Intelligent insect-catching robot based on deep learning

Abstract

The invention relates to the field of intelligent insect catching, in particular to an intelligent insect catching robot based on deep learning, which comprises a camera, a comprehensive acquisition node, a smell trapping agent conveying pipeline, a smell trapping agent storage tank, an MCU (Microprogrammed Control Unit) processor, a battery, a pest detention box, a first wheel, a second wheel, a rotary spray head and a motor driver, the intelligent processing module is deployed in the MCU processor and comprises a data acquisition unit, an environment sensing unit, a path planning unit, a trapping control unit and a decision adjustment unit, the environment sensing unit constructs a target environment map through moving object recognition and visual image analysis, the path planning unit optimizes a path in real time to cope with new obstacles, and the trapping unit is controlled to trap pests; the decision adjustment unit adjusts trapping decisions in real time by constructing and optimizing a pest recognition model, the accuracy and efficiency of pest catching are remarkably improved, and excellent pest catching performance and adaptability are shown.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent insect trapping, and particularly to an intelligent insect trapping robot based on deep learning. Background Art

[0002] As a place for cooking food, the kitchen often generates food residues, water vapor, and grease during the cooking process, making the kitchen environment humid and oily. Such an environment is extremely attractive to various pests, such as cockroaches and flies, seriously affecting people's health and quality of life. At present, there are many deficiencies in the prevention and control methods for kitchen pests. Existing simple insect trapping devices cannot distinguish between pests and non-pests. At the same time, these devices cannot automatically adjust the insect trapping strategy according to the activity patterns of kitchen pests and environmental changes, and have poor adaptability. Therefore, it is of great practical significance to develop an intelligent robot that can draw a map of the target environment, provide a basis for path planning, navigate autonomously in a complex and changeable environment, and actively search for pests.

[0003] Chinese Patent Publication No.: CN102037946B discloses a bionic automatic insect trap that uses light-emitting diodes to lure flying insects and kill them according to the phototaxis of flying insects. However, this solution still has problems such as difficulty in identifying dynamic changes in the environment, inability to actively avoid obstacles, and poor adaptability to the environment, and urgently needs an intelligent device that can navigate autonomously in a complex and changeable environment and actively search for pests to solve these problems. Summary of the Invention

[0004] Therefore, the present invention provides an intelligent insect trapping robot based on deep learning to overcome the problems in the prior art that the insect trapping device has insufficient adaptability to complex environments and cannot actively avoid obstacles to search for pests.

[0005] To achieve the above object, the present invention provides an intelligent insect trapping robot based on deep learning, including:

[0006] A camera, which is connected to the MCU processor and is used for collecting first target environment information;

[0007] An integrated acquisition node, which is connected to the MCU processor and is used for collecting second target environment information;

[0008] An odor attractant delivery pipeline, which is connected to an odor attractant storage tank and a rotary nozzle and is used for delivering the odor attractant;

[0009] An odor attractant storage tank, which is connected to the odor attractant delivery pipeline and the MCU processor and is used for storing the odor attractant;

[0010] An MCU processor, which is connected to a camera, a lidar, an odor attractant storage tank, a battery, a pest holding box, and a second wheel. The MCU processor includes an intelligent processing module for intelligently controlling an intelligent insect-catching robot based on deep learning;

[0011] A battery, which is connected to the MCU processor and the integrated acquisition node, for providing power to the intelligent insect-catching robot;

[0012] A pest holding box, which is connected to the MCU processor and the motor driver, and is provided with a movable baffle for holding pests;

[0013] A first wheel, which is connected to the pest holding box, for the movement of the intelligent insect-catching robot;

[0014] A second wheel, which is connected to the MCU processor, for the movement of the intelligent insect-catching robot;

[0015] A rotary nozzle, which is connected to the odor attractant delivery pipeline, for dispersing the odor attractant at multiple angles;

[0016] A motor driver, which is connected to the pest holding box, for controlling the movable baffle in the pest holding box 7.

[0017] Further, the intelligent processing includes:

[0018] A data acquisition unit for acquiring target insect-catching data;

[0019] An environment perception unit for performing moving object recognition and visual image analysis on the environment where the intelligent insect-catching robot is located according to the target insect-catching data, and obtaining a target environment map;

[0020] A path planning unit for performing path planning according to the target environment map to obtain a target planned path, and also for real-time optimization of the target planned path;

[0021] A control trapping unit for controlling the intelligent insect-catching robot to reach the target insect-catching point according to the target planned path, and also for generating a trapping decision according to the target insect-catching data, and trapping pests according to the trapping decision to obtain trapped pests;

[0022] A decision adjustment unit for identifying and analyzing the trapped pests, and for real-time adjustment of the trapping decision according to the identification and analysis results.

[0023] Further, the data acquisition unit acquires the target insect-catching data, and the target insect-catching data includes first target environment information and second target environment information. Among them, the first target environment information is acquired through a camera, and the second target environment information is acquired through the integrated acquisition node.

[0024] Further, the environment perception unit performs moving object recognition on the environment where the intelligent insect-catching robot is located according to the moving object recognition method, and obtains a moving object recognition result. The moving object recognition method includes: acquiring the current point cloud of the first object according to the lidar, acquiring the environmental point cloud within the historical sampling time period, performing average calculation, and obtaining target static data, and inputting the target static data into the point cloud processing library to obtain a static data set;

[0025] The environment perception unit takes the closest point cloud corresponding to the current point cloud of the first object in the static data model as the target closest point cloud, calculates the distance d of the target point cloud according to the Euclidean distance formula, and sets where Xs is the position value of the target closest point cloud on the X-axis of the space coordinate system, Ys is the position value of the target closest point cloud on the Y-axis of the space coordinate system, Zs is the position value of the target closest point cloud on the Z-axis of the space coordinate system, Xc is the position value of the current point cloud of the first object on the X-axis of the space coordinate system, Yc is the position value of the current point cloud of the first object on the Y-axis of the space coordinate system, and Zc is the position value of the current point cloud of the first object on the Z-axis of the space coordinate system.

[0026] Further, the environment perception unit calculates the centroid coordinates Cu of the current dynamic point cloud of the first object in the U-th frame and the centroid coordinates Cu2 of the current dynamic point cloud of the first object in the U2-th frame, where:

[0027] U2 - U = 1;

[0028]

[0029] where xu is the position value of the centroid of the current dynamic point cloud of the first object in the U-th frame on the X-axis of the space coordinate system, yu is the position value of the centroid of the current dynamic point cloud of the first object in the U-th frame on the Y-axis of the space coordinate system, zu is the position value of the centroid of the current dynamic point cloud of the first object in the U-th frame on the Z-axis of the space coordinate system, xu2 is the position value of the centroid of the current dynamic point cloud of the first object in the U2-th frame on the X-axis of the space coordinate system, yu2 is the position value of the centroid of the current dynamic point cloud of the first object in the U2-th frame on the Y-axis of the space coordinate system, zu2 is the position value of the centroid of the current dynamic point cloud of the first object in the U2-th frame on the Z-axis of the space coordinate system, xi is the position value of the i-th point in the current dynamic point cloud of the first object on the X-axis of the space coordinate system, yi is the position value of the i-th point in the current dynamic point cloud of the first object on the Y-axis of the space coordinate system, zi is the position value of the i-th point in the current dynamic point cloud of the first object on the Z-axis of the space coordinate system, and N is the number of points in the current dynamic point cloud of the first object;

[0030] The environmental perception unit calculates the displacement △x in the X-axis direction, the displacement △y in the Y-axis direction, and the displacement △z in the Z-axis direction of the centroid of the current dynamic point cloud of the first object in the spatial coordinate system according to the centroid coordinate Cu of the current dynamic point cloud of the first object in the U-th frame and the centroid coordinate Cu2 of the current dynamic point cloud of the first object in the U2-th frame. It is set that △x = xu - xu2, △y = yu - yu2, and △z = zu - zu2;

[0031] The environmental perception unit calculates the velocity component Vx in the X-axis direction, the velocity component Vy in the Y-axis direction, and the velocity component Vz in the Z-axis direction of the current dynamic point cloud of the first object in the spatial coordinate system according to the displacement △x in the X-axis direction, the displacement △y in the Y-axis direction, and the displacement △z in the Z-axis direction of the centroid of the current dynamic point cloud of the first object. It is set that is the time interval;

[0032] The environmental perception unit calculates the centroid velocity V of the current dynamic point cloud of the first object according to the velocity component Vx in the X-axis direction, the velocity component Vy in the Y-axis direction, and the velocity component Vz in the Z-axis direction of the centroid velocity of the current dynamic point cloud of the first object. It is set that

[0033] Further, the environmental perception unit performs visual image analysis on the environment where the intelligent insect-catching robot is located according to the visual image analysis method to obtain the visual image analysis result. The visual image analysis method includes: photographing pests through a camera, detecting pests in the first frame image to obtain the target pest pixels, and finding the target pest pixels in the consecutive frames after the first frame image through the optical flow algorithm to obtain the movement trajectory of the pests. The movement trajectory of the pests is used as the visual image analysis result;

[0034] The environmental perception unit annotates and records the results of moving object recognition and image analysis in the MCU processor to obtain the target environmental map.

[0035] Further, the path planning unit performs path planning on the target environmental map according to the path planning method to obtain the target planned path.

[0036] Further, when the path planning unit performs real-time optimization on the target planned path, after a new obstacle appears, the path planning unit obtains the density ρ of pests at the target insect-catching point through the camera, compares the density ρ of pests at the target insect-catching point with the preset density ρ0, judges the importance of the target insect-catching point according to the comparison result, and updates the target planned path according to the judgment result.

[0037] Further, when the control trapping unit controls the intelligent insect-catching robot to reach the target insect-catching point according to the target planned path, by obtaining the distance d2 between the current position and the target insect-catching point in the target insect-catching data, comparing the distance d2 between the current position and the target insect-catching point with the preset target distance d20, judging the position state between the current position and the target insect-catching point according to the comparison result, and making a decision on the release of the odor trapping agent at the current position according to the judgment result;

[0038] The control trapping unit generates a trapping decision according to the target insect-catching data through a trapping decision generation method.

[0039] Further, the control trapping unit obtains the odor trapping agent emission time E in the target insect-catching data, compares the odor trapping agent emission time E with the preset emission time E0, judges the sufficiency of the odor trapping agent emission time according to the comparison result, and processes the release of the odor trapping agent according to the judgment result.

[0040] Compared with the prior art, the beneficial effect of the present invention is that by using a camera, a lidar and a variety of sensors to cooperate to collect target insect-catching data, and using an intelligent processing module to plan paths and avoid new obstacles, the intelligent insect-catching robot can actively avoid obstacles to find pests, thereby improving the adaptability of the intelligent insect-catching robot to complex environments. Description of the Drawings

[0041] Figure 1 is a schematic structural diagram of the intelligent insect-catching robot based on deep learning in this embodiment;

[0042] Figure 2 is a schematic structural diagram of the intelligent processing module in this embodiment. Detailed Embodiments

[0043] 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.

[0044] The preferred embodiments of the present invention will be described below with reference to the drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present invention and do not limit the protection scope of the present invention.

[0045] Please refer to Figure 1 shown, which is a schematic structural diagram of the intelligent insect-catching robot based on deep learning in this embodiment. The device includes:

[0046] A camera 1, which is connected to an MCU processor 5 and is used for collecting first target environmental information;

[0047] The integrated acquisition node 2 is connected to the MCU processor 5 and is used for acquiring the second target environmental information;

[0048] The odor attractant delivery pipeline 3 is connected to the odor attractant storage tank 4 and the rotary nozzle 10 and is used for delivering the odor attractant;

[0049] The odor attractant storage tank 4 is connected to the odor attractant delivery pipeline 3 and the MCU processor 5 and is used for storing the odor attractant;

[0050] The MCU processor 5 is connected to the camera 1, the lidar 2, the odor attractant storage tank 4, the battery 6, the pest holding box 7 and the second wheel 9. The MCU processor 5 includes an intelligent processing module for intelligently controlling the intelligent insect-catching robot based on deep learning;

[0051] The battery 6 is connected to the MCU processor 5 and the integrated acquisition node 2 and is used for supplying power to the intelligent insect-catching robot;

[0052] The pest holding box 7 is connected to the MCU processor 5 and the motor driver 11 and is provided with a movable baffle for holding pests;

[0053] The first wheel 8 is connected to the pest holding box 7 and is used for the movement of the intelligent insect-catching robot;

[0054] The second wheel 9 is connected to the MCU processor 5 and is used for the movement of the intelligent insect-catching robot;

[0055] The rotary nozzle 10 is connected to the odor attractant delivery pipeline 3 and is used for dispersing the odor attractant at multiple angles;

[0056] The motor driver 11 is connected to the pest holding box 7 and is used for controlling the movable baffle in the pest holding box 7.

[0057] Specifically, the intelligent insect-catching robot is applied to kitchen pest trapping. The information of the environment where the intelligent insect-catching robot is located is acquired through the collaborative operation of the lidar and the camera. The intelligent insect-catching robot is moved to the target pest-trapping point according to the optimal path and the odor attractant is released to trap and hold the pests, so as to remove the pests in the space and improve the accuracy of pest trapping.

[0058] Specifically, the camera 1 acquires the first target environmental information, and the first target environmental information includes object images, pest images and pest density;

[0059] When the comprehensive acquisition node 2 acquires the second target environmental information, it acquires the second target environmental information through a lidar, an optical sensor, a position sensor, a time sensor, an infrared sensor, a direction sensor, a counting sensor, an odor sensor, a timer, a temperature sensor, a wind direction sensor, and a counter;

[0060] The odor attractant delivery pipe 3 uses high-purity copper as the pipe material to deliver the odor attractant;

[0061] The odor attractant storage tank 4 uses stainless steel as the material to store the odor attractant;

[0062] The MCU processor 5 intelligently processes the first target environmental information and the second target environmental information, and controls the intelligent insect-catching robot to trap pests;

[0063] The battery 6 uses a storage battery to provide electrical energy for the normal insect-catching activities of the intelligent insect-catching robot;

[0064] The pest holding box 7 uses a cuboid semi-transparent plastic box integrated with a movable baffle to hold pests. At the same time, the pest holding box 7 is integrated with a thin-film pressure sensor to calculate the number of pests;

[0065] The first wheel 8 and the second wheel 9 use solid wheels to carry the intelligent insect-catching robot to move to the target insect-catching point according to the target planned path;

[0066] The rotary nozzle 10 is composed of a nozzle body, a rotating mechanism, a nozzle, and a sealing component. After the intelligent insect-catching robot reaches the target insect-catching point, the odor attractant stored in the odor attractant storage tank 4 is released in a 360° rotation to trap pests;

[0067] The motor driver 11 is composed of a control circuit, a power drive circuit, a protection circuit, and a shell with heat dissipation performance, and controls the opening and closing of the movable baffle to separate the pests from the outside world, thereby holding the pests.

[0068] Specifically, the first target environmental information refers to object images, pest images, and pest density. In this embodiment, the object images are not limited, such as device images and human body images. The pest images refer to the image data of pests in the scenario of trapping pests. The pest density refers to the number of pests per unit area. Assuming the pest density is ρ2, the number of pests identified in the pest image is n1, and the coverage area of the camera's field of view is g square meters. The pest density ρ2 is obtained according to the calculation formula ρ2 = n1 / g. The second target environmental information refers to pest image resolution, intelligent pest-catching robot position, new obstacle staying time, new obstacle passing times, new obstacle movement direction, the number of times the new obstacle's movement direction is consistent with that of the intelligent pest-catching robot, odor attractant concentration, odor attractant emission time, environmental temperature, wind direction, and pest quantity. In this embodiment, the manner of collecting the second target environmental information by the comprehensive acquisition node 2 is not limited. For example, it can be collected by a lidar. The odor attractant refers to chemical agents and natural substances developed according to the biological characteristics and olfactory preferences of different pests to attract pests. The high-purity copper refers to copper with a purity of over 99.95%. The stainless steel refers to steel that does not rust in weak corrosive media such as the atmosphere and fresh water. In this embodiment, the type of the storage battery is not limited, such as lithium batteries and nickel-metal hydride batteries. The movable baffle refers to a structural component used to control the access channel of the pest holding box. The thin-film pressure sensor refers to a sensitive element used to detect pressure changes and can calculate the number of pests in the pest holding box. In this embodiment, the material of the semi-transparent plastic box is not limited, such as a polyvinyl chloride box. In this embodiment, the material of the solid wheel is not limited, such as a solid rubber wheel and a solid metal wheel. The nozzle body refers to a component with a water flow channel inside for guiding water flow. The rotating mechanism refers to a component that drives the rotation shaft to rotate through the impact force of water flow, so as to evenly spray water flow by the nozzle. The nozzle refers to a component that forms an atomization effect on the sprayed water flow. The sealing component refers to a component used between each connection part to prevent water leakage and ensure the normal operation of the nozzle. The control circuit refers to the part that receives external input signals and processes these signals according to preset logic and algorithms. The power drive circuit refers to the part that converts the weak electrical signals output by the control circuit into strong electrical signals capable of driving the motor to operate, provides sufficient power and current for the motor, and drives the motor to rotate. The protection circuit refers to a circuit module used to monitor and protect the motor and the driver itself from various faults and abnormal conditions. In this embodiment, the housing with heat dissipation performance is not limited, such as a metal housing.

[0069] Specifically, the camera 1 helps the intelligent insect-catching robot accurately identify obstacles by collecting the first target environmental information, improving the accuracy of insect-catching. The comprehensive acquisition node 2 accurately collects the second target environmental information, enhancing the accuracy and efficiency of insect-catching. The odor trapping agent delivery pipeline 3 and the odor trapping agent storage tank 4 are made of high-purity copper and stainless steel respectively, ensuring the efficacy of the trapping agent and improving the insect-catching benefit. The MCU processor 5 adjusts the trapping process in real time to improve the insect-catching efficiency. The battery 6 provides continuous power supply to ensure the normal operation of the intelligent insect-catching robot. The pest detention box 7 centrally detains pests to prevent their spread. The first wheel 8 and the second wheel 9 ensure the stable movement of the robot. The rotating sprinkler 10 rotates to expand the spraying range, enhancing the insect-catching effect. The motor driver 11 precisely controls the movable baffle to reduce energy consumption and improve the insect-catching efficiency.

[0070] Please refer to Figure 2 as shown, which is a schematic structural diagram of the intelligent processing module of this embodiment. The intelligent processing module includes:

[0071] A data acquisition unit for acquiring target insect-catching data;

[0072] An environmental perception unit for identifying moving objects and analyzing visual images of the environment where the intelligent insect-catching robot is located according to the target insect-catching data, obtaining a target environmental map. The environmental perception unit is connected to the data acquisition unit;

[0073] A path planning unit for performing path planning according to the target environmental map to obtain a target planned path, and also for real-time optimization of the target planned path. The path optimization unit is connected to the environmental perception unit;

[0074] A control trapping unit for controlling the intelligent insect-catching robot to reach the target insect-catching point according to the target planned path, and also for generating a trapping decision according to the target insect-catching data and trapping pests according to the trapping decision to obtain detained pests. The odor trapping unit is connected to the path optimization unit;

[0075] A decision adjustment unit for identifying and analyzing the detained pests and making real-time adjustments to the trapping decision according to the identification and analysis results. The decision adjustment unit is connected to the control trapping unit.

[0076] Specifically, the intelligent processing module is applied to the MCU processor of the intelligent insect-catching robot in this embodiment. By intelligently processing the environmental information collected by the intelligent insect-catching robot, the target planned path of the intelligent insect-catching robot is obtained. The intelligent insect-catching robot reaches the target insect-catching point according to the target planned path, and traps pests according to the trapping decision. At the same time, the target planned path is optimized in real time and the trapping decision is adjusted in real time, so as to ensure the accuracy and efficiency in the insect-catching process. The data acquisition unit provides a strong information basis for subsequent analysis by acquiring the first target environmental information and the second target environmental information. The environmental perception unit obtains the detailed situation of the environment where the intelligent insect-catching robot is located by performing moving object recognition and visual image analysis on the first target environmental information and the second target environmental information, ensuring the accuracy and effectiveness of subsequent insect-catching. The path planning unit plans and optimizes the insect-catching path according to the detailed situation of the environment where the intelligent insect-catching robot is located, so that the intelligent insect-catching robot can quickly reach the target insect-catching point to trap pests and improve the efficiency of insect-catching. The control trapping unit generates a trapping decision according to the target trapping data to trap pests and efficiently detain the pests, improving the effectiveness of insect-catching. The decision adjustment unit adjusts the trapping decision in real time by identifying and analyzing the detained pests, making the trapping decision more suitable for the current environment and improving the efficiency and effectiveness of insect-catching.

[0077] Specifically, the data acquisition unit acquires target trapping data, and the target trapping data includes the first target environmental information and the second target environmental information. Among them, the first target environmental information is acquired through a camera, and the second target environmental information is acquired through a comprehensive acquisition node.

[0078] Specifically, the data acquisition unit continuously updates the environmental information of the camera and the comprehensive acquisition node in real time, so as to ensure the efficient pursuit of pests.

[0079] Specifically, the environmental perception unit performs moving object recognition on the environment where the intelligent insect-catching robot is located according to the moving object recognition method, and obtains the moving object recognition result. The moving object recognition method includes: obtaining the current point cloud of the first object according to the lidar, obtaining the environmental point cloud in the historical sampling time period and performing average calculation to obtain the target static data, and inputting the target static data into the point cloud processing library to obtain the static data set;

[0080] The environmental perception unit takes the closest point cloud corresponding to the current point cloud of the first object in the static data model as the target closest point cloud, calculates the distance d of the target point cloud according to the Euclidean distance formula, and sets Among them, Xs is the position value of the target's nearest point cloud on the X-axis of the spatial coordinate system, Ys is the position value of the target's nearest point cloud on the Y-axis of the spatial coordinate system, Zs is the position value of the target's nearest point cloud on the Z-axis of the spatial coordinate system, Xc is the position value of the current point cloud of the first object on the X-axis of the spatial coordinate system, Yc is the position value of the current point cloud of the first object on the Y-axis of the spatial coordinate system, Zc is the position value of the current point cloud of the first object on the Z-axis of the spatial coordinate system. Compare the target point cloud distance d with the preset point cloud distance d0, judge the state of the current point cloud of the first object according to the comparison result, and output the state of the current point cloud of the first object according to the judgment result, where:

[0081] When d ≤ d0, the environmental perception unit determines that the state of the current point cloud of the first object is static, and outputs the state of the current point cloud of the first object as the current static point cloud;

[0082] When d > d0, the environmental perception unit determines that the state of the current point cloud of the first object is dynamic, and outputs the state of the current point cloud of the first object as the current dynamic point cloud.

[0083] Specifically, the intelligent insect-catching robot refers to an intelligent insect-catching robot based on deep learning. Set the distance between the current point cloud of the first object and each point cloud in the static dataset as Ci, where i = 1, 2, 3... i-1, i. Compare the distances Ci between the point clouds with each other to obtain the closest distance between the current point cloud of the first object and each point cloud in the static dataset, and use it as the closest point cloud. The point cloud refers to a three-dimensional spatial data expression for obtaining the spatial coordinates of a large number of discrete points on the surface of an object. The current point cloud of the first object refers to a dataset describing the information of the first object collected by a lidar at the current moment. The historical sampling time period refers to multiple specific time lengths used to collect lidar point cloud data when constructing a static data model. The multiple specific time lengths include n2 sampling time lengths t. Set n2 = 240 and t = 5 seconds. In this embodiment, the multiple specific time lengths are not limited. For example, n2 = 360 can be set. In this embodiment, the sampling time length t is not limited. For example, t = 6s can be set. The environmental point cloud refers to a dataset describing the information of the objects in the environment where the intelligent insect-catching robot is located, collected by a lidar. The average calculation refers to adding up the values of the environmental point cloud collected within the sampling time length t in the X, Y, and Z axis directions respectively, and then dividing by the total number of environmental point clouds collected within the sampling time length t to obtain data representing the static characteristics of the environment where the intelligent insect-catching robot is located within the sampling time length t. The point cloud processing library refers to a set of pre-written program codes for performing noise reduction, filtering, and fitting operations on target static data. In this embodiment, the selection method of the point cloud processing library is not limited. For example, the PCL point cloud library can be used. The target point cloud distance refers to the distance between the current point cloud of the first object and the target closest point cloud. In this embodiment, the preset point cloud distance is not limited. For example, d0 = 0.2m can be set.

[0084] Specifically, the environmental perception unit constructs a static dataset by inputting target static data into the point cloud processing library, effectively determines the state of the current point cloud of the first object, provides a recognition basis for subsequent moving object recognition, and thus improves the efficiency of pest recognition.

[0085] Specifically, the environmental perception unit calculates the centroid coordinate Cu of the current dynamic point cloud of the first object in the U-th frame and the centroid coordinate Cu2 of the current dynamic point cloud of the first object in the U2-th frame, where:

[0086] U2 - U = 1;

[0087]

[0088] Among them, xu is the value of the position of the centroid of the current dynamic point cloud of the first object in the U-th frame on the X-axis of the spatial coordinate system, yu is the value of the position of the centroid of the current dynamic point cloud of the first object in the U-th frame on the Y-axis of the spatial coordinate system, zu is the value of the position of the centroid of the current dynamic point cloud of the first object in the U-th frame on the Z-axis of the spatial coordinate system, xu2 is the value of the position of the centroid of the current dynamic point cloud of the first object in the U2-th frame on the X-axis of the spatial coordinate system, yu2 is the value of the position of the centroid of the current dynamic point cloud of the first object in the U2-th frame on the Y-axis of the spatial coordinate system, zu2 is the value of the position of the centroid of the current dynamic point cloud of the first object in the U2-th frame on the Z-axis of the spatial coordinate system, xi is the value of the position of the i-th point in the current dynamic point cloud of the first object on the X-axis of the spatial coordinate system, yi is the value of the position of the i-th point in the current dynamic point cloud of the first object on the Y-axis of the spatial coordinate system, zi is the value of the position of the i-th point in the current dynamic point cloud of the first object on the Z-axis of the spatial coordinate system, and N is the number of points in the current dynamic point cloud of the first object;

[0089] The environmental perception unit calculates the displacement △x of the centroid of the current dynamic point cloud of the first object in the X-axis direction, the displacement △y in the Y-axis direction, and the displacement △z in the Z-axis direction of the spatial coordinate system according to the centroid coordinate Cu of the current dynamic point cloud of the first object in the U-th frame and the centroid coordinate Cu2 of the current dynamic point cloud of the first object in the U2-th frame, and sets △x = xu - xu2, △y = yu - yu2, △z = zu - zu2;

[0090] The environmental perception unit calculates the velocity component Vx of the current dynamic point cloud of the first object in the X-axis direction, the velocity component Vy in the Y-axis direction, and the velocity component Vz in the Z-axis direction of the spatial coordinate system according to the displacement △x of the centroid of the current dynamic point cloud of the first object in the X-axis direction, the displacement △y in the Y-axis direction, and the displacement △z in the Z-axis direction, and sets △t is the time interval;

[0091] The environmental perception unit calculates the centroid velocity V of the current dynamic point cloud of the first object according to the velocity component Vx of the centroid velocity of the current dynamic point cloud of the first object in the X-axis direction, the velocity component Vy in the Y-axis direction, and the velocity component Vz in the Z-axis direction of the spatial coordinate system, and sets The centroid velocity V of the current dynamic point cloud of the first object is compared with the preset velocity V0, the moving state of the current dynamic point cloud of the first object is judged according to the comparison result, and the recognition result of the first object is output according to the judgment result, where:

[0092] When V > V0, the environmental perception unit determines that the moving state of the current dynamic point cloud is fast moving, and outputs the recognition result of the first object as a pest to be determined;

[0093] When V ≤ V0, the environmental perception unit determines that the moving state of the current dynamic point cloud is slow movement, and outputs the first object recognition result as the device to be determined;

[0094] The environmental perception unit obtains the number of points m included in the point cloud of the pest to be determined through a lidar, compares the number of points m included in the point cloud of the pest to be determined with a preset number m0, determines the scale of the point cloud of the pest to be determined according to the comparison result, and outputs the recognition result of the pest to be determined according to the determination result, where:

[0095] When m < m0, the environmental perception unit determines that the scale of the point cloud of the pest to be determined is small, outputs the recognition result of the pest to be determined as a pest, and outputs the recognition result of the pest to be determined as the recognition result of the moving object;

[0096] When m ≥ m0, the environmental perception unit determines that the scale of the point cloud of the pest to be determined is large, outputs the recognition result of the pest to be determined as a device, and outputs the recognition result of the pest to be determined as the recognition result of the moving object.

[0097] Specifically, the time interval △t refers to the time interval between two adjacent frames of the current dynamic point cloud of the first object, the centroid coordinate refers to the coordinate value of the mass center of the object formed by the point cloud in space, the number of points refers to the number of discrete points that make up the point cloud, the centroid refers to the mass center representing the object formed by the point cloud, the timestamp information refers to the accurate time data recorded synchronously by the lidar when collecting each point cloud data, the space coordinate system refers to a mathematical framework composed of mutually perpendicular coordinate axes of the X-axis, Y-axis, and Z-axis to determine the position of an object in three-dimensional space, the velocity component refers to the magnitude of the velocity of the object in the direction of each coordinate axis of the space coordinate system, and the centroid velocity refers to the comprehensive velocity value that comprehensively considers the velocity components of the centroid of the current dynamic point cloud of the first object in the three coordinate axes of the space coordinate system.

[0098] Specifically, the environmental perception unit analyzes the dynamic point cloud to accurately judge the object state, initially identify the object type, and then further refine the identification from the dimension of the point cloud scale to determine whether the object is a device or a pest, improving the accuracy and reliability of the identification.

[0099] Specifically, the environmental perception unit performs visual image analysis on the environment where the intelligent insect-catching robot is located according to the visual image analysis method, and obtains the visual image analysis result. The visual image analysis method includes: photographing the pest through a camera, detecting the pest in the first frame image to obtain the target pest pixels, and using the optical flow algorithm to find the target pest pixels in the consecutive frames after the first frame image to obtain the moving trajectory of the pest, and taking the moving trajectory of the pest as the visual image analysis result;

[0100] The environment perception unit annotates and records the results of moving object recognition and image analysis in the MCU processor to obtain a target environment map.

[0101] Specifically, the first frame image refers to the image of the pest captured for the first time by the camera when performing visual image analysis on the environment where the intelligent insect-catching robot is located. The target pest pixels refer to the pixel points that jointly depict the shape and contour characteristics of the pest through its own color and brightness attributes. The optical flow algorithm refers to an algorithm used to analyze the movement of objects in a video sequence. For example, it is assumed that the movement of an object is continuous within a short period of time, and the movement trajectory of the pest in the image is tracked by calculating the displacement of corresponding pixel points in two adjacent frames. The continuous frames refer to a series of frames arranged in chronological order that record the scene changing over time.

[0102] Specifically, the environment perception unit obtains a target environment map based on the results of moving object recognition and image analysis, improving the insect-catching success rate per unit time.

[0103] Specifically, the path planning unit performs path planning on the target environment map according to the path planning method to obtain a target planned path. The path planning method includes:

[0104] Step S01: Construct a backtracking tree with the target environment map as the search space;

[0105] Step S02: Obtain the current position of the intelligent insect-catching robot in the target environment map through the position sensor integrated with the lidar, set it as the starting point, obtain the pest position through the results of moving object recognition, and set the pest position as the target insect-catching point;

[0106] Step S03: Create a backtracking tree at the starting point that only contains the starting point;

[0107] Step S04: Generate target potential positions where the intelligent insect-catching robot may move towards the target insect-catching point within the search space. Take the target potential positions as sampling points, move a distance j from the starting point towards the sampling point to obtain a second node, add the second node to the backtracking tree, and connect the starting point and the second node to obtain a first path. Obtain the Euclidean distance d1 between the second node and the target insect-catching point through the lidar. Compare the Euclidean distance d1 between the second node and the target insect-catching point with the preset insect-catching distance d01, judge the position distance between the second node and the target insect-catching point according to the comparison result, and output the path in the backtracking tree according to the judgment result, where:

[0108] When d1 ≥ d01, the path planning unit determines that the positional distance between the second node and the target insect-catching point is far, sets the second node as the new starting point, and repeats steps S01 to S04 until d1 < d01;

[0109] When d1 < d01, the path planning unit determines that the positional distance between the second node and the target insect-catching point is near, stops constructing the backtracking tree, and outputs all paths in the backtracking tree as the target planning path.

[0110] Specifically, the target potential position refers to the position points that the intelligent insect-catching robot may reach when moving towards the target insect-catching point. The search space refers to the spatial range within which the intelligent insect-catching robot can conduct searches and explorations during path planning. The backtracking tree is a tree-shaped data structure used to explore and record the possible paths of the intelligent insect-catching robot. It is constructed starting from the starting point of the intelligent insect-catching robot. By continuously generating sampling points and taking the node closest to the sampling point as the starting point of the new path, a new node is formed after moving a certain distance towards the sampling point and added to the tree. The node refers to the position where the intelligent insect-catching robot arrives at a certain moment. The Euclidean distance is a measurement method in mathematics used to measure the straight-line distance between two points in three-dimensional space. In this embodiment, the preset insect-catching distance d01 is not limited. For example, d01 = 10 cm is set.

[0111] Specifically, the path planning unit conducts path planning by constructing a backtracking tree, quickly generates the target planning path, enabling the intelligent insect-catching robot to quickly reach the range within a circle with the target insect-catching point as the center and a radius of d01 to carry out trapping work, thereby improving the insect-catching efficiency.

[0112] Specifically, when the path planning unit performs real-time optimization on the target planning path, after a new obstacle appears, the path planning unit obtains the density ρ of pests at the target insect-catching point through a camera, compares the density ρ of pests at the target insect-catching point with the preset density ρ0, judges the importance of the target insect-catching point according to the comparison result, and updates the target planning path according to the judgment result, where:

[0113] When ρ ≤ ρ0, the path planning unit determines that the importance of the target insect-catching point is low, does not update the target planning path, and the intelligent insect-catching robot stops moving and waits for the new obstacle to pass before continuing to move along the target planning path;

[0114] When ρ > ρ0, the path planning unit determines that the importance of the target insect-catching point is high, updates the target planning path, sets the position p1 where the intelligent insect-catching robot discovers the new obstacle as the new starting point of the backtracking tree, and regenerates the target planning path;

[0115] The path planning unit obtains the direction of the new obstacle's movement through the direction sensor in the comprehensive acquisition node, compares the direction of the new obstacle's movement with the direction of the target planned path. When the direction of the new obstacle's movement is consistent with the direction of the target planned path, it obtains the number of times b that the new obstacle's movement direction is consistent with the target planned path in a day, compares the number of times b that the new obstacle's movement direction is consistent with the target planned path in a day with the preset number of times b0, judges the frequency of the new obstacle's movement direction being consistent with the target planned path according to the comparison result, and corrects the target planned path of the intelligent insect-catching robot according to the judgment result, where:

[0116] When b ≤ b0, the path planning unit determines that the frequency of the new obstacle's movement direction being consistent with the target planned path is low frequency, and does not correct the target planned path of the intelligent insect-catching robot;

[0117] When b > b0, the path planning unit determines that the frequency of the new obstacle's movement direction being consistent with the target planned path is high frequency, corrects the target planned path of the intelligent insect-catching robot, sets the position p2 where the intelligent insect-catching robot discovers the new obstacle as the new starting point of the backtracking tree, and regenerates the target planned path;

[0118] The path planning unit obtains the staying time t1 of the new obstacle through the time sensor in the comprehensive acquisition node, compares the staying time t1 of the new obstacle with the preset staying time t10, judges the staying condition of the new obstacle according to the comparison result, and optimizes the target planned path of the intelligent insect-catching robot according to the judgment result, where:

[0119] When t1 ≤ t10, the path planning unit determines that the staying condition of the new obstacle is short-term staying, and does not optimize the target planned path of the intelligent insect-catching robot;

[0120] When t1 > t10, the path planning unit determines that the staying condition of the new obstacle is long-term staying, optimizes the target planned path of the intelligent insect-catching robot, and sets the position p3 where the intelligent insect-catching robot discovers the new obstacle as the new starting point of the backtracking tree, and regenerates the target planned path;

[0121] The path planning unit obtains the number of times a that the new obstacle passes through in a preset duration through the infrared sensor in the comprehensive acquisition node, compares the number of times a that the new obstacle passes through in the preset duration with the preset number of times a0, judges the frequency of the new obstacle passing through in the preset duration according to the comparison result, and adjusts the staying condition of the new obstacle according to the judgment result, where:

[0122] When a ≤ a0, the path planning unit determines that the frequency of the new obstacle passing through in a day is low frequency, and does not adjust the staying condition of the new obstacle;

[0123] When a > a0, the path planning unit determines that the frequency of the new obstacle passing by in a day is high, adjusts the staying condition of the new obstacle, and directly determines the staying condition of the new obstacle as a long-term stay.

[0124] Specifically, in this embodiment, the preset duration is not limited. For example, if the preset duration is set to 24 hours, the new obstacle refers to an object that is detected in real time through a camera and a lidar and that hinders the progress of the intelligent insect-catching robot on the walking path of the intelligent insect-catching robot, such as kitchen equipment. In this embodiment, the preset staying time t10 is not limited. For example, if t10 = 30s is set, the preset passing number a0 is not limited. For example, if a0 = 40 times is set, the preset number b0 is not limited. For example, if b0 = 6 times is set.

[0125] Specifically, the path planning unit analyzes the new obstacle during the movement of the intelligent insect-catching robot, and re-plans the path in a timely manner according to the analysis result to ensure that the intelligent insect-catching robot always moves towards the area with dense pests, thereby improving the insect-catching success rate.

[0126] Specifically, when the control trapping unit controls the intelligent insect-catching robot to reach the target insect-catching point according to the target planned path, by obtaining the distance d2 between the current position and the target insect-catching point in the target insect-catching data, compares the distance d2 between the current position and the target insect-catching point with the preset target distance d20, judges the position state between the current position and the target insect-catching point according to the comparison result, and makes a decision on the release of the odor attractant at the current position according to the judgment result, where:

[0127] When d2 ≤ d20, the control trapping unit determines that the position state between the current position and the target insect-catching point is a short distance, and releases the odor attractant at the current position;

[0128] When d2 > d20, the control trapping unit determines that the position state between the current position and the target insect-catching point is a long distance, does not release the odor attractant at the current position, and continues to move according to the target planned path until d2 ≤ d20;

[0129] The control trapping unit generates a trapping decision according to the target insect-catching data through a trapping decision generation method, and the trapping decision generation method includes:

[0130] Step G01, set the initial direction G2 of the release of the odor attractant by obtaining the direction where the pests are located through the target environmental map;

[0131] Step G02, use a rotary nozzle to release the odor attractant at a concentration of 15% and a odor emission speed of 20 mL per second in the initial direction;

[0132] Step G03: Obtain the current wind direction G1 through the wind direction sensor in the comprehensive acquisition node, calculate the emission direction deviation H according to the formula H = |G1 - G2|, compare the emission direction deviation H with the preset direction deviation H0, judge the deviation degree between the current wind direction and the initial direction according to the comparison result, and adjust the initial direction according to the judgment result, where:

[0133] When H ≤ H0, the control trapping unit determines that the deviation degree between the current wind direction and the direction where the pest is located is small, and does not adjust the initial direction;

[0134] When H > H0, the control trapping unit determines that the deviation degree between the current wind direction and the direction where the pest is located is large, and adjusts the initial direction until H ≤ H0.

[0135] Specifically, in this embodiment, the preset target distance d20 is not limited. For example, d20 = 10 cm is set. In this embodiment, the preset direction deviation H0 is not limited. For example, H0 = 30°. The current position refers to the position point where the intelligent insect-catching robot is located in the search space at the current moment.

[0136] Specifically, the control trapping unit determines whether to release the odor trapping agent by comparing the distance between the current position and the target insect-catching point, and at the same time adjusts the release direction of the odor trapping agent in real time, precisely controls the use timing of the odor trapping agent, and improves the utilization efficiency of the odor trapping agent.

[0137] Specifically, the control trapping unit obtains the odor trapping agent emission time E in the target insect-catching data, compares the odor trapping agent emission time E with the preset emission time E0, judges the sufficiency of the odor trapping agent emission time according to the comparison result, and processes the release of the odor trapping agent according to the judgment result, where:

[0138] When E < E0, the control trapping unit determines that the sufficiency of the odor trapping agent emission time is insufficient, and the intelligent insect-catching robot continues to emit the odor trapping agent;

[0139] When E ≥ E0, the control trapping unit determines that the sufficiency of the odor trapping agent emission time is sufficient, and the intelligent insect-catching robot stops emitting the odor trapping agent;

[0140] The control trapping unit obtains the odor trapping agent concentration F through the odor sensor in the comprehensive acquisition node, compares the odor trapping agent concentration F with the preset concentration F0, judges the odor trapping agent concentration state according to the comparison result, and adjusts the sufficiency of the odor trapping agent emission time according to the judgment result, where:

[0141] When F < F0, the control trapping unit determines that the concentration state of the odor attractant is insufficient and adjusts the sufficiency of the odor attractant emission time to insufficient;

[0142] When F ≥ F0, the control trapping unit determines that the concentration state of the odor attractant is sufficient and adjusts the sufficiency of the odor attractant emission time to sufficient;

[0143] The control trapping unit obtains the current ambient temperature W through the temperature sensor in the comprehensive acquisition node, compares the current ambient temperature W with each preset ambient temperature range. The preset ambient temperature ranges include the first preset ambient temperature W1 and the second preset ambient temperature W2. Set W1 = 15°C and W2 = 30°C. Judge the degree of the current ambient temperature according to the comparison result, and update the concentration state of the odor attractant according to the judgment result, where:

[0144] When W < W1, the control trapping unit determines that the degree of the current ambient temperature is too low, updates the concentration state of the odor attractant, sets the updated odor attractant concentration as F1, sets F1 = W1 / (W1 + W)×F, and updates the sufficiency of the odor attractant emission time according to the updated odor attractant concentration F1 and the preset concentration F0, where:

[0145] When W1 ≤ W ≤ W2, the control trapping unit determines that the degree of the current ambient temperature is appropriate and does not update the concentration state of the odor attractant;

[0146] When W > W2, the control trapping unit determines that the degree of the current ambient temperature is too high, updates the concentration state of the odor attractant, sets the updated odor attractant concentration as F2, sets F2 = W2 / (W + W2)×F, and updates the sufficiency of the odor attractant emission time according to the updated odor attractant concentration F2 and the preset concentration F0.

[0147] Specifically, when the temperature is abnormal, the control trapping unit adjusts the concentration of the odor attractant to ensure that the odor attractant maintains good attraction, enhances the trapping effect on pests, and thus improves the pest trapping ability of the intelligent pest trapping robot under different environmental conditions.

[0148] When the control trapping unit traps pests according to the trapping decision, the initial state of the movable baffle is set to the open state. According to the volume V0 of a single pest, the volume Vc of the pest holding box, and the initial pressure J0 of the pest holding box, according to the formula: Calculate the current pressure J in the pest holding box, where n is the number of pests. Compare the current pressure J in the pest holding box with the preset pressure J1, judge the current capacity in the pest holding box according to the comparison result, and control the movable baffle according to the judgment result, where:

[0149] When J < J1, the control trapping unit determines that the current capacity in the pest holding box is sufficient and does not control the movable baffle.

[0150] When J ≥ J1, the control trapping unit determines that the current capacity in the pest holding box is insufficient, controls the movable baffle, and quickly closes the movable baffle through the motor driver. The pests in the pest holding box at this time are taken as the trapped pests.

[0151] Specifically, the volume of a single pest refers to the volume of a single pest recorded by the lidar in the pest holding box. The initial pressure of the pest holding box refers to the pressure when there are no pests in the pest holding box. The current pressure in the pest holding box refers to the pressure when there are pests in the pest holding box under the current situation.

[0152] The control trapping unit realizes the intelligent detention of pests by combining the real-time pressure value monitored by the membrane pressure sensor, centralizes the management of pests, and thus improves the pest-catching efficiency.

[0153] Specifically, when the decision-making adjustment unit identifies and analyzes the trapped pests and makes real-time adjustments to the trapping decision according to the identification and analysis results, and constructs the pest identification model based on the pest image dataset, 80% of the pest image dataset is divided into the identification training set, 20% of the pest image dataset is divided into the identification verification set. The pre-trained weights that have been trained on a large-scale image dataset are used to initialize the identification model. The image size in the selected YOLOv5 version parameters is set to 416×416, the number of categories is set to 128, the learning rate for training is set to 0.01, the batch size is set to 16, and the number of training epochs is set to 200 rounds. The identification training set is input into the selected YOLOv5 version after setting the parameters for training. The parameters are continuously adjusted through the backpropagation algorithm, and the identification verification set is input into the selected YOLOv5 version after training to optimize the parameters of the selected YOLOv5 version. The selected YOLOv5 version with an accuracy meeting 90% is output as the pest identification model.

[0154] The decision-making adjustment unit calculates the pest identification rate α according to the formula α = S1 / S2×100%, where S1 is the number of pests output by the pest identification model and S2 is the number of pests in the pest holding box obtained by the membrane pressure sensor. The pest identification rate α is compared with the preset identification rate α0, and the compliance of the pest identification rate is judged according to the comparison result, and the performance of the identification model is updated according to the judgment result, where:

[0155] When α ≥ α0, the decision-making adjustment unit determines that the compliance of the pest identification rate is qualified and does not update the performance of the identification model.

[0156] When α < α0, the decision adjustment unit determines that the compliance of the pest recognition rate does not meet the standard, updates the performance of the recognition model, sets the resolution of the camera image shooting parameters to the highest, uses a ring flash to re-take pictures of the pests, and adds pictures similar to the pest images to the recognition training set to re-train the pest recognition model, obtaining an updated target pest recognition model;

[0157] The decision adjustment unit obtains the resolution R of the relevant pest image, compares the resolution R of the relevant pest image with the preset resolution R0, determines the compliance of the resolution of the relevant pest image according to the comparison result, and adjusts the pest recognition rate according to the determination result, where:

[0158] When R < R0, the decision adjustment unit determines that the compliance of the resolution of the relevant pest image does not meet the standard, updates the pest recognition rate, sets the adjusted pest recognition rate to α1, sets α1 = R0 / (R0 - R) × α, and adjusts the pest recognition rate according to the adjusted pest recognition rate α1 and the preset recognition rate α0;

[0159] When R ≥ R0, the decision adjustment unit determines that the compliance of the resolution of the relevant pest image meets the standard and does not adjust the pest recognition rate. At this time, the usage duration P of the odor attractant is obtained, the usage duration P of the odor attractant is compared with the preset usage duration P0, and P0 = 10 days is set. The attractiveness of the odor attractant to pests is determined according to the comparison result, and the odor attractant is processed according to the determination result, where:

[0160] When P ≤ P0, the decision adjustment unit determines that the attractiveness of the odor attractant to pests is strong and does not process the odor attractant;

[0161] When P > P0, the decision adjustment unit determines that the attractiveness of the odor attractant to pests is weak and replaces the odor attractant.

[0162] Specifically, the resolution refers to the amount of information stored in an image measured by the number of pixels that can be displayed in the horizontal and vertical directions. The ring flash refers to a ring-shaped lighting device surrounding the camera lens. The pest image dataset refers to images of relevant pests collected through the Internet. Annotation tools are used to draw bounding boxes around each pest image in the collected relevant pest images and label the pest categories to which they belong. The labeled relevant pest images are used as the pest image dataset. The large-scale image dataset refers to a large object detection, segmentation, and caption dataset created by the Microsoft team. The pre-trained weights refer to the model parameters with rich image features pre-trained on the large-scale image dataset. The selected YOLOv5 version refers to an object detection algorithm that predicts multiple bounding boxes and the probability of object categories within the boxes in a single image at one time. The image size refers to the width and height of the image input into the model. The number of classes refers to the total number of different object classes that the model needs to identify. The learning rate refers to a parameter that controls the step size of the model when updating parameters in each iteration. The batch size refers to the number of image samples that the model processes simultaneously in one training iteration. The number of training epochs refers to the number of times the model completely traverses the entire training dataset. The backpropagation algorithm refers to an algorithm used in the model training process to calculate the gradient of the loss function with respect to the model parameters and update the parameters. The recognition training set refers to the dataset in the pest image dataset used to train the selected YOLOv5 version. The recognition validation set refers to the dataset in the pest image dataset used to adjust the trained selected YOLOv5 version.

[0163] Specifically, the decision-making adjustment unit identifies and analyzes the pests in the pest detention box by constructing a pest recognition model, effectively avoiding mis-trapping, and at the same time optimizing the trapping decision-making in a targeted manner, thereby improving the pest trapping efficiency and effectiveness of the intelligent pest-catching robot.

[0164] So far, the technical solution of the present invention has been described in combination with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand 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 diagram of the structure of an intelligent insect-catching robot based on deep learning, characterized in that, Including: A camera, connected to an MCU processor, for collecting first target environmental information; A comprehensive collection node, connected to the MCU processor, for collecting second target environmental information; An odor attractant delivery pipeline, connected to an odor attractant storage tank and a rotary nozzle, for delivering odor attractant; An odor attractant storage tank, connected to the odor attractant delivery pipeline and the MCU processor, for storing odor attractant; An MCU processor, connected to the camera, lidar, odor attractant storage tank, battery, pest holding box and second wheel. The MCU processor includes an intelligent processing module for intelligently controlling the intelligent insect-catching robot; A battery, connected to the MCU processor and the comprehensive collection node, for supplying power to the intelligent insect-catching robot; A pest holding box, connected to the MCU processor and a motor driver, provided with a movable baffle for holding pests; A first wheel, connected to the pest holding box, for the movement of the intelligent insect-catching robot; A second wheel, connected to the MCU processor, for the movement of the intelligent insect-catching robot; A rotary nozzle, connected to the odor attractant delivery pipeline, for dispersing odor attractant at multiple angles; A motor driver, connected to the pest holding box, for controlling the movable baffle in the pest holding box 7.

2. The intelligent insect-catching robot based on deep learning according to claim 1, wherein the intelligent processing module includes: A data acquisition unit for acquiring target insect-catching data; An environment perception unit for performing moving object recognition and visual image analysis on the environment where the intelligent insect-catching robot is located according to the target insect-catching data to obtain a target environmental map; A path planning unit for performing path planning according to the target environmental map to obtain a target planned path, and also for real-time optimization of the target planned path; A control trapping unit for controlling the intelligent insect-catching robot to reach a target insect-catching point according to the target planned path, and also for generating a trapping decision according to the target insect-catching data and trapping pests according to the trapping decision to obtain trapped pests; A decision adjustment unit for identifying and analyzing the trapped pests and making real-time adjustments to the trapping decision according to the identification and analysis results.

3. The intelligent bug-catching robot based on deep learning according to claim 2, wherein The data acquisition unit acquires target insect-catching data, and the target insect-catching data includes first target environmental information and second target environmental information. Among them, the first target environmental information is acquired by the camera, and the second target environmental information is acquired by the comprehensive collection node.

4. The intelligent bug-catching robot based on deep learning according to claim 2, characterized in that The environment perception unit performs moving object recognition on the environment where the intelligent insect-catching robot is located according to the moving object recognition method to obtain a moving object recognition result. The moving object recognition method includes: obtaining the current point cloud of the first object according to the lidar, obtaining the environmental point cloud within a historical sampling time period and performing average calculation to obtain target static data, and inputting the target static data into a point cloud processing library to obtain a static data set; The environment perception unit takes the nearest point cloud corresponding to the current point cloud of the first object in the static data model as the target nearest point cloud, and calculates the distance d of the target point cloud according to the Euclidean distance formula.

5. The intelligent insect-catching robot based on deep learning according to claim 4, characterized in that, The environmental perception unit calculates the centroid coordinate Cu of the first object's current dynamic point cloud in the U-th frame and the centroid coordinate Cu2 of the first object's current dynamic point cloud in the U2-th frame; The environmental perception unit calculates the displacement △x in the X-axis direction, the displacement △y in the Y-axis direction, and the displacement △z in the Z-axis direction of the centroid of the first object's current dynamic point cloud based on the centroid coordinate Cu of the first object's current dynamic point cloud in the U-th frame and the centroid coordinate Cu2 of the first object's current dynamic point cloud in the U2-th frame; The environmental perception unit calculates the velocity component Vx in the X-axis direction, the velocity component Vy in the Y-axis direction, and the velocity component Vz in the Z-axis direction of the first object's current dynamic point cloud based on the displacement △x in the X-axis direction, the displacement △y in the Y-axis direction, and the displacement △z in the Z-axis direction of the centroid of the first object's current dynamic point cloud; The centroid velocity V of the first object's current dynamic point cloud is calculated based on the velocity component Vx in the X-axis direction, the velocity component Vy in the Y-axis direction, and the velocity component Vz in the Z-axis direction of the centroid velocity of the first object's current dynamic point cloud in the space coordinate system; 6. The intelligent insect-catching robot based on deep learning according to claim 5, characterized in that The environmental perception unit performs visual image analysis on the environment where the intelligent insect-catching robot is located according to the visual image analysis method to obtain the visual image analysis result. The visual image analysis method includes: photographing pests through a camera, detecting pests in the first frame image to obtain target pest pixels, and searching for target pest pixels in consecutive frames after the first frame image through the optical flow algorithm to obtain the movement trajectory of the pests, and taking the movement trajectory of the pests as the visual image analysis result; The environmental perception unit annotates and records the results of moving object recognition and the visual image analysis result in the MCU processor to obtain the target environmental map; 7. The intelligent insect-catching robot based on deep learning according to claim 2, wherein The path planning unit performs path planning on the target environmental map according to the path planning method to obtain the target planned path; 8. The intelligent insect-catching robot based on deep learning according to claim 7, characterized in that When the path planning unit performs real-time optimization on the target planned path, after a new obstacle appears, the path planning unit obtains the density ρ of pests at the target insect-catching point through a camera, compares the density ρ of pests at the target insect-catching point with the preset density ρ0, judges the importance of the target insect-catching point according to the comparison result, and updates the target planned path according to the judgment result; 9. The intelligent insect-catching robot based on deep learning according to claim 2, characterized in that, When the control trapping unit controls the intelligent insect-catching robot to reach the target insect-catching point according to the target planned path, by obtaining the distance d2 between the current position and the target insect-catching point in the target insect-catching data, compares the distance d2 between the current position and the target insect-catching point with the preset target distance d20, judges the position state between the current position and the target insect-catching point according to the comparison result, and makes a decision on the release of the odor attractant at the current position according to the judgment result; The control trapping unit generates a trapping decision according to the target insect-catching data through the trapping decision generation method.

10. The intelligent insect-catching robot based on deep learning according to claim 9, wherein The control trapping unit obtains the odor attractant emission time E in the target insect trapping data, compares the odor attractant emission time E with the preset emission time E0, judges the adequacy of the odor attractant emission time according to the comparison result, and processes the odor attractant release according to the judgment result.

Images