the correspondence between the optimization features and the image optimization strategies, the corresponding image optimization strategies are given by the pre-built optimization strategy library.

5. The automatic detection method for small-volume pests according to claim 4, characterized in that: Taking pest image data as input, the trained pest target detection model is used to identify the insect species in the pest image, and the location information of the pest is obtained based on the imaging position.

6. The automatic detection method of small-volume pests according to claim 5, characterized in that: Align different types of pest-related data based on timestamps and spatial locations; After extracting the insect feature information in the pest image, the location of the insect is taken as the target location, the environmental features within the target location are extracted, and the sound features of the insect are extracted, which are summarized as a pest feature data set.

7. The automatic detection method of small-volume pests according to claim 6, characterized in that: Using pest features as input, the trained pest recognition model identifies the insect species and marks the pests according to the time nodes and location information of the target pests in the target area. If the number of target pests that appear during the collection period exceeds expectations, the insect density in the area is calculated based on the location information and time nodes of the target pests.

8. The method for automatically detecting small-volume pests according to claim 7, characterized in that: After receiving the pest warning instruction, the sensor network collects environmental condition data and pest data in each sub-area of ​​the pest detection area, performs correlation analysis on the two and obtains the correlation coefficient; If the correlation coefficient exceeds the expected value, the trained pest outbreak prediction model is used to predict the number of pests in each sub-region and obtain the pest prediction value.

9. The automatic detection method of small-volume pests according to claim 8, characterized in that: The alarm threshold of insect quantity is set after constructing the insect quantity value based on the insect quantity in the sub-area; When the number of first-level alarm instructions received within the preset alarm period exceeds expectations, an alarm value is generated based on the time node when the first-level alarm instructions are received and the degree to which the number of predicted pests exceeds the alarm threshold.

10. The automatic detection method of small-volume pests according to claim 9, characterized in that: After receiving the second-level alarm command, the changing trend data of insect quantity and environmental data are summarized, a detection report is generated, and the distribution and diffusion path of insects are displayed through a dynamic map.

11. The automatic detection method of small-volume pests according to claim 10, characterized in that: Taking pest control of small insects as the target word, a knowledge graph for small pest control is pre-constructed; taking insect quantity, species data, environmental data and tree-related data as input, the knowledge graph for small pest control provides corresponding emergency treatment plans for pest outbreaks in the pest monitoring area.

12. Small-volume pest automatic detection system, characterized by: include, The environmental analysis unit collects environmental condition data in the pest monitoring area and verifies the similarity between the environmental condition data and the target value. If the similarity exceeds expectations, the sensor network collects pest data; The image recognition unit optimizes the collected low-quality pest images in a targeted manner and uses the trained pest target detection model to identify the insect species and location information in the pest images; The pest verification unit verifies the insect species based on the pest characteristics of small insects, calculates the insect density in the area based on the location information and time nodes of the target pests, and issues pest warning instructions to the outside when the insect density exceeds expectations; The alarm unit uses the trained pest outbreak prediction model to predict the number of pests. If the prediction result exceeds the expectation, a first-level alarm command is issued to the outside. The pest response unit, if the first-level alarm command does not obtain an effective response for multiple consecutive times, constructs an alert value based on the alarm data, issues a second-level alarm command when the alert value exceeds the alert threshold, and outputs a pest emergency treatment plan based on the small-volume pest control knowledge graph.

Images