Intelligent deinsectization device and method based on machine learning

Abstract

The invention relates to the technical field of interpretable artificial intelligence, in particular to an intelligent deinsectization device and method based on machine learning, and the device comprises an infrared sensor, a high-definition camera, a chemical sensor, a sound sensor, a connecting rod, an electric shock bat, a pulley block, a chemical storage tank, a chemical nozzle, an MCU processor, a motor driver and a protective housing. The intelligent module is deployed in the MCU processor, the intelligent module obtains target pest data in multiple dimensions through a data obtaining unit, constructs a model through a pest image recognition unit and a pest movement recognition unit through feature extraction, plans and adjusts a pest catching path through a path planning unit, and flexibly selects a pest killing mode through an intelligent pest killing unit; according to the method, the pest recognition precision and the pest catching efficiency are remarkably improved, and the pest problem is efficiently solved according to the environment and the pest condition.

Description

Technical Field

[0001] The present invention relates to the technical field of interpretable artificial intelligence, and particularly to an intelligent insect extermination device and method based on machine learning. Background Art

[0002] Kitchens in urban residents' families are prone to attracting various pests due to food residues, water vapor, etc. For a long time, traditional insect extermination methods have dominated. As a common means, chemical agent spraying can, to a certain extent, inhibit the number of pests. However, frequent and large-scale use makes pests extremely resistant to chemical agents, resulting in a gradual decline in the insecticidal effect of the agents. Physical insect extermination methods have problems such as limited monitoring range and poor pest species recognition ability. Moreover, most existing insect extermination devices lack intelligent and precise functions. They are difficult to dynamically adjust insect extermination strategies according to the real-time behavior of pests, environmental changes, and the potential impact of pests on the ecosystem and human health. With the increasingly strict requirements of people for food safety and the quality of the living environment, it is of great significance to develop an efficient, intelligent, and precise insect extermination technology.

[0003] Chinese Patent Publication No.: CN117397657A discloses a new type of intelligent insect extermination device, including a controller, a light intensity sensor, and a trapping lamp. The light intensity sensor and the trapping lamp are both connected to the controller. When the light intensity sensor detects a decrease in the external light intensity, the trapping lamp is controlled to turn on through the controller. The controller is powered on, and the controller is connected to a waterproof switch. It also includes a base structure. The upper end of the base structure is provided with a lamp tube structure, and a sliding suction device is fitted outside the lamp tube structure. An air duct is provided between the sliding suction device and the base structure. However, this solution still has problems such as the lack of comprehensive and precise monitoring of pest species, quantity, biological characteristics, and behavior patterns, and a single insect extermination method. There is a need for a device that can accurately identify and predict pest species and movement patterns in complex environments and can flexibly select insect extermination methods according to actual needs to solve these problems. Summary of the Invention

[0004] To this end, the present invention provides an intelligent insect extermination device and method based on machine learning to overcome the problems in the prior art of low insect extermination efficiency and a single insect extermination method due to the lack of accurate identification of pest species and movement patterns in complex environments.

[0005] To achieve the above object, the present invention provides an intelligent insect extermination device and method based on machine learning, including:

[0006] An infrared sensor, which is connected to the protective housing and is used to collect the body temperature of pests;

[0007] A high-definition camera, which is connected to the protective housing and the MCU processor and is used to collect pest images and pest videos;

[0008] A chemical sensor, which is connected to a protective housing and is used for collecting pest pheromones;

[0009] A sound sensor, which is connected to a protective housing and is used for collecting pest sound frequencies;

[0010] A connecting rod, which is connected to an electric shock racket, an electric shock driver and a protective housing and is used for delivering electric power to the electric shock racket;

[0011] An electric shock racket, which is connected to the connecting rod and is used for hitting pests;

[0012] A pulley block, which includes a first pulley, a second pulley, a third pulley and a fourth pulley. The pulley block is connected to the protective housing and is used for driving the device to move;

[0013] A chemical drug storage tank, which is connected to a drug spray head and a protective housing and is used for storing chemical drugs;

[0014] A drug spray head, which is connected to the chemical drug storage tank and is used for spraying the chemical drugs in the chemical drug storage tank;

[0015] An MCU processor, which is connected to a high-definition camera and a protective housing and contains an intelligent module inside and is used for processing target pest-catching data;

[0016] A motor driver, which is connected to the connecting rod and the protective housing and is used for controlling the connecting rod;

[0017] A protective housing, which is connected to an infrared sensor, a high-definition camera, a chemical sensor, a sound sensor, a connecting rod, a first pulley, a second pulley, a third pulley, a fourth pulley, a chemical drug storage tank, an MCU processor and an electric shock driver, and is used for connecting device components, protecting the device and dissipating heat.

[0018] Furthermore, the intelligent module includes:

[0019] A data acquisition unit, which is used for acquiring target pest data;

[0020] A pest image recognition unit, which is used for extracting the target comprehensive features of the target pest data, and is also used for constructing a pest recognition model according to a pest recognition model construction method and outputting a pest recognition result according to the target comprehensive features of the target pest data;

[0021] A pest movement recognition unit, which is used for constructing a pest movement prediction model according to a pest movement prediction model construction method and outputting a pest movement prediction result according to the target pest data;

[0022] A path planning unit, which is used for planning a target pest-catching path according to the pest movement prediction result and making real-time adjustments to the target pest-catching path;

[0023] An intelligent pest control unit, which is used to set the pest control priority according to the target pest data, select the pest control method according to the pest identification result, and is also used to optimize the pest identification model according to the pest killing rate.

[0024] Furthermore, the data acquisition unit acquires the target pest data, and the target pest data includes pest image data, pest biological data, and pest impact data. The pest impact data includes the pest reproduction speed, the number of allergens produced by the pests, and the pest disease transmission ability value.

[0025] Furthermore, the target comprehensive features of the target pest data include pest image features, pest sound frequency features, pest body temperature features, and pest pheromone features. The pest image recognition unit extracts the pest image features in the target comprehensive features of the target pest data through a pest image feature extraction method. The pest image feature extraction method includes:

[0026] Step G01: Perform Gaussian filtering on the pest image data in the target pest data to obtain a target smoothed image;

[0027] Step G02: Calculate the first-order partial derivative of the target smoothed image to obtain the target gradient information;

[0028] Step G03: Perform non-maximum suppression on the target gradient information to obtain the limit pixel points;

[0029] Step G04: Perform double-threshold processing on the limit pixel points to obtain the pest image features;

[0030] The pest image recognition unit extracts the pest sound frequency features in the target comprehensive features of the target pest data through a pest sound frequency feature extraction method. The pest sound frequency feature extraction method includes: inputting the pest sound frequency in the pest biological data of the target pest data into a filter to obtain the target sound frequency, and performing discrete cosine transform on the target sound frequency to obtain the pest sound frequency features;

[0031] The pest image recognition unit extracts the pest body temperature features in the target comprehensive features of the target pest data through a pest body temperature feature extraction method. The pest body temperature feature extraction method includes: performing infrared thermal imaging processing on the pest body temperature in the pest biological data of the target pest data to obtain a pest thermal image, and then calculating the average temperature of the pest thermal image to obtain the pest body temperature features;

[0032] The pest image recognition unit extracts the pheromone feature of the target integrated feature in the target pest data through the pest pheromone feature extraction method. The pest pheromone feature extraction method includes: performing comprehensive chemical sensor processing on the pest pheromone in the pest biological data of the target pest data to obtain the pest pheromone feature;

[0033] The pest image recognition unit uses the pest image feature, pest sound frequency feature, pest body temperature feature, and pest pheromone feature as the target integrated feature of the target pest data.

[0034] Further, when the pest image recognition unit constructs a pest recognition model based on the pest feature database, the process includes:

[0035] Step K01, dividing 70% of the data in the pest feature database into an identification training set, 20% of the data in the pest feature database into an identification validation set, and 10% of the data in the pest feature database into an identification test set;

[0036] Step K02, selecting a convolutional neural network as the initial pest recognition model and initializing the weights and biases of the initial pest recognition model;

[0037] Step K03, using a convolutional kernel with preset parameters in the convolutional layer of the initial pest recognition model parameters, selecting the Adam optimizer and the cross-entropy loss function to train the initial pest recognition model, loading the identification training set into the initial pest recognition model, performing forward propagation through the initial pest recognition model, and calculating the output value of the initial pest recognition model;

[0038] Step K04, calculating the loss function value according to the output value of the initial pest recognition model, calculating the gradient through the backpropagation algorithm, and updating the weights and biases of the initial pest recognition model, repeating the processes of forward propagation, loss function calculation, and backpropagation to obtain the calculated pest recognition model;

[0039] Step K05, inputting the identification validation set into the calculated pest recognition model to optimize the parameters of the calculated pest recognition model to obtain the optimized pest recognition model;

[0040] Step K06, performing a correct rate test on the optimized pest recognition model through the identification test set, and outputting the optimized pest recognition model with a correct rate meeting 90% as the pest recognition model;

[0041] Step K07, inputting the target integrated feature of the target pest data into the pest recognition model to obtain the pest recognition result.

[0042] Further, when the path planning unit plans the target insect-catching path, it sets the current location of the device as the starting point, obtains the location where the pest arrives after a preset time according to the pest movement prediction result, takes it as the target insect-catching point, and according to the A* search algorithm, obtains all paths from the starting point to the target insect-catching point, takes them as the set of possible paths, and then obtains the optimal insect-killing path from the set of possible paths according to the insect-killing cost function. The pest movement prediction result is obtained in real time for the device according to the preset frequency, and the predicted pest speed B in the pest movement prediction result is obtained. The predicted pest speed B is compared with each preset pest biological speed. The preset pest biological speeds include the first preset pest biological speed B1 and the second preset pest biological speed B2. It is set that B1 = 5m / s and B2 = 10m / s. The change situation of the pest movement speed is judged according to the comparison result, and the optimal insect-killing path of the device is adjusted according to the judgment result, where:

[0043] When B > B2, the path planning unit determines that the change situation of the pest movement speed is abnormal, adjusts the optimal insect-killing path of the device, and obtains the location where the pest arrives after a preset time according to the pest movement prediction result again, takes it as the new target insect-catching point, and obtains the optimal insect-killing path according to the A* search algorithm;

[0044] When B1 ≤ B ≤ B2, the path planning unit determines that the change situation of the pest movement speed is normal and does not adjust the optimal insect-killing path of the device;

[0045] When B < B1, the path planning unit determines that the change situation of the pest movement speed is abnormal, adjusts the optimal insect-killing path of the device, and obtains the location where the pest arrives after a preset time according to the pest movement prediction result again, takes it as the new target insect-catching point, and obtains the optimal insect-killing path according to the A* search algorithm.

[0046] Further, the path planning unit calculates the pest movement speed change value C according to the formula C = B - (B2 - B1), compares the pest movement speed change value C with the preset speed change value C0, judges the change amplitude of the pest movement speed according to the comparison result, and adjusts the preset frequency according to the judgment result, where:

[0047] When C ≤ C0, the path planning unit determines that the change amplitude of the pest movement speed is abnormal and does not update the preset frequency;

[0048] When C > C0, the path planning unit determines that the change amplitude of the pest movement speed is abnormal, updates the preset frequency, and shortens the amplitude of the preset frequency;

[0049] The path planning unit obtains the pest movement direction D and the real-time pest movement direction D0 in the pest movement prediction result, calculates the deviation E of the pest movement direction according to the formula E = |D - D0|, compares the deviation E of the pest movement direction with the preset deviation E0, judges the pest movement direction status according to the comparison result, and adjusts the pest movement speed according to the judgment result, where:

[0050] When E ≤ E0, the path planning unit determines that the pest movement direction status is normal and does not adjust the pest movement speed;

[0051] When E > E0, the path planning unit determines that the pest movement direction status is abnormal, adjusts the pest movement speed, and directly determines that the change in the pest movement speed is abnormal.

[0052] Further, when setting the pest control priority, the intelligent pest control unit obtains the pest reproduction speed F, compares the pest reproduction speed F with each preset reproduction speed, and the preset reproduction speeds include the first preset reproduction speed F1 and the second preset reproduction speed F2. It is set that F1 = 42 pests per day and F2 = 0.56 pests per day. The pest reproduction speed level is judged according to the comparison result, and the pest control priority is set according to the judgment result, where:

[0053] When F < F1, the intelligent pest control unit determines that the pest reproduction speed level is low and sets it as the third priority for pest control;

[0054] When F1 ≤ F ≤ F2, the intelligent pest control unit determines that the pest reproduction speed level is medium and sets it as the second priority for pest control;

[0055] When F > F2, the intelligent pest control unit determines that the pest reproduction speed level is high and sets it as the first priority for pest control.

[0056] Further, the intelligent pest control unit compares the pest movement speed B with the preset pest movement speed B0, judges the influence degree of the pest movement speed according to the comparison result, and updates the pest reproduction speed level according to the judgment result, where:

[0057] When B ≤ B0, the intelligent pest control unit determines that the influence degree of the pest movement speed is low and does not update the pest reproduction speed level;

[0058] When B > B0, the intelligent pest control unit determines that the influence degree of the pest movement speed is high, updates the pest reproduction speed level, sets the update coefficient as α, α = 1.3 - 0.3e -0.7*(B-B0) , and the updated pest reproduction speed is F1, F1 = α × F;

[0059] The intelligent pest control unit obtains the quantity H of allergens produced by pests, the pest disease transmission ability value R, and the pest disease transmission ability value J. It calculates the threat value K of pests to human health according to the formula K = 0.3×H + 0.5×R + 0.2J, compares the threat value K of pests to human health with the preset threat value K0, judges the threat level of pests to human health according to the comparison result, and corrects the movement speed of pests according to the judgment result, where:

[0060] When K ≤ K0, the intelligent pest control unit determines that the threat level of pests to human health is low and does not correct the movement speed of pests;

[0061] When K > K0, the intelligent pest control unit determines that the threat level of pests to human health is high and corrects the movement speed of pests. Set the updated threat coefficient as β, β = 1.48 - 0.2e -0.1*(K-K0) , and the corrected movement speed of pests is B3, B3 = β×B;

[0062] Furthermore, the intelligent pest control unit selects a pest control method according to the pest identification result and initializes the selected pest control method as the electric shock method;

[0063] The intelligent pest control unit obtains the pest density ρ and compares it with the preset pest density ρ0, judges the pest quantity status according to the comparison result, and adjusts the pest control method according to the judgment result, where:

[0064] When ρ ≤ ρ0, the intelligent pest control unit determines that the pest quantity status is normal and does not adjust the pest control method;

[0065] When ρ > ρ0, the intelligent pest control unit determines that the pest quantity status is abnormal and adjusts the pest control method, adjusting the electric shock method to the spray pest control method;

[0066] After the pest control robot reaches the preset working duration, the intelligent pest control unit obtains the current pest quantity ρ1 and the pest quantity ρ2 one week ago, and according to the formula calculates the pest killing rate M, compares the pest killing rate M with the preset pest killing rate M0, judges the pest killing situation according to the comparison result, and replaces the chemical drug in the chemical drug storage tank according to the judgment result, where:

[0067] When M > M0, the intelligent pest control unit determines that the pest killing situation is normal and does not replace the chemical drug in the chemical drug storage tank;

[0068] When M ≤ M0, the intelligent pest control unit determines that the pest killing situation is abnormal, replaces the chemical drug in the chemical drug storage tank, recalculates the pest killing rate after replacing the chemical drug in the chemical drug storage tank, and if M ≤ M0, optimizes the pest recognition model. The process includes:

[0069] The intelligent pest control unit obtains the accuracy rate A1 of the pest recognition model and compares it with the preset model accuracy rate A0, judges the accuracy rate situation of the pest recognition model according to the comparison result, and optimizes the pest recognition model according to the judgment result, where:

[0070] When A1 < A0, the intelligent pest control unit determines that the accuracy rate situation of the pest recognition model is abnormal, optimizes the pest recognition model, independently constructs a pest recognition model with the target pest data collected by the sound sensor, infrared sensor and chemical sensor, sets the weight of the target pest data collected by the sound sensor as ω1, the weight of the target pest data collected by the infrared sensor as ω2, and the weight of the target pest data collected by the chemical sensor as ω3, where ω1 + ω2 + ω3 = 1, distributes the weights in multiple batches according to the preset ratio, determines the final pest recognition result until the intelligent pest control unit determines that the accuracy rate situation of the pest recognition model is normal;

[0071] When A1 ≥ A0, the intelligent pest control unit determines that the accuracy rate situation of the pest recognition model is normal and adjusts the electric shock method to the spray pest control method.

[0072] Compared with the prior art, the beneficial effects of the present invention are as follows. Through multi-sensor fusion, comprehensive and accurate monitoring and recognition of pests are realized, a pest recognition model and a pest movement prediction model are constructed, so as to accurately plan the pest catching path, set the pest control priority and select the pest control method according to the types and movement patterns of pests, accurately identify the types and movement patterns of pests in complex environments, flexibly select a variety of pest control methods, and improve the pest control efficiency. Description of the Drawings

[0073] Figure 1 It is a schematic structural diagram of the intelligent pest control device based on machine learning in this embodiment;

[0074] Figure 2 It is a schematic structural diagram of the intelligent module in this embodiment. Detailed Embodiments

[0075] In order to make the purpose and advantages of the present invention clearer, the present invention will be further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0076] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principle of the present invention and do not limit the protection scope of the present invention.

[0077] It should be noted that in the description of the present invention, the terms indicating directions or positional relationships such as "upper", "lower", "left", "right", "inner", "outer", etc. are based on the directions or positional relationships shown in the drawings. This is only for convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0078] In addition, it should also be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installation", "connection", and "connection" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0079] Please refer to Figure 1 as shown, which is a schematic structural diagram of the intelligent pest control device based on machine learning in this embodiment. The device includes:

[0080] An infrared sensor 1, which is connected to the protection shell 15 and is used to collect the body temperature of pests;

[0081] A high-definition camera 2, which is connected to the protection shell 15 and the MCU processor 13 and is used to collect pest images and pest videos;

[0082] A chemical sensor 3, which is connected to the protection shell 15 and is used to collect pest pheromones;

[0083] A sound sensor 4, which is connected to the protection shell 15 and is used to collect the sound frequency of pests;

[0084] A connecting rod 5, which is connected to the electric shock bat 6, the electric shock driver 14 and the protection shell 15 and is used to deliver electric power to the electric shock bat;

[0085] An electric shock bat 6, which is connected to the connecting rod 5 and is used to strike pests;

[0086] A pulley group, which includes a first pulley 7, a second pulley 8, a third pulley 9 and a fourth pulley 10. The pulley group is connected to the protection shell 15 and is used to drive the device to move;

[0087] A chemical drug storage tank 11, which is connected to the drug sprayer 12 and the protection shell 15 and is used to store chemical drugs;

[0088] The drug spray head 12, which is connected to the chemical drug storage tank 11 and is used to spray the chemical drug in the chemical drug storage tank;

[0089] The MCU processor 13, which is connected to the high-definition camera 2 and the protective housing 15 and is used to process the target insect-catching data;

[0090] The motor driver 14, which is connected to the connecting rod 5 and the protective housing 15 and is used to control the connecting rod 5;

[0091] The protective housing 15, which is connected to the infrared sensor 1, the high-definition camera 2, the chemical sensor 3, the sound sensor 4, the connecting rod 5, the first pulley 7, the second pulley 8, the third pulley 9, the fourth pulley 10, the chemical drug storage tank 11, the MCU processor 13 and the electric shock driver 14, is used to connect the device components, protect the device and dissipate heat.

[0092] Specifically, the device is applied to pest killing. By collecting pest image data and pest biological data through the high-definition camera and multiple sensors, the monitoring accuracy and comprehensiveness of pests are improved, the traces and positions of pests can be accurately found. The pest position is determined by the MCU processor, and the electric shock driver controls the connecting rod to deliver electricity to the electric shock paddle. The electric shock paddle can hit the pests in time. The chemical drug in the chemical drug storage tank can be sprayed through the drug spray head to eliminate the pests. And the appropriate pest control method can be selected according to the environmental situation, which can efficiently handle the pest problem, reduce the harm of pests to the environment and improve the insect-catching efficiency.

[0093] Specifically, the infrared sensor 1, the chemical sensor 3 and the sound sensor 4 collect pest biological data, and the infrared thermal imaging is used to collect the body temperature of pests. The high-definition camera 2 collects pest image data. The connecting rod 5 uses a conductive material to deliver electricity to the electric shock paddle. The electric shock paddle 6 is composed of a steel wire mesh and hits the pests. The first pulley 7, the second pulley 8, the third pulley 9 and the fourth pulley 10 use rubber pulleys to drive the device to move. The chemical drug storage tank 11 uses stainless steel as the tank material to store the chemical drug. The drug spray head 12 uses a pressure-type spray head to spray the chemical drug in the chemical drug storage tank. The MCU processor 13, as the central processor of the device, processes the pest biological data and pest image data and controls the motor driver 14. The motor driver 14 controls the connecting rod 5. The protective housing 15 uses a hard heat-dissipating material to connect and protect the device and dissipate heat.

[0094] Specifically, the pest biological data refers to pest sound frequency, pest body temperature, and pest pheromone. The pest sound frequency is collected by a sound sensor, the pest body temperature characteristics are collected by an infrared sensor, and the pest pheromone is collected by a chemical sensor. The pest image data refers to pest images and pest videos, which are collected by a high-definition camera. The infrared thermography is a technology that uses the infrared radiation emitted by the pest itself to image the surface body temperature distribution of the pest. In this embodiment, the conductive material is not limited, and those skilled in the relevant art can freely select according to actual needs as long as the conductivity requirement is met, such as copper and aluminum. The steel wire mesh is a hard mesh woven from steel materials. In this embodiment, the pressure type nozzle is not limited, such as a centrifugal nozzle and a fan-shaped nozzle. In this embodiment, the hard heat dissipation material is not limited, such as copper and aluminum.

[0095] Please refer to Figure 2 as shown, which is a schematic structural diagram of the intelligent module of this embodiment. The intelligent module includes:

[0096] A data acquisition unit for acquiring target pest data;

[0097] A pest image recognition unit for extracting the target comprehensive features of the target pest data, and also for constructing a pest recognition model according to the pest recognition model construction method, and outputting a pest recognition result according to the target comprehensive features of the target pest data. The pest image recognition unit is connected to the data acquisition unit;

[0098] A pest movement recognition unit for constructing a pest movement prediction model according to the pest movement prediction model construction method, and outputting a pest movement prediction result according to the target pest data. The pest movement recognition unit is connected to the data acquisition unit;

[0099] A path planning unit for planning a target pest-catching path according to the pest movement prediction result and making real-time adjustments to the target pest-catching path. The path planning unit is connected to the pest movement recognition unit;

[0100] An intelligent pest control unit for setting pest control priorities according to the target pest data, selecting pest control methods according to the pest recognition result, and also for optimizing the pest recognition model according to the pest killing rate. The intelligent pest control unit is connected to the pest image recognition unit.

[0101] Specifically, the intelligent module is applied to the MCU processor of the intelligent pest control device described in this embodiment. It accurately identifies the types and movement patterns of pests through multi-sensor fusion. At the same time, it flexibly plans the pest-catching path, sets the pest control priority, and selects the pest control method according to the pest characteristics, improving the pest control efficiency. The data acquisition unit obtains target pest data from multiple dimensions, enhancing the pest control accuracy rate. The pest image recognition unit extracts the target comprehensive features of the target pest data, effectively improving the pest recognition accuracy rate. At the same time, it fuses pest image features, pest sound frequency features, pest body temperature features, and pest pheromone features, broadening the recognition dimension, accurately distinguishing pests in complex environments, and enhancing the adaptability of the device. It also outputs the pest recognition result by constructing a pest recognition model, thus adapting to diverse pest species and improving the pest recognition accuracy. The pest movement recognition unit accurately predicts the future movement trend of pests by constructing a pest movement prediction model, providing a reliable analysis basis for the path planning unit and enhancing the pest-catching efficiency. The path planning unit takes the current position of the device as the starting point, determines the target pest-catching point based on the pest movement prediction result, and selects the best pest control path from numerous possible paths by means of the A* search algorithm, thus ensuring that the device moves efficiently to the target position. At the same time, it continuously monitors the changes in the pest movement speed and the deviation of the pest movement direction and compares them with the preset values to timely judge the changes in the pest movement state, thereby re-planning the best pest control path, making the device always fit the pest movement changes, and enhancing the accuracy of the pest-catching path and the pest control efficiency. The intelligent pest control unit scientifically sets the pest control priority according to the pest reproduction speed, pest movement speed, and the threat value to human health, ensuring that the pests with the greatest harm are processed first, rationally allocating pest control resources, and flexibly switching the pest control method according to the pest recognition result and pest density, improving the pest control effectiveness and the adaptability of the device. At the same time, it judges the pest control effect by calculating the pest killing rate, timely replaces chemical drugs, and optimizes the pest recognition model to continuously improve the pest control effect.

[0102] Specifically, the data acquisition unit obtains the target pest data, and the target pest data includes pest image data, pest biological data, and pest impact data. The pest impact data includes pest reproduction speed, the number of allergens produced by pests, and the pest disease transmission ability value.

[0103] It can be understood that this embodiment does not limit the acquisition method of the pest impact data. Relevant technical personnel in the field can freely choose according to actual needs as long as the requirement of obtaining the pest impact data is met, such as obtaining it from the Internet.

[0104] Specifically, the data acquisition unit obtains the target pest data from multiple dimensions, facilitating subsequent precise pest control based on the target pest data and improving the pest control efficiency and accuracy rate.

[0105] Specifically, the target comprehensive features of the target pest data include pest image features, pest sound frequency features, pest body temperature features, and pest pheromone features. The pest image recognition unit extracts the pest image features in the target comprehensive features of the target pest data through a pest image feature extraction method, and the pest image feature extraction method includes:

[0106] Step G01, perform Gaussian filtering on the pest image data in the target pest data to obtain a target smoothed image;

[0107] Step G02, perform a first-order partial derivative calculation on the target smoothed image to obtain target gradient information;

[0108] Step G03, perform non-maximum suppression on the target gradient information to obtain limit pixel points;

[0109] Step G04, perform a double-threshold processing on the limit pixel points to obtain pest image features;

[0110] The pest image recognition unit extracts the pest sound frequency features in the target comprehensive features of the target pest data through a pest sound frequency feature extraction method, and the pest sound frequency feature extraction method includes: inputting the pest sound frequency in the pest biological data of the target pest data into a filter to obtain a target sound frequency, and performing a discrete cosine transform on the target sound frequency to obtain pest sound frequency features;

[0111] The pest image recognition unit extracts the pest body temperature features in the target comprehensive features of the target pest data through a pest body temperature feature extraction method, and the pest body temperature feature extraction method includes: performing infrared thermal imaging processing on the pest body temperature in the pest biological data of the target pest data to obtain a pest thermal image, and then calculating the average temperature of the pest thermal image to obtain pest body temperature features;

[0112] The pest image recognition unit extracts the pest pheromone features in the target comprehensive features of the target pest data through a pest pheromone feature extraction method, and the pest pheromone feature extraction method includes: performing comprehensive chemical sensor processing on the pest pheromone in the pest biological data of the target pest data to obtain pest pheromone features;

[0113] The pest image recognition unit uses the pest image features, pest sound frequency features, pest body temperature features, and pest pheromone features as the target comprehensive features of the target pest data;

[0114] When the pest image recognition unit constructs a pest recognition model based on the pest feature database, the process includes:

[0115] Step K01: Divide 70% of the data in the pest feature database into an identification training set, 20% of the data in the pest feature database into an identification validation set, and 10% of the data in the pest feature database into an identification test set;

[0116] Step K02: Select a convolutional neural network as the initial pest identification model, and initialize the weights and biases of the initial pest identification model;

[0117] Step K03: Use a convolutional kernel with preset parameters in the convolutional layer of the initial pest identification model parameters, select the Adam optimizer and the cross-entropy loss function to train the initial pest identification model, load the identification training set into the initial pest identification model, perform forward propagation through the initial pest identification model, and calculate the output value of the initial pest identification model;

[0118] Step K04: Calculate the loss function value based on the output value of the initial pest identification model, calculate the gradient through the backpropagation algorithm, and update the weights and biases of the initial pest identification model. Repeat the processes of forward propagation, loss function calculation, and backpropagation to obtain the calculated pest identification model;

[0119] Step K05: Input the identification validation set into the calculated pest identification model to optimize the parameters of the calculated pest identification model and obtain the optimized pest identification model;

[0120] Step K06: Conduct a correct rate test on the optimized pest identification model through the identification test set, and output the optimized pest identification model with a correct rate meeting 90% as the pest identification model;

[0121] Step K07: Input the target comprehensive features of the target pest data into the pest identification model to obtain the pest identification result.

[0122] Specifically, the recognition training set refers to the data set in the pest feature database used to train the initial pest recognition model. The recognition validation set refers to the data set in the pest feature database used to adjust the calculated pest recognition model. The recognition test set refers to the data set in the pest feature database used to evaluate the optimized pest recognition model. The Gaussian filtering refers to a digital signal processing technology that removes noise in an image and makes the image edges softer. The first-order partial derivative calculation refers to the calculation of the horizontal gradient and vertical gradient of an image to obtain the gradient magnitude and direction angle. For example, if the coordinates of the i-th point in the target smoothed image in the plane coordinate system are I(x, y), the horizontal gradient L(x) of the target smoothed image is calculated according to the formula L(x) = I(x + 1, y) - I(x, y), where I(x + 1, y) refers to the point on the same horizontal line as the i-th point in the target smoothed image, but translated one unit length in the positive X-axis direction of the plane coordinate system compared to the i-th point in the target smoothed image. The vertical gradient L(y) of the target smoothed image is calculated according to the formula L(y) = I(x, y + 1) - I(x, y), where I(x, y + 1) refers to the point on the same vertical line as the i-th point in the target smoothed image, but translated one unit length in the positive Y-axis direction of the plane coordinate system compared to the i-th point in the target smoothed image. Then, according to the gradient magnitude S(x, y) of the target smoothed image is calculated. Finally, according to the direction angle θ(x, y) of the target smoothed image is calculated. The target gradient information includes the gradient magnitude and direction angle of pest images and pest videos. The non-maximum suppression refers to an algorithm that retains the pixels with the local maximum gradient magnitude and removes the blurred edges. The extreme pixel points refer to the pixels with the local maximum gradient magnitude retained after non-maximum suppression. The double-threshold processing refers to setting two thresholds, high and low, and connecting the pixels above the low threshold and connected to the high threshold edge. The pest image feature refers to the extreme pixel points after double-threshold processing. The filter refers to a component that removes noise in the sound frequency. The discrete cosine transform refers to a mathematical transformation method that converts a signal from the time domain to the frequency domain. The pest thermal image refers to an image of the temperature distribution in the pest body area formed after infrared thermal imaging processing. The average temperature calculation refers to traversing the pixel points of the temperature distribution image in the pest body area, accumulating the temperature values expressed by the pixel points, and dividing by the number of pixel points to calculate the average temperature of the pest body area. For example, if the temperature of the v-th pixel point in the temperature distribution image of the pest body area is U(v), where v = 1, 2, 3... v - 1, v, according to the formula Calculate the average temperature ε(v) of the pest body area, where v is the number of all pixels in the temperature distribution image of the pest body area. The chemical sensor comprehensive processing refers to the technology that pheromone molecules of pests react with the metal oxide semiconductor sensor array to change the electrical properties of the metal oxide semiconductor sensor, so as to identify pest pheromones. The pest feature database refers to the pest image recognition results corresponding to the pest image features, pest sound frequency features, pest body temperature features, and pest pheromone features in historical materials. This embodiment does not limit the initialization method, such as normal distribution random initialization. This embodiment does not limit the setting method of preset parameters, such as empirical setting and adaptive adjustment.

[0123] Specifically, the pest image recognition unit extracts the target comprehensive features of the target pest data, effectively improving the pest recognition accuracy. At the same time, it integrates pest image features, pest sound frequency features, pest body temperature features, and pest pheromone features, broadens the recognition dimension, accurately distinguishes pests in a complex environment, improves the adaptability of the device, and also outputs the pest recognition result by constructing a pest recognition model, so as to adapt to various pest species and improve the pest recognition accuracy.

[0124] Specifically, the pest movement recognition unit constructs a pest movement prediction model according to the pest movement prediction model construction method. The pest movement prediction model construction method includes:

[0125] Step J01: Divide 70% of the data in the pest movement database into a movement training set, divide 20% of the data in the pest feature database into a movement validation set, and divide 10% of the data in the pest feature database into a movement test set;

[0126] Step J02: Select a convolutional neural network as the initial pest movement prediction model and initialize the weights and biases of the pest movement prediction model;

[0127] Step J03: Use a convolutional kernel with preset parameters in the convolutional layer of the initial pest movement prediction model parameters, select the Adam optimizer and the cross-entropy loss function to train the initial pest movement prediction model, load the movement training set into the initial pest movement prediction model, perform forward propagation through the initial pest movement prediction model, and calculate the output value of the initial pest movement prediction model;

[0128] Step J04: Calculate the loss function value according to the output value of the initial pest movement prediction model, calculate the gradient through the backpropagation algorithm, and update the weights and biases of the initial pest movement prediction model. Repeat the processes of forward propagation, loss function calculation, and backpropagation to obtain the calculated pest movement prediction model;

[0129] Step J05: Input the motion validation set into the calculated pest motion prediction model, optimize the parameters of the calculated pest motion prediction model to obtain the optimized pest motion prediction model.

[0130] Step J06: Test the accuracy of the optimized pest motion prediction model using the motion test set, and output the optimized pest motion prediction model with an accuracy rate meeting 90% as the pest motion prediction model.

[0131] Step J07: Sample the pest images and pest videos collected by the high-definition camera at a fixed frame rate to obtain the target image sequence.

[0132] Step J08: Input the target image sequence into the gated recurrent unit to obtain the pest motion prediction result.

[0133] Specifically, the motion training set refers to the data set in the pest motion database used to train the initial pest motion prediction model. The motion validation set refers to the data set in the pest motion database used to adjust the calculated pest motion prediction model. The motion test set refers to the data set in the pest motion database used to evaluate the optimized pest motion prediction model. The pest motion database refers to the pest motion direction and pest motion speed corresponding to the pest images and pest videos in the historical materials. In this embodiment, the fixed frame rate is not limited. For example, if the fixed frame rate is set to 25 frames per second, the target image sequence refers to a series of consecutive image sequences extracted by sampling the pest images at 25 frames per second. The gated recurrent unit refers to a model used to predict the trend of sequence data. The pest motion prediction result includes the pest motion direction and the pest motion speed.

[0134] Specifically, the pest motion recognition unit accurately predicts the future motion trend of pests by constructing a pest motion prediction model, provides a reliable analysis basis for the path planning unit, and improves the pest-catching efficiency.

[0135] Specifically, when the path planning unit plans the target insect-catching path, it sets the current position of the device as the starting point, obtains the position where the pest arrives after a preset time according to the pest movement prediction result, uses it as the target insect-catching point, and according to the A* search algorithm, obtains all paths from the starting point to the target insect-catching point, uses them as the set of possible paths, and then obtains the optimal insect-killing path in the set of possible paths according to the insect-killing cost function. It obtains the pest prediction speed B in the pest movement prediction result in real time for the device according to a preset frequency, compares the pest prediction speed B with each preset pest biological speed. The preset pest biological speeds include the first preset pest biological speed B1 and the second preset pest biological speed B2. Set B1 = 5m / s and B2 = 10m / s. It judges the change situation of the pest movement speed according to the comparison result, and adjusts the optimal insect-killing path of the device according to the judgment result, where:

[0136] When B > B2, the path planning unit determines that the change situation of the pest movement speed is abnormal, adjusts the optimal insect-killing path of the device, re-obtains the position where the pest arrives after a preset time according to the pest movement prediction result, uses it as the new target insect-catching point, and obtains the optimal insect-killing path according to the A* search algorithm;

[0137] When B1 ≤ B ≤ B2, the path planning unit determines that the change situation of the pest movement speed is normal and does not adjust the optimal insect-killing path of the device;

[0138] When B < B1, the path planning unit determines that the change situation of the pest movement speed is abnormal, adjusts the optimal insect-killing path of the device, re-obtains the position where the pest arrives after a preset time according to the pest movement prediction result, uses it as the new target insect-catching point, and obtains the optimal insect-killing path according to the A* search algorithm;

[0139] The path planning unit calculates the pest movement speed change value C according to the formula C = B - (B2 - B1), compares the pest movement speed change value C with the preset speed change value C0, judges the change amplitude of the pest movement speed according to the comparison result, and adjusts the preset frequency according to the judgment result, where:

[0140] When C ≤ C0, the path planning unit determines that the change amplitude of the pest movement speed is abnormal and does not update the preset frequency;

[0141] When C > C0, the path planning unit determines that the change amplitude of the pest movement speed is abnormal, updates the preset frequency, and shortens the preset frequency amplitude;

[0142] The path planning unit obtains the pest movement direction D and the real-time pest movement direction D0 in the pest movement prediction result, calculates the deviation E of the pest movement direction according to the formula E = |D - D0|, compares the deviation E of the pest movement direction with the preset deviation E0, judges the pest movement direction status according to the comparison result, and adjusts the pest movement speed according to the judgment result, where:

[0143] When E ≤ E0, the path planning unit determines that the pest movement direction status is normal and does not adjust the pest movement speed;

[0144] When E > E0, the path planning unit determines that the pest movement direction status is abnormal, adjusts the pest movement speed, and directly determines that the change in the pest movement speed is abnormal.

[0145] Specifically, the device refers to an intelligent pest control device based on machine learning. The A* search algorithm refers to an existing search algorithm for finding the shortest path from the starting point to the target pest-catching point. By setting up an open list and a closed list, initializing the open list and the closed list, adding the starting point to the open list, calculating the shortest path from the starting point of the open list to the target pest-catching point according to the pest control cost function, and moving the starting point to the closed list, and continuing to process the current position of the device until the best pest control path is found. The pest control cost function refers to a function that comprehensively considers the actual cost from the starting point to the current position of the device and the estimated cost from the current position of the device to the target pest-catching point to evaluate the path priority. For example, according to the formula Calculate the pest control cost function δ(n′) of the current position of the device, where φ(n′) is the actual cost from the starting point to the current position of the device, is the estimated cost from the current position of the device to the target pest-catching point. In this embodiment, the preset frequency is not limited. For example, the preset frequency is set to 1 second / time. In this embodiment, the speed change value C0 is not limited. For example, C0 = 0.25 m / s. The amplitude shortening refers to reducing the frequency of the preset frequency. For example, reducing the preset frequency of 1 second / time to 0.5 second / time.

[0146] Specifically, the path planning unit takes the current position of the device as the starting point, determines the target pest-catching point based on the pest movement prediction result, and selects the best pest control path from many possible paths by means of the A* search algorithm, so as to ensure that the device moves efficiently to the target position. At the same time, by real-time monitoring the change in the pest movement speed and the deviation of the pest movement direction and comparing them with the preset values, it timely judges the change in the pest movement state, thereby re-planning the best pest control path, making the device always fit the pest movement change, and improving the accuracy of the pest-catching path and the pest control efficiency.

[0147] Specifically, when setting the pest control priority for the intelligent pest control unit, the breeding speed F of pests is obtained, and the breeding speed F of pests is compared with each preset breeding speed. The preset breeding speeds include the first preset breeding speed F1 and the second preset breeding speed F2. It is set that F1 = 42 pests per day and F2 = 0.56 pests per day. According to the comparison result, the pest breeding speed level is judged, and the pest control priority is set according to the judgment result, where:

[0148] When F < F1, the intelligent pest control unit determines that the pest breeding speed level is low and sets it as the third priority for pest control;

[0149] When F1 ≤ F ≤ F2, the intelligent pest control unit determines that the pest breeding speed level is medium and sets it as the second priority for pest control;

[0150] When F > F2, the intelligent pest control unit determines that the pest breeding speed level is high and sets it as the first priority for pest control;

[0151] The intelligent pest control unit compares the pest movement speed B with the preset pest movement speed B0, judges the influence degree of the pest movement speed according to the comparison result, and updates the pest breeding speed level according to the judgment result, where:

[0152] When B ≤ B0, the intelligent pest control unit determines that the influence degree of the pest movement speed is low and does not update the pest breeding speed level;

[0153] When B > B0, the intelligent pest control unit determines that the influence degree of the pest movement speed is high and updates the pest breeding speed level. The update coefficient is set as α, α = 1.3 - 0.3e -0.7*(B-B0) , and the updated breeding speed of pests is F1, F1 = α × F;

[0154] The intelligent pest control unit obtains the number of allergens H produced by pests and the pest disease transmission ability value R, pest disease transmission ability value J. The threat value K of pests to human health is calculated according to the formula K = 0.3 × H + 0.5 × R + 0.2J. The threat value K of pests to human health is compared with the preset threat value K0, the threat degree of pests to human health is judged according to the comparison result, and the pest movement speed is corrected according to the judgment result, where:

[0155] When K ≤ K0, the intelligent pest control unit determines that the threat degree of pests to human health is low and does not correct the pest movement speed;

[0156] When K > K0, the intelligent pest control unit determines that the threat degree of pests to human health is high and corrects the pest movement speed. The update threat coefficient is set as β, β = 1.48 - 0.2e -0.1*(K-K0), the corrected pest movement speed is B3, and B3 = β × B;

[0157] The intelligent pest control unit selects a pest control method according to the pest recognition result, and initializes the selection of the pest control method to the electric shock method;

[0158] The intelligent pest control unit obtains the pest density ρ and compares it with the preset pest density ρ0, judges the pest quantity status according to the comparison result, and adjusts the pest control method according to the judgment result, where:

[0159] When ρ ≤ ρ0, the intelligent pest control unit determines that the pest quantity status is normal and does not adjust the pest control method;

[0160] When ρ > ρ0, the intelligent pest control unit determines that the pest quantity status is abnormal, adjusts the pest control method, and adjusts the electric shock method to the spray pest control method;

[0161] After the intelligent pest control unit reaches the preset working duration, it obtains the current pest quantity ρ1 and the pest quantity ρ2 one week ago, and calculates the pest killing rate M according to the formula compares the pest killing rate M with the preset pest killing rate M0, judges the pest killing situation according to the comparison result, and replaces the chemical drug in the chemical drug storage tank according to the judgment result, where:

[0162] When M > M0, the intelligent pest control unit determines that the pest killing situation is normal and does not replace the chemical drug in the chemical drug storage tank;

[0163] When M ≤ M0, the intelligent pest control unit determines that the pest killing situation is abnormal, replaces the chemical drug in the chemical drug storage tank, recalculates the pest killing rate after replacing the chemical drug in the chemical drug storage tank, and if M ≤ M0, optimizes the pest recognition model. The process includes:

[0164] The intelligent pest control unit obtains the accuracy rate A1 of the pest recognition model and compares it with the preset model accuracy rate A0, judges the accuracy rate situation of the pest recognition model according to the comparison result, and optimizes the pest recognition model according to the judgment result, where:

[0165] When A1 < A0, the intelligent pest control unit determines that the accuracy of the pest recognition model is abnormal, optimizes the pest recognition model, independently constructs a pest recognition model with the target pest data collected by the sound sensor, the infrared sensor, and the chemical sensor, sets the weight of the target pest data collected by the sound sensor as ω1, the weight of the target pest data collected by the infrared sensor as ω2, and the weight of the target pest data collected by the chemical sensor as ω3, where ω1 + ω2 + ω3 = 1, allocates weights in multiple batches according to a preset ratio, determines the final pest recognition result until the intelligent pest control unit determines that the accuracy of the pest recognition model is normal;

[0166] When A1 ≥ A0, the intelligent pest control unit determines that the accuracy of the pest recognition model is normal and adjusts the electrocution method to a spray pest control method.

[0167] Specifically, the electrocution method refers to the method in which the motor driver controls the connecting rod and conveys the electric quantity through the connecting rod to the electric shock paddle to beat the pests. The spray pest control method refers to the method in which the drug nozzle evenly sprays the chemical drug in the chemical drug storage tank in the form of mist to the target pest-catching point under the action of pressure. In this embodiment, the preset pest killing rate M0 is not limited. For example, M0 = 95%. In this embodiment, the preset pest movement speed B0 is not limited. For example, B0 = 1.5 m / s. In this embodiment, the preset threat value K0 is not limited. For example, K0 = 0.5. The pest density refers to the number of pests per unit area. The pest density is set as ρ3, the number of pests identified in the pest image is n, and the coverage area of the field of view of the high-definition camera is z square meters. The pest density ρ3 is obtained according to the calculation formula ρ2 = n / z. In this embodiment, the preset working duration is not limited. For example, the preset working duration is set as one week. In this embodiment, the preset ratio is not limited. For example, ω1 = 0.2, ω2 = 0.5, ω3 = 0.3. The multiple-batch weight allocation refers to adjusting the preset ratio multiple times to allocate the weight of the pest recognition model.

[0168] Specifically, the intelligent pest control unit scientifically sets the pest control priority according to the pest reproduction speed, the pest movement speed, and the threat value to human health, ensures that the pests with the greatest harm are processed first, rationally allocates pest control resources, flexibly switches the pest control method according to the pest recognition result and the pest density, improves the pest control effectiveness and the adaptability of the device, and at the same time judges the pest control effect by calculating the pest killing rate, timely replaces the chemical drug and optimizes the pest recognition model to continuously improve the pest control effect.

[0169] So far, the technical solution of the present invention has been described in connection with the preferred embodiments shown in the accompanying drawings. However, it is easily understood by those skilled in the art that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the protection scope of the present invention.

Claims

1. Schematic structural diagram of an intelligent insecticidal device based on machine learning, characterized in that, Comprising: An infrared sensor, which is connected to the protective housing and is used to collect the body temperature of pests; A high-definition camera, which is connected to the protective housing and the MCU processor and is used to collect pest images and pest videos; A chemical sensor, which is connected to the protective housing and is used to collect pest pheromones; A sound sensor, which is connected to the protective housing and is used to collect the sound frequency of pests; A connecting rod, which is connected to the electric shock bat, the electric shock driver and the protective housing and is used to transmit electric power to the electric shock bat; An electric shock bat, which is connected to the connecting rod and is used to strike pests; A pulley block, which includes a first pulley, a second pulley, a third pulley and a fourth pulley. The pulley block is connected to the protective housing and is used to drive the device to move; A chemical drug storage tank, which is connected to the drug spray head and the protective housing and is used to store chemical drugs; A drug spray head, which is connected to the chemical drug storage tank and is used to spray the chemical drugs in the chemical drug storage tank; An MCU processor, which is connected to the high-definition camera and the protective housing and contains an intelligent module inside and is used to process target pest control data; A motor driver, which is connected to the connecting rod and the protective housing and is used to control the connecting rod; A protective housing, which is connected to the infrared sensor, the high-definition camera, the chemical sensor, the sound sensor, the connecting rod, the first pulley, the second pulley, the third pulley, the fourth pulley, the chemical drug storage tank, the MCU processor and the electric shock driver, and is used to connect device components, protect the device and dissipate heat.

2. The intelligent pest control device based on machine learning according to claim 1, wherein the intelligent module includes: A data acquisition unit, which is used to acquire target pest data; A pest image recognition unit, which is used to extract the target comprehensive features of the target pest data, and is also used to construct a pest recognition model according to the pest recognition model construction method, and output a pest recognition result according to the target comprehensive features of the target pest data; A pest movement recognition unit, which is used to construct a pest movement prediction model according to the pest movement prediction model construction method, and output a pest movement prediction result according to the target pest data; A path planning unit, which is used to plan a target pest control path according to the pest movement prediction result and make real-time adjustments to the target pest control path; An intelligent pest control unit, which is used to set the pest control priority according to the target pest data, select a pest control method according to the pest recognition result, and is also used to optimize the pest recognition model according to the pest killing rate.

3. The intelligent insect extermination device based on machine learning according to claim 2, wherein The data acquisition unit acquires the target pest data, and the target pest data includes pest image data, pest biological data and pest impact data, and the pest impact data includes pest reproduction speed, the number of allergens produced by pests, and the pest disease transmission ability value.

4. The intelligent insect extermination device based on machine learning according to claim 3, characterized in that The target comprehensive features of the target pest data include pest image features, pest sound frequency features, pest body temperature features and pest pheromone features. The pest image recognition unit extracts the pest image features in the target comprehensive features of the target pest data through a pest image feature extraction method, and the pest image feature extraction method includes: Step G01, perform Gaussian filtering on the pest image data in the target pest data to obtain a target smoothed image; Step G02: Calculate the first-order partial derivative of the target smoothed image to obtain the target gradient information; Step G03: Perform non-maximum suppression on the target gradient information to obtain the limit pixel points; Step G04: Perform double-threshold processing on the limit pixel points to obtain the pest image features; The pest image recognition unit extracts the pest sound frequency feature in the target comprehensive feature of the target pest data through the pest sound frequency feature extraction method. The pest sound frequency feature extraction method includes: inputting the pest sound frequency in the pest biological data of the target pest data into a filter to obtain the target sound frequency, and performing discrete cosine transform on the target sound frequency to obtain the pest sound frequency feature; The pest image recognition unit extracts the pest body temperature feature in the target comprehensive feature of the target pest data through the pest body temperature feature extraction method. The pest body temperature feature extraction method includes: performing infrared thermal imaging processing on the pest body temperature in the pest biological data of the target pest data to obtain the pest thermal image, and then calculating the average temperature of the pest thermal image to obtain the pest body temperature feature; The pest image recognition unit extracts the pest pheromone feature in the target comprehensive feature of the target pest data through the pest pheromone feature extraction method. The pest pheromone feature extraction method includes: performing comprehensive chemical sensor processing on the pest pheromone in the pest biological data of the target pest data to obtain the pest pheromone feature; The pest image recognition unit takes the pest image features, pest sound frequency features, pest body temperature features, and pest pheromone features as the target comprehensive features of the target pest data.

5. The intelligent pest control device based on machine learning according to claim 4, wherein When the pest image recognition unit constructs a pest recognition model based on the pest feature database, the process includes: Step K01: Divide 70% of the data in the pest feature database into the recognition training set, 20% of the data in the pest feature database into the recognition validation set, and 10% of the data in the pest feature database into the recognition test set; Step K02: Select a convolutional neural network as the initial pest recognition model and initialize the weights and biases of the initial pest recognition model; Step K03: Use a convolutional kernel with preset parameters in the convolutional layer of the initial pest recognition model parameters, select the Adam optimizer and the cross-entropy loss function to train the initial pest recognition model, load the recognition training set into the initial pest recognition model, perform forward propagation through the initial pest recognition model, and calculate the output value of the initial pest recognition model; Step K04: Calculate the loss function value according to the output value of the initial pest recognition model, calculate the gradient through the backpropagation algorithm, and update the weights and biases of the initial pest recognition model. Repeat the processes of forward propagation, loss function calculation, and backpropagation to obtain the calculated pest recognition model; Step K05: Input the recognition validation set into the calculated pest recognition model to optimize the parameters of the calculated pest recognition model and obtain the optimized pest recognition model; Step K06: Use the recognition test set to test the accuracy of the optimized pest recognition model, and output the optimized pest recognition model with an accuracy rate meeting 90% as the pest recognition model. Step K07: Input the target comprehensive features of the target pest data into the pest recognition model to obtain the pest recognition result.

6. The intelligent insecticidal device based on machine learning according to claim 2, characterized in that, When the path planning unit plans the target pest-catching path, it sets the current position of the device as the starting point, obtains the position where the pest arrives after a preset time according to the pest movement prediction result, uses it as the target pest-catching point, and according to the A* search algorithm, obtains all paths from the starting point to the target pest-catching point, uses them as the set of possible paths, and then obtains the optimal pest control path from the set of possible paths according to the pest control cost function. The pest movement prediction result is obtained in real time for the device at a preset frequency, and the pest prediction speed B in the pest movement prediction result is obtained. The pest prediction speed B is compared with each preset pest biological speed, and the preset pest biological speeds include the first preset pest biological speed B1 and the second preset pest biological speed B2. The change situation of the pest movement speed is judged according to the comparison result, and the optimal pest control path of the device is adjusted according to the judgment result, where: When B > B2, the path planning unit determines that the change situation of the pest movement speed is abnormal, adjusts the optimal pest control path of the device, and re-obtains the position where the pest arrives after a preset time according to the pest movement prediction result, uses it as the new target pest-catching point, and obtains the optimal pest control path according to the A* search algorithm. When B1 ≤ B ≤ B2, the path planning unit determines that the change situation of the pest movement speed is normal and does not adjust the optimal pest control path of the device. When B < B1, the path planning unit determines that the change situation of the pest movement speed is abnormal, adjusts the optimal pest control path of the device, and re-obtains the position where the pest arrives after a preset time according to the pest movement prediction result, uses it as the new target pest-catching point, and obtains the optimal pest control path according to the A* search algorithm.

7. The intelligent insect extermination device based on machine learning according to claim 6, wherein, The path planning unit calculates the pest movement speed change value C according to the pest movement speed change value formula, compares the pest movement speed change value C with the preset speed change value C0, judges the change range of the pest movement speed according to the comparison result, and adjusts the preset frequency according to the judgment result, where: When C ≤ C0, the path planning unit determines that the change range of the pest movement speed is abnormal and does not update the preset frequency. When C > C0, the path planning unit determines that the change range of the pest movement speed is abnormal, updates the preset frequency, and shortens the preset frequency by a certain amount. The path planning unit obtains the pest movement direction D and the pest real-time movement direction D0 in the pest movement prediction result, calculates the deviation E of the pest movement direction according to the pest movement direction deviation formula, compares the deviation E of the pest movement direction with the preset deviation E0, judges the pest movement direction situation according to the comparison result, and adjusts the pest movement speed according to the judgment result, where: When E ≤ E0, the path planning unit determines that the movement direction status of the pest is normal and does not adjust the movement speed of the pest; When E > E0, the path planning unit determines that the movement direction status of the pest is abnormal, adjusts the movement speed of the pest, and directly determines that the change in the movement speed of the pest is abnormal.

8. The intelligent insect extermination device based on machine learning according to claim 2, characterized in that, When setting the pest control priority, the intelligent pest control unit obtains the reproduction speed F of the pest, compares the reproduction speed F of the pest with each preset reproduction speed, and the preset reproduction speeds include the first preset reproduction speed F1 and the second preset reproduction speed F2. According to the comparison results, the intelligent pest control unit judges the reproduction speed level of the pest and sets the pest control priority according to the judgment results, where: When F < F1, the intelligent pest control unit determines that the reproduction speed level of the pest is low and sets it as the third priority for pest control; When F1 ≤ F ≤ F2, the intelligent pest control unit determines that the reproduction speed level of the pest is medium and sets it as the second priority for pest control; When F > F2, the intelligent pest control unit determines that the reproduction speed level of the pest is high and sets it as the first priority for pest control.

9. The intelligent insecticidal device based on machine learning according to claim 8, wherein The intelligent pest control unit compares the movement speed B of the pest with the preset pest movement speed B0, judges the influence degree of the pest movement speed according to the comparison results, and updates the reproduction speed level of the pest according to the judgment results, where: When B ≤ B0, the intelligent pest control unit determines that the influence degree of the pest movement speed is low and does not update the reproduction speed level of the pest; When B > B0, the intelligent pest control unit determines that the influence degree of the pest movement speed is high, updates the reproduction speed level of the pest, sets the update coefficient as α, and obtains the updated reproduction speed F1 of the pest; The intelligent pest control unit obtains the number H of allergens produced by the pest and the pest disease transmission ability value R, pest disease transmission ability value J, calculates the threat value K of the pest to human health according to the threat value formula of human health, compares the threat value K of the pest to human health with the preset threat value K0, judges the threat degree of the pest to human health according to the comparison results, and corrects the movement speed of the pest according to the judgment results, where: When K ≤ K0, the intelligent pest control unit determines that the threat degree of the pest to human health is low and does not correct the movement speed of the pest; When K > K0, the intelligent pest control unit determines that the threat degree of the pest to human health is high, corrects the movement speed of the pest, sets the update threat coefficient as β, and obtains the updated pest movement speed B3.

10. The intelligent insect extermination device based on machine learning according to claim 9, characterized in that, The intelligent pest control unit selects the pest control method according to the pest recognition result and initializes the selection of the pest control method as the electric shock method; The intelligent pest control unit compares the pest density ρ with the preset pest density ρ0, judges the pest quantity status according to the comparison results, and adjusts the pest control method according to the judgment results, where: When ρ ≤ ρ0, the intelligent pest control unit determines that the pest quantity status is normal and does not adjust the pest control method; When ρ > ρ0, the intelligent pest control unit determines that the pest quantity status is abnormal, adjusts the pest control method, and adjusts the electric shock method to the spray pest control method; After the intelligent pest control unit reaches the preset working duration, it obtains the pest killing rate M, compares the pest killing rate M with the preset pest killing rate M0, judges the pest killing situation according to the comparison result, and replaces the chemical medicine in the chemical medicine storage tank according to the judgment result, where: When M > M0, the intelligent pest control unit determines that the pest killing situation is normal and does not replace the chemical medicine in the chemical medicine storage tank; When M ≤ M0, the intelligent pest control unit determines that the pest killing situation is abnormal, replaces the chemical medicine in the chemical medicine storage tank, recalculates the pest killing rate after replacing the chemical medicine in the chemical medicine storage tank, and if M ≤ M0, optimizes the pest recognition model. The process includes: The intelligent pest control unit obtains the accuracy rate A1 of the pest recognition model and compares it with the preset model accuracy rate A0, judges the accuracy rate situation of the pest recognition model according to the comparison result, and optimizes the pest recognition model according to the judgment result, where: When A1 < A0, the intelligent pest control unit determines that the accuracy rate situation of the pest recognition model is abnormal, optimizes the pest recognition model, independently constructs a pest recognition model from the target pest data collected by the sound sensor, infrared sensor, and chemical sensor, sets the weight of the target pest data collected by the sound sensor as ω1, the weight of the target pest data collected by the infrared sensor as ω2, and the weight of the target pest data collected by the chemical sensor as ω3, where ω1 + ω2 + ω3 = 1, allocates the weights in multiple batches according to the preset ratio, and determines the final pest recognition result until the intelligent pest control unit determines that the accuracy rate situation of the pest recognition model is normal; When A1 ≥ A0, the intelligent pest control unit determines that the accuracy rate situation of the pest recognition model is normal and adjusts the electric shock method to the spray pest control method.

Images