A recognition system and method for a robot suitable for night orchard tree root pest inspection

Abstract

The application provides a recognition system and method of a robot suitable for night orchard tree root pest inspection, the system relies on an autonomous pathfinding chassis, and utilizes a front visual sensor to perceive a path for automatic cruising between fruit tree rows. In the pest recognition method, an infrared camera is used to collect tree root part images, first, a filtering algorithm is used to process the images, to realize feature enhancement of target pests and background interference suppression; second, angle rotation and Gaussian blur and other data enhancement technologies are used to improve the model processing capacity; finally, a deep learning model based on a YOLOv8 architecture is relied on to analyze the images in real time, to accurately identify the pest type and spatial position. Then, the system drives a pesticide spraying device to perform directional pesticide spraying according to the recognition result. The method is suitable for tree root part pest detection under night lightless conditions, and through real-time intervention, the probability of pest occurrence can be reduced, and the orchard yield can be improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent production equipment technology for forestry and fruit industry and machine recognition technology, and in particular to a robot identification system and method suitable for nighttime inspection of tree root pests in orchards. Background Technology

[0002] Currently, there is considerable research both domestically and internationally on automated pest inspection robots for orchards. The main technical approaches focus on utilizing machine vision, multispectral imaging, and artificial intelligence algorithms to collect images and identify pests and diseases on the leaves, branches, and other above-ground parts of fruit trees during daytime when lighting conditions are good. These studies are primarily applied to the automated monitoring of diurnal pests such as aphids, spider mites, and fruit borers. However, existing solutions suffer from limitations in application scenarios and misaligned monitoring times. Field surveys and observations have revealed that most pests that damage the roots of fruit trees, such as scarab beetle larvae, cutworms, and weevils, are primarily active and forage at night. These pests typically emerge from the soil after nightfall and crawl upwards along the tree roots; their harmful behavior is relatively concealed and they exhibit nocturnal characteristics. Therefore, most existing inspection robots primarily operate during the daytime, and the conditions and types of pests they identify do not coincide with the activity of the target pests they need to eliminate, making them less effective against specific seasonal pests.

[0003] Currently, there is limited research on the above issues. Chinese patent application CN202510730519.3 discloses an intelligent field pest inspection method based on heat source feature recognition. This method uses an improved algorithm to extract multi-scale feature maps containing pest heat source features, combines preset pest morphology, heat source intensity, movement and other parameters to screen suspected pest areas, and achieves spatial positioning through parallax calculation.

[0004] The aforementioned invention provides a method for pest inspection based on heat source feature recognition, but it has some shortcomings in identifying pests at night. This method uses a binocular camera detection approach, which relies heavily on external light sources and is suitable for detecting larger pests during the day. However, target pests are generally small and emit little heat, so heat source recognition alone cannot completely detect pests present at the tree roots. At the same time, the working hours of existing inspection robots are misaligned with the activity time of target pests, making it difficult to directly capture the location of pests and causing the best time for pest control to be missed. Furthermore, in terms of pest control and removal, manual intervention is still required, which cannot be done in real time and has a certain lag, resulting in a certain loss of fruit tree yield. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a robot identification system and method suitable for nighttime orchard root pest inspection. It employs a single-frame image-based infrared weak target detection method, utilizing single-frame images captured by an infrared camera and combining them with a background suppression algorithm to extract weak infrared targets, thereby improving the reliability of image recognition in complex backgrounds. An autonomous pathfinding chassis equipped with a depth camera ensures the device can move between rows of fruit trees at night in the absence of light. Simultaneously, a deep learning model trained on the YOLOv8 architecture analyzes the images to perform real-time identification of pest types and locations. Finally, the system drives a pesticide spraying device to spray pesticides directionally onto the root areas of the identified pests, achieving real-time pest control, reducing the probability of pest occurrence, decreasing manual labor, and increasing fruit tree yield.

[0006] The present invention achieves the above-mentioned technical objectives through the following technical means.

[0007] A robot identification system for nighttime orchard root pest inspection includes an automatic identification chassis, a pest identification infrared camera, and a pesticide spraying device.

[0008] Both the insect identification infrared camera and the pesticide spraying device are located above the automatic identification walking chassis, with the insect identification infrared camera located on both sides in front of the pesticide spraying device.

[0009] The automatic walking chassis is equipped with a vision sensor, a sensor gimbal, a wheel motor, a Mecanum wheel, a battery pack, a pesticide storage tank, and a processing unit. The vision sensor is mounted on the sensor gimbal at the front end of the automatic walking chassis. The Mecanum wheel is connected to the wheel motor. The pesticide storage tank is located above the automatic walking chassis at the rear end. The battery pack is placed inside the automatic walking chassis. The processing unit is located in the middle of the upper part of the automatic walking chassis, between the camera bracket and the pesticide storage tank.

[0010] The pesticide spraying device includes a nozzle, a control valve, a hose, and a hydraulic pump; the hose connects the control valve and the pesticide storage tank; the hydraulic pump is installed above the control valve and connected to the nozzle.

[0011] The processing unit analyzes the coordinates of the pest, the angle of the gimbal, and the depth information provided by the RGB-D camera. It then calculates the three-dimensional spatial coordinates of the target point relative to the robot chassis coordinate system through inverse transformation. Subsequently, it opens the control valve and uses a hydraulic pump to draw the pesticide from the pesticide storage tank through a hose. Finally, it sprays the pesticide to the designated location through a nozzle.

[0012] The above solution includes the following steps:

[0013] Step S1: When the automatic identification walking chassis remains stationary, use the pest identification infrared camera to collect images of the root area of ​​each fruit tree from multiple angles. Through the image acquisition unit, capture images of the pest-prone areas of the fruit tree roots along the walking route from different directions and angles to obtain infrared image data.

[0014] Step S2: The acquired infrared image data is filtered by spatial domain to suppress the background of the infrared image and obtain a residual image, which more clearly highlights the small areas where the target pest has a temperature difference with the background.

[0015] Step S3: Using the YOLOv8 architecture, train a deep learning model to identify whether there are pests in the processed infrared image, and determine the specific type of pest in the image based on the trained model.

[0016] Step S4: If the image is determined to contain the target pest, the camera will capture images of the fruit tree area where the pest is located and send them to the processing unit for targeted pesticide spraying.

[0017] In the above scheme, the vision sensor is a three-dimensional vision sensor, which uses an RGB-D depth camera to simultaneously acquire color images and depth images. The structured light method is used, in which a specific wavelength of infrared laser is emitted by the structured light camera, and after a certain encoding, it is projected onto the obstacle. The position and depth information of the obstacle are obtained by calculating the distortion of the returned encoded pattern.

[0018] In the above scheme, the positioning and navigation technology used by the automatic walking chassis is based on machine vision. The vision sensor is installed at the front end of the automatic walking chassis through a sensor gimbal. The vision sensor obtains positioning information by converting pixel position, pixel distance to camera, camera coordinate system and robot coordinate system. After analysis by the processing unit, the automatic walking chassis travels the correct route.

[0019] In the above scheme, when the insect identification infrared camera acquires images, it adopts an infrared filtering method to achieve insect detection by estimating and suppressing the background of the infrared image. During the detection process, the original image is spatially filtered to estimate the background of the infrared image. The difference between the original image and the estimated background is calculated to generate a residual image. Then, the residual image is thresholded to detect insects.

[0020] In the above scheme, in step S3, before training the data in the deep learning model, the LabelImg tool is used to label the pests in the orchard tree root images taken at night and distinguish their different types; at the same time, the confidence test of the labeling quality of all images is performed to obtain the training set, validation set and test set.

[0021] In the above scheme, in step S3, the deep learning model is trained using the YOLOv8 architecture, the model is trained using the completed training set, and the data is enhanced using angle rotation, Gaussian blur and grid processing techniques.

[0022] In the above scheme, after the deep learning model identifies the presence of pests in the image, the processing unit analyzes the coordinates of the pests, the gimbal angle, and the depth information provided by the RGB-D camera, and calculates the three-dimensional spatial coordinates of the target point relative to the robot chassis coordinate system through inverse transformation; then the control valve is opened, and the pesticide is drawn out from the pesticide storage tank through a hose by a hydraulic pump, and finally the pesticide is sprayed to the designated location through the nozzle.

[0023] The above-mentioned plan is applicable to pest control in greenhouse vegetables.

[0024] The beneficial effects of this invention are as follows:

[0025] 1. This invention enables the robot to autonomously plan and adjust its route based on the actual terrain and planting layout of the orchard. Equipped with visual sensors, the robot can achieve self-positioning and navigation, adapting to irregular terrain. Furthermore, when the infrared camera is capturing images, the robot can autonomously stop and remain stable, allowing for multi-angle imaging of the fruit tree roots, thereby obtaining clearer and more complete images of pest characteristics.

[0026] 2. This invention uses infrared thermal imaging technology to replace the traditional visual recognition method that relies on visible light, enabling the robot to work in the dark. It solves the problem of the existing device's detection period coinciding with the active periods of pests, and focuses on detecting tree root areas where pests are prevalent, thereby reducing the probability of orchard pests and diseases.

[0027] 3. This invention integrates infrared recognition and pesticide spraying devices onto a single mobile platform, enabling immediate prevention and control upon detection. After detecting and locating pests, the system drives the spraying device to target the affected areas. This integrated operation not only improves the speed of pest control and effectively curbs the spread and reproduction of pests at night, but also reduces the problem of untimely pesticide application in traditional separate inspection and control models, thus improving the timeliness and reliability of pest control in orchards.

[0028] 4. This invention improves the system's recognition performance and reliability by using spatial filtering technology to enhance the acquired infrared images in real time. This processing method effectively suppresses noise interference from complex backgrounds and enhances the thermal contrast between the target pest and its surrounding environment, thereby significantly improving the image signal-to-noise ratio. The enhanced image provides subsequent deep learning models with more salient features, directly improving the accuracy and success rate of pest detection, and reducing false recognitions caused by image blurring or background interference.

[0029] 5. This invention achieves intelligent and precise application of pesticides by identifying the location of pests. This method not only effectively replaces traditional manual inspection and spraying operations, reducing manpower and labor intensity, but also significantly reduces the total amount of pesticides used through targeted application, thereby improving operational efficiency while reducing production costs and environmental pollution. Attached Figure Description

[0030] Figure 1 Therefore, the overall structural distribution diagram of the robot for nighttime orchard root pest inspection.

[0031] Figure 2 Therefore, the workflow diagram of the robotic pest identification device for nighttime orchard root pest inspection is as follows.

[0032] Figure 3 A schematic diagram of the chassis structure of the robot used for nighttime orchard root pest inspection.

[0033] Figure 4 A schematic diagram of the Mecanum wheel for a robot used to inspect for root pests in orchards at night.

[0034] Figure 5 This is a schematic diagram of the structure of a robotic pesticide spraying device for nighttime orchard root pest inspection.

[0035] Figure 6 Therefore, the control flowchart for the robot pesticide spraying during nighttime orchard root pest inspection is shown below.

[0036] The attached figures are labeled as follows:

[0037] 101-Vision sensor; 102-Sensor gimbal; 103-Wheel motor; 104-Wheel; 105-Battery pack; 106-Pesticide storage box; 107-Processing unit; 301-Sprayer head; 302-Control valve; 303-Hose; 303-Hydraulic pump. Detailed Implementation

[0038] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0039] In the description of this invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "axial", "radial", "vertical", "horizontal", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing this invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0040] Combined with appendix Figure 1 As shown, a robot identification system for nighttime orchard root pest inspection includes an automatic identification chassis 1, a pest identification infrared camera 2, and a pesticide spraying device 3.

[0041] The insect identification infrared camera 2 is located above the automatic identification walking chassis 1, the pesticide spraying device 3 is located above the automatic identification walking chassis 1, the insect identification infrared camera 2 is located on both sides in front of the pesticide spraying device 3, and the pesticide spraying device 3 is located in the middle of the rear of the insect identification infrared camera 2.

[0042] A robot identification system for nighttime inspection of tree root pests in orchards, not limited to nighttime pest and disease inspection in orchards, but also applicable to nighttime pest control in greenhouse vegetables, etc.

[0043] Combined with appendix Figure 2 As shown, a method for identifying pests on tree roots in orchards at night includes the following steps:

[0044] Step S1: When the automatic identification walking chassis 1 remains stationary, the pest identification infrared camera 2 is used to collect images of the root area of ​​each fruit tree from multiple angles. Through the image acquisition unit, images are captured from different directions and angles for the pest-prone areas of the fruit tree roots along the walking route to obtain infrared image data.

[0045] Specifically, in step S1, to achieve high-precision recognition, the system first controls the automatic recognition chassis to come to a complete stop, thus eliminating image motion blur caused by platform movement or vibration, and establishing a stable spatial reference system for subsequent image registration and 3D information estimation. Subsequently, driven by a controllable pan-tilt unit, the insect-identifying infrared camera sequentially captures images of the same tree root area from multiple angles, including left, front, and right, according to a preset program. The core principle of this multi-angle acquisition strategy is to overcome visual blind spots caused by the complex shape of tree roots, branches, leaves, or soil obstruction under a single viewpoint by changing the perspective, and to obtain more comprehensive information about the target area using the principle of multiple views. Simultaneously, the image acquisition unit adjusts the integration time and gain of the infrared sensor to ensure that it can capture raw thermal image data with a superior signal-to-noise ratio in nighttime environments. Essentially, this utilizes the subtle thermodynamic differences between the living insect and the surrounding soil, tree trunk, and other background elements to form detectable target features on the infrared spectrum, thereby providing high-quality raw data input for subsequent image enhancement and intelligent recognition modules.

[0046] During implementation, when the inspection robot, based on its navigation system, determines that it has arrived at the preset detection point on the target fruit tree, its automatic identification chassis immediately decelerates and comes to a complete stop, entering a zero-speed lock state to ensure stability. Subsequently, the controllable pan-tilt unit of the pest detection infrared camera begins operation, driving the lens to sequentially deflect to multiple predetermined angles according to a preset program, including a left-front oblique view, a view directly at the base of the tree root, and a right-front oblique view. At each angle, the camera adjusts its focal length and exposure parameters to capture high-definition images of the area centered on the tree root, covering the ground surface to the base of the trunk, acquiring a set of infrared thermal images containing different perspectives. After all images are captured, the image data, along with the corresponding location information, is transmitted to the processing unit, awaiting the next stage of image processing.

[0047] Step S2: The acquired infrared image data is filtered by spatial domain to suppress the background of the infrared image and obtain a residual image, which more clearly highlights the small areas where the target pest has a temperature difference with the background.

[0048] Specifically, in step S2, the system utilizes the difference in spatial frequency distribution between the target and the background in the infrared image for feature separation. Because the background has a large area of ​​continuity and a slowly changing temperature distribution, its energy is mainly concentrated in the low-frequency part of the spatial domain; while smaller pest targets, due to the subtle thermal gradient difference between themselves and their surroundings, appear as high-frequency detail components in the image. This step employs a spatial filtering algorithm based on background estimation and difference: First, a low-pass filter with an appropriately sized window is applied to the original infrared image, and a smoothed estimate of the background thermal field is obtained through convolution. This estimate effectively preserves the low-frequency background information of the image and filters out high-frequency targets and noise. Subsequently, a pixel-by-pixel difference operation is performed between the original image and the estimated background image to generate a residual image. In this residual image, the gray values ​​of the background region are suppressed to a near-zero uniform distribution, while target regions with thermal anomalies are highlighted as high-contrast bright or dark spots, greatly improving the signal-to-noise ratio of the image and providing optimized input with feature focusing and minimal background interference for subsequent accurate recognition based on deep learning.

[0049] During implementation, the system performs spatial filtering background suppression processing on the acquired raw infrared images: First, the image processing module performs convolution operations on each input image according to pre-set algorithm parameters to generate a smooth background estimation map, preserving the low-frequency thermal distribution features of large-scale backgrounds such as soil and tree trunks; then, through pixel-by-pixel difference operations, the original image is subtracted from the background estimation map to obtain the residual image; this processing can significantly suppress background interference and highlight pest targets with thermal radiation differences from the background as high-contrast feature points, thereby providing high signal-to-noise ratio input data for subsequent recognition stages.

[0050] Step S3: Using the YOLOv8 architecture, train a deep learning model to identify whether pests exist in the processed infrared image, and determine the specific type of pests in the image based on the trained model.

[0051] Specifically, in step S3, the preprocessed residual image is first fed into the CSPDarknet backbone network of the model. This network extracts multi-level features from low-level edges to high-level semantics in the image layer by layer through cross-stage local connectivity and spatial pyramid pooling structure. Subsequently, the feature pyramid network (FPN) and path aggregation network (PAN) structure perform bidirectional fusion of multi-scale features. The fused feature map then enters the decoupled detection head, and the class probability and bounding box coordinates of each preset anchor box are obtained through classification branch and regression branch respectively. Then, the non-maximum suppression algorithm is applied to filter the prediction boxes with the highest confidence and no overlap, and finally outputs the existence status and specific category of the pest, thus completing the pest identification from image features.

[0052] Step S4: If the image is determined to contain the target pest, the camera will capture images of the fruit tree area where the pest is located and send them to the processing system for targeted pesticide spraying.

[0053] Specifically, in step S4, after the model identifies and locates the pest, the system uses inverse kinematics analysis to determine the three-dimensional spatial coordinates of the target pest relative to the robot body, combining the pest's pixel coordinates in the infrared image with the known camera position, installation posture, and positional information provided by multi-angle views and depth sensors. Subsequently, the processing system generates corresponding control commands based on these spatial coordinates to drive the nozzle to align with the target area. Finally, the hydraulic pump is activated to build pressure, and the control valve is opened to spray pesticide onto the target area.

[0054] During implementation, based on the pixel coordinates of the identified pest images, the three-dimensional spatial coordinates of the target area relative to the robot chassis coordinate system are analyzed. Subsequently, the processing system calculates the deflection angle required to drive the nozzle through inverse kinematics based on the relationship between these spatial coordinates and the pose of the current nozzle gimbal. It then generates control commands to align the nozzle with the target area, and after restarting, sprays continuously for 2-5 seconds before closing the valve and stopping the pump. Finally, the system records the location of this application and resets the nozzle.

[0055] Before training the data in the deep learning model, the LabelImg tool was used to label the pests in the orchard tree root images taken at night and to distinguish their different types; at the same time, the confidence level of the labeling quality of all images was tested to obtain the training set, validation set and test set.

[0056] Combined with appendix Figure 3 As shown, the automatic walking chassis 1 includes a vision sensor 101, a sensor gimbal 102, a wheel motor 103, wheels 104, a battery pack 105, a pesticide storage tank 106, and a processing unit 107. The vision sensor 101 is mounted on the sensor gimbal 102 at the front end of the device. The wheels 104 are connected to the wheel motor 103. The pesticide storage tank 106 is located above the device at the rear end. The battery pack 105 is placed inside the device. The processing unit 107 is located in the middle of the upper part of the device, between the camera bracket 102 and the pesticide storage tank 106.

[0057] The positioning and navigation technology adopted by the automatic walking chassis 1 is mainly based on machine vision. The vision sensor 101 is installed at the front of the chassis through the sensor gimbal 102. The vision sensor obtains positioning information by converting pixel position, pixel distance to camera, camera coordinate system and robot coordinate system. After analysis by the processing unit 107, the chassis travel route is controlled.

[0058] Combined with appendix Figure 4As shown, the autonomous walking chassis uses Mecanum wheels, which have unique omnidirectional motion capabilities, enabling the robot to move laterally and diagonally without turning between narrow rows of fruit trees, greatly improving its mobility and passability in complex environments. Furthermore, this feature allows the infrared camera to accurately target each tree root area to be inspected and achieve stable parking, ensuring the acquisition of stable and clear images. At the same time, the ability to rotate in place and flexibly avoid obstacles optimizes the inspection path and improves the efficiency of nighttime inspection operations.

[0059] Combined with appendix Figure 5 As shown, the pesticide spraying device 3 includes a nozzle 301, a control valve 302, a hose 303, and a hydraulic pump 304; the hose 303 connects the control valve 302 and the pesticide storage tank 106; the hydraulic pump 304 is installed above the control valve 302 and connected to the nozzle 301.

[0060] After receiving the work instruction, the pesticide spraying device 3 first uses the hydraulic pump 304 to pump the pesticide from the pesticide storage tank 106 through the delivery hose 303, then opens the control valve 302, and finally sprays the pesticide precisely to the area where the identified pests are located through the nozzle 301.

[0061] Combined with appendix Figure 6 As shown, the pesticide spraying device control system, after the deep learning model identifies the location of the tree roots where the pest is located, the processing unit 107 opens the control valve 302, and the pesticide is drawn out from the pesticide storage tank 106 through the hose 303 by the hydraulic pump 304, and finally sprayed to the designated location through the nozzle 301.

[0062] The control process is as follows: First, the preprocessed image is input into the trained deep learning model for recognition and analysis; then, the system determines whether there is a target pest in the current image based on the model output. If no pest is found, the control valve remains closed, and the system directly proceeds to the processing loop of the next image. If a pest is found, the system first drives the nozzle of the pesticide spraying device to turn towards the target area, then opens the control valve to accurately spray the identified pest. After spraying, the system continues to process the subsequently acquired images, and this process is repeated cyclically.

[0063] A method for identifying pests on tree roots in orchards at night, and its specific implementation process:

[0064] The autonomous walking chassis 1 propels the entire device forward along a preset path. During movement, the RGB module in the vision sensor 101 mounted on the sensor gimbal 102 acquires a color image of the path ahead for path recognition. The processing unit 107 fuses the visual and depth information, driving the wheel motor 103 to move the Mecanum wheels 104 in combination. When the robot reaches the fruit tree based on its positioning information, the chassis stops and enters a locked state. Subsequently, the processing unit 107 sends instructions to the gimbal of the insect-identifying infrared camera 2, driving its lens to capture multi-view images and sending the image data and location information back to the processing unit 107. The image processing module within the processing unit 107 processes the acquired raw infrared images, employing a spatial filtering background suppression algorithm to enhance the target pests into high-contrast bright spots, improving the image's signal-to-noise ratio. The pre-processed residual image is then input into a deep learning recognition model. This model is trained based on the YOLOv8 architecture, and its training set contains a large number of nighttime infrared images labeled with different pests, employing data augmentation techniques such as angle rotation and Gaussian blur. The final output includes the presence of pests, their specific type, and their coordinates in the image. If no pests are detected, the robot chassis moves to the next target point. If pests are detected, the system enters the pesticide spraying process. First, the processing unit calculates the three-dimensional spatial coordinates of the target point relative to the robot chassis coordinate system based on the pest's spatial coordinates, camera intrinsic parameters, and gimbal angle. Then, based on the spatial coordinates, the nozzle 301 is driven to align with the target area, and the hydraulic pump 304 is started to establish stable pressure in the pipeline. After the pressure stabilizes, the control valve 302 is opened, and the pesticide, driven by pressure, is pumped from the storage tank 106 through the hose 303 and sprayed onto the root area of ​​the identified pest. After 2-5 seconds, the control valve closes, the hydraulic pump stops, the nozzle resets, and the system moves to the next row of fruit trees.

[0065] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0066] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention without departing from the principles and spirit of the present invention.

Claims

1. A recognition system for a robot suitable for nighttime inspection of tree root pests in orchards, characterized in that, It includes an automatic identification walking chassis (1), an insect identification infrared camera (2), and a pesticide spraying device (3). The insect identification infrared camera (2) and the pesticide spraying device (3) are both located above the automatic identification walking chassis (1), and the insect identification infrared camera (2) is located on both sides in front of the pesticide spraying device (3). The automatic walking chassis (1) is equipped with a vision sensor (101), a sensor gimbal (102), a wheel motor (103), a Mecanum wheel (104), a battery pack (105), a pesticide storage box (106), and a processing unit (107). The vision sensor (101) is mounted on the sensor gimbal (102) at the front end of the automatic walking chassis (1). The Mecanum wheel (104) is connected to the wheel motor (103). The pesticide storage box (106) is located above the automatic walking chassis (1) at the rear end. The battery pack (105) is placed inside the automatic walking chassis (1). The processing unit (107) is located in the middle of the automatic walking chassis (1), between the camera bracket (102) and the pesticide storage box (106). The pesticide spraying device (3) includes a nozzle (301), a control valve (302), a hose (303), and a hydraulic pump (304); the hose (303) connects the control valve (302) and the pesticide storage tank (106); the hydraulic pump (304) is installed above the control valve (302) and connected to the nozzle (301); The processing unit (107) analyzes the coordinates of the pest, the angle of the gimbal and the depth information provided by the RGB-D camera, calculates the three-dimensional spatial coordinates of the target point relative to the robot chassis coordinate system through inverse transformation, then opens the control valve (302) and draws the pesticide from the pesticide storage tank (106) through the hose (303) via the hydraulic pump (304), and finally sprays the pesticide to the designated location through the nozzle (301).

2. The identification method of the identification system for a robot suitable for nighttime orchard root pest inspection according to claim 1, characterized in that, Includes the following steps: Step S1: When the automatic identification walking chassis (1) remains stationary, the pest identification infrared camera (2) is used to collect images of the root area of ​​each fruit tree from multiple angles. Through the image acquisition unit, images are captured from different directions and angles for the pest-prone areas of the fruit tree roots along the walking route to obtain infrared image data. Step S2: The acquired infrared image data is filtered by spatial domain to suppress the background of the infrared image and obtain a residual image, which more clearly highlights the small areas where the target pest has a temperature difference with the background. Step S3: Using the YOLOv8 architecture, train a deep learning model to identify whether there are pests in the processed infrared image, and determine the specific type of pest in the image based on the trained model. Step S4: If the image is determined to be a target pest, the camera will capture the area of ​​the fruit tree where the pest is located and send it to the processing unit (107), and pesticide will be sprayed in a targeted manner.

3. The identification method of the identification system for a robot suitable for nighttime orchard root pest inspection according to claim 2, characterized in that, The vision sensor (101) is a three-dimensional vision sensor that uses an RGB-D depth camera to simultaneously acquire color images and depth images. It uses structured light method, which emits infrared lasers of a specific wavelength through the structured light camera, projects them onto the obstacle after a certain encoding, and calculates the distortion of the returned encoded pattern to obtain the position and depth information of the obstacle.

4. The identification method of the identification system for a robot suitable for nighttime orchard root pest inspection according to claim 2, characterized in that, The positioning and navigation technology used in the automatic walking chassis (1) is based on machine vision. The vision sensor (101) is installed at the front end of the automatic walking chassis (1) through the sensor gimbal (102). The vision sensor obtains positioning information by converting pixel position, pixel distance to camera, camera coordinate system and robot coordinate system. After analysis by the processing unit (107), the automatic walking chassis (1) is controlled to travel the route.

5. The identification method of the identification system for a robot suitable for nighttime orchard root pest inspection according to claim 2, characterized in that, When the insect identification infrared camera (2) acquires images, it adopts an infrared filtering method to detect insects by estimating and suppressing the background of the infrared image. During the detection process, the original image is spatially filtered to estimate the background of the infrared image. The original image and the estimated background are then compared to generate a residual image. Finally, the residual image is thresholded to detect insects.

6. The identification method of the identification system for a robot suitable for nighttime orchard root pest inspection according to claim 2, characterized in that, In step S3, before training the data in the deep learning model, the LabelImg tool is used to label the pests in the orchard tree root images taken at night and distinguish their different types; at the same time, the confidence level of the labeling quality of all images is tested to obtain the training set, validation set and test set.

7. The identification method of the identification system for a robot suitable for nighttime orchard root pest inspection according to claim 1, characterized in that, In step S3, the deep learning model is trained using the YOLOv8 architecture, utilizing the completed training set to train the model, and employing angle rotation, Gaussian blur, and grid processing techniques to enhance the data.

8. The identification method of the identification system for a robot suitable for nighttime orchard root pest inspection according to claim 1, characterized in that, After the deep learning model identifies the presence of pests in the image, the processing unit (107) analyzes the coordinates of the pests, the angle of the gimbal, and the depth information provided by the RGB-D camera, and calculates the three-dimensional spatial coordinates of the target point relative to the robot chassis coordinate system through inverse transformation; then the control valve (302) is opened, and the pesticide is drawn out from the pesticide storage tank (106) through the hose (303) by the hydraulic pump (304), and finally the pesticide is sprayed to the designated location through the nozzle (301).

9. The identification method of the identification system for a robot suitable for nighttime orchard root pest inspection according to claim 1, characterized in that, Suitable for pest control in greenhouse vegetables.

Images