A weeding device and method based on video recognition

Abstract

The present invention discloses an electronic ejection weeding device, comprising multiple weeding mechanisms, including a main body carrier, a shovel blade, an electromagnetic shovel mechanism, a camera assembly, an electromagnetic shovel controller, and a video algorithm processor. During use, the camera collects image data; the video algorithm processor processes the data transmitted by the camera and transmits weeding instructions to the electromagnetic shovel controller, which controls the electromagnetic shovel to remove weeds. The present invention also discloses a weeding method, including a video algorithm. Compared to existing technologies, the present invention is more accurate and utilizes a learning algorithm. Similarly, for contour recognition, the device achieves higher accuracy and speed.

Description

Technical Field

[0001] The present invention relates to a mechanical weeding technology, and in particular to a weeding device and method based on video recognition. Background Art

[0002] Existing weeding devices have the technical problem of being easily damaged. The components of the weeder are relatively complex and easily damaged. In addition, they are not effective on rugged or obstructed terrain and are prone to getting stuck or damaged, affecting their effectiveness and lifespan. Existing video recognition weeding methods, such as the one disclosed in CN108271765B, are based on HSV color recognition to identify sample contours, but are affected by environmental factors such as light and wind. Models established without deep learning have no accuracy. In addition, the above patent uses HSV color recognition to first obtain a binary plant image, and then performs noise reduction processing to obtain plant contours, which is too redundant. Summary of the Invention

[0003] In order to solve the above-mentioned defects of the prior art, the present invention provides an electronic ejection weeding device, which can increase the service life of the weeding device and adapt to the complex terrain environment of field operations.

[0004] To achieve the above technical objectives, the present invention adopts the following technical solution: including a weeding mechanism, which is provided in multiple groups, each group of the weeding mechanism includes: a main body carrying box, including a shovel box, an axial upper cover connected to the upper edge of the front end of the shovel box, a tail end closing cover connected to the upper edge of the rear end of the shovel box, a traction ring provided at the front end of the shovel box, an electromagnetic shovel opening window provided on the side of the shovel box, a battery slot provided inside the shovel box, and a load bracket fixed to the bottom of the shovel box;

[0005] The shovel blade is in the shape of an inverted trapezoid and is fixed to the rear of the main body carrying box through shovel blade fixing brackets on both sides;

[0006] The electromagnetic shovel mechanism includes multiple sets of electromagnetic shovels arranged in sequence at 45 degrees along the X axis, and a drive battery;

[0007] An electromagnetic shovel controller is used to control the electromagnetic shovel to remove weeds;

[0008] The video algorithm processor processes the weed image data and transmits the weeding instruction to the electromagnetic shovel controller.

[0009] Preferably, the electromagnetic shovel includes a coil, a shovel box, a shovel rod, an electromagnetic shovel blade, and a return spring.

[0010] Preferably, the electromagnetic shovel includes multiple sets of coils.

[0011] Preferably, the weeding mechanism is provided with a camera assembly, which is arranged inside the shovel box. The camera assembly includes a camera, a camera fixing bracket, and a shock-absorbing cotton block. The shock-absorbing cotton block is clamped between the camera and the load bracket. A glass window is provided on the side of the shovel box; the camera and the shovel head of the electromagnetic shovel are in the same direction.

[0012] Preferably, the electromagnetic shovel includes a U-shaped rubber head.

[0013] Preferably, the load bracket is supported from the bottom by a plurality of support studs, the bottoms of the support studs are riveted to the inner bottom of the main body load box, and the support studs are fixed with nuts and washers.

[0014] The present invention also provides a weeding method based on video recognition, characterized in that the video algorithm of the video algorithm processor includes the following steps:

[0015] The camera collects image data;

[0016] The video algorithm processor processes the data transmitted by the camera and transmits the grass-cutting instruction to the electromagnetic shovel controller;

[0017] Used to control the electromagnetic shovel to remove weeds.

[0018] Preferably, the video algorithm of the video algorithm processor includes the following steps:

[0019] S1: Use the camera on the weeding device to collect images of different weeds, different weed numbers between ridges, and weeds with different leaf occlusions in the field to increase data diversity. At the same time, add data samples under different weather conditions so that the model can adapt to different weather conditions.

[0020] S2: Based on the Pycharm operating platform, Python scripts are used to enhance the collected data, enrich the data set, and improve the robustness of the detection network model. At the same time, image dehazing technology is used based on the Pytorch framework to process the data set images to increase the image contrast;

[0021] S3: Extract SURF features from samples and build a data set;

[0022] S4: Labelimg is used to annotate the dataset so that the specific location coordinates of the target weeds can be read during network training. The dataset is divided into a validation set (val), a test set (text), and a training set (train).

[0023] S5: Replace the original CIoU loss function in the YOLOv8 network with G-IoU and specify the IoU threshold to increase the movement speed of the prediction box, which can effectively improve training speed and inference accuracy;

[0024] S6: Adjust the weight parameters of the three loss functions of the YOLOv8 network to achieve higher recognition accuracy;

[0025] S7: The collected data is transmitted to the network for weed identification.

[0026] Preferably, the G-IoU function introduces the minimum true box, with a value range of (-1, 1). When the IoU threshold is n, the algorithm is as follows:

[0027]

[0028] Among them, G is the minimum bounding box that can be surrounded by n bounding boxes, A is the real A bounding box of target recognition, B is the predicted B bounding box of target recognition, until the predicted box n. During the operation, G is calculated first, then the intersection of n bounding boxes is calculated, and then G minus the intersection is calculated. The ratio of this value to the area of ​​G is calculated, and then the IOU of all target recognition bounding boxes is subtracted from this ratio to get GIOU.

[0029] Preferably, the SURF feature extraction includes:

[0030] Image reading, image display;

[0031] Image grayscale conversion, RGB image to grayscale image;

[0032] Feature point detection: Use the Hessian matrix to detect feature points. Use the second-order derivative matrix in the x and y directions to measure the local curvature of the function. The determinant value represents the amount of change around the pixel point. The feature point needs to be the extreme point of the determinant value.

[0033] Feature point positioning: locate feature points by interpolating feature point neighboring information;

[0034] Direction positioning: By calculating the Haar wavelet transform of the pixels around the feature point in the x and y directions, and adding the transformation values ​​in the x and y directions within a certain angle range in the xy plane to form a vector, the longest of all vectors is the direction of the feature point;

[0035] Feature description: After the direction of the feature point is selected, the surrounding pixels are used as a reference to establish a descriptor;

[0036] Feature matching: Extract feature descriptors from multiple images, use the SURF algorithm for feature matching, and obtain matching points.

[0037] The technical solution provided by the present disclosure can achieve at least the following beneficial effects:

[0038] 1. The machine uses narrow, angled, thin-bladed blades attached to the main carrier to remove weeds between ridges. The blade's shape and size are determined by the field's dimensions. The main carrier features a front-retracting, rear-exiting design. Retracting forward reduces resistance between the carrier and the ground, while extending backward increases stability.

[0039] 2. Choose the electromagnetic catapult rod shovel weeding method, which is lower in cost, simpler to implement, consumes less power, is more efficient and more convenient than the laser method.

[0040] 3. The electromagnetic shovel can be driven by multiple sets of coils, which can make the effective movement stroke of the shovel rod completely equal to the length of the protective cover, minimizing the lateral size of the main load box. Compared with the electromagnetic spring lock with a single set of coils, the theoretical maximum force point of the lock core is only at the center point of the coil.

[0041] 4. The electromagnetic shovel can be driven by multiple sets of coils, which can effectively reduce power loss. Compared with the electromagnetic spring lock with a single set of coils, this device does not need to generate tension to counteract the reset spring at all times. This device directly generates kinetic energy from electrical energy, and consumes less power.

[0042] 5. The electromagnetic shovel can be used in two forms: cutting shovel and pulling shovel. For thick rhizomes, a cutting shovel can be used to cut them off. For small rhizomes, a U-shaped rubber head pulling shovel can be used. The lateral impact force and the large friction of the rubber head can be used to pull out smaller weeds.

[0043] 6. Compared with the HSV color model, the present invention is more accurate and adopts a learning algorithm. Compared with the patent in the background technology, it has higher accuracy and better speed for contour recognition.

[0044] 7. SURF algorithm adopted in the present invention An important advantage of SURF feature extraction technology is that it can process multiple images at the same time, thereby speeding up the processing speed, and another important advantage is that it can ignore the noise in the image, which can effectively detect important features in the image.

[0045] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic structural diagram of the electronic ejection weeding device of the present invention;

[0047] Figure 2 This is a schematic diagram of the structure of the electronic ejection weeding device of the present invention. Figure 2 ;

[0048] Figure 3 1. It is a schematic diagram of the top view of the electronic ejection weeding device of the present invention;

[0049] Figure 4 2. It is a schematic diagram of the load support structure of the electronic ejection weeding device of the present invention;

[0050] Figure 5 Schematic diagram of the electromagnetic shovel action of the electronic ejection weeding device of the present invention;

[0051] Figure 6 Schematic diagram of the multi-coil electromagnetic shovel structure of the electronic ejection weeding device of the present invention;

[0052] Figure 7 Schematic diagram of the multi-coil electromagnetic shovel in action of the electronic ejection weeding device of the present invention;

[0053] Figure 8 This is a system architecture diagram of the electronic ejection weed removal device, controller, and cloud processor of the present invention;

[0054] Figure 9 It is a schematic diagram of the shovel-shaped electronic ejection weeding device of the present invention;

[0055] Figure 10 It is a schematic flow chart of the weeding method of the present invention;

[0056] Figure 11 It is a schematic diagram of the video algorithm flow of the present invention;

[0057] Figure 12 Schematic diagram of the G-IoU function recognition frame of the present invention.

[0058] Reference numerals:

[0059] 1. Main body carrying box; 2. Shovel blade; 3. Electromagnetic shovel mechanism; 4. Camera assembly; 5. Electromagnetic shovel controller; 6. Video algorithm processor;

[0060] 11. Shovel box; 12. Axial upper cover; 13. Tail end closing cover; 14. Towing ring; 15. Electromagnetic shovel opening window; 16. Battery slot; 17. Load bracket;

[0061] 21. Shovel blade fixing bracket;

[0062] 31. Electromagnetic shovel; 32. Driving battery; 311. Coil; 312. Shovel box; 313. Shovel rod; 314. Electromagnetic shovel blade;

[0063] 315, return spring; 316, U-shaped rubber head; 317, permanent magnet core; P1, de-energized state; P2, energized state; C1, cutting shovel; C2, pulling shovel;

[0064] 41. Camera; 42. Camera mounting bracket; 43. Shock-absorbing cotton pad; 44. Glass window. DETAILED DESCRIPTION

[0065] The present invention will be described more fully below with reference to the accompanying drawings, in which exemplary embodiments of the present invention are illustrated. Obviously, all features disclosed in this specification, or steps in all methods or processes disclosed, except for mutually exclusive features and / or steps, can be combined in any way. Any feature disclosed in this specification, unless otherwise stated, can be replaced by other equivalent or alternative features having similar purposes, that is, unless otherwise described, each feature is only an example of a series of equivalent or similar features. The technical solution of the present invention is described in many aspects below in conjunction with the figures and embodiments.

[0066] like Figure 1 As shown, the electronic ejection weeding device of one embodiment of the present invention includes a weeding mechanism, which is provided with multiple groups, each group of weeding mechanisms includes: a main body carrier box 1, the main body carrier box 1 includes a shovel box 11, an axial upper cover 12 connected to the upper edge of the front end of the shovel box, a tail end cover 13 connected to the upper edge of the rear end of the shovel box, and a traction ring 14 provided at the front end of the shovel box. The position relationship of the traction ring 14 is shown in FIG. Figure 4 The electromagnetic shovel opening window 15 is set on the side of the shovel box, and the battery slot 16 is set inside the shovel box. The load bracket 17 is fixed to the bottom of the shovel box 11. The position relationship between the load bracket 17 and the shovel box 11 is detailed in Figure 4 ;

[0067] like Figure 2 As shown, the weeding mechanism also includes a shovel blade. The shovel blade 2 is in an inverted trapezoidal shape and is fixed to the rear of the main body carrying box 1 through the shovel blade fixing brackets 21 on both sides. In addition, Figure 3 The connection position relationship between the shovel blade 2 and the main body carrier box 1 and the position relationship between the shovel blade fixing bracket 21 and the main body carrier box 1 are shown.

[0068] like Figure 3 As shown, the weeding mechanism also includes an electromagnetic shovel mechanism 3, comprising multiple sets of electromagnetic shovels 31 arranged at 45° along the X-axis, and a drive battery 32. The electromagnetic shovels utilize a long-stroke DC push-pull electromagnet lock as their primary structure. The electromagnetic shovels are connected to the electromagnetic shovel controller via two wires, each carrying a 12V DC signal with an average current of 1A.

[0069] like Figure 1 As shown, the weeding mechanism also includes an electromagnetic shovel controller for controlling the electromagnetic shovel to remove weeds; a video algorithm processor for processing weed image data and then transmitting a weeding instruction to the electromagnetic shovel controller.

[0070] Continue to refer to Figure 3 The electromagnetic shovel 31 includes a coil 311, a shovel box 312, a shovel rod 313, an electromagnetic shovel blade 314, and a return spring 415. The shovel rod 313 includes a limit cap at the bottom. In this embodiment, the movement principle of the battery shovel is as follows: Figure 5When the electromagnetic shovel is not powered on, the weak thrust of the tail end return spring 315 pushes the shovel rod 313 back into the shovel box. When powered on, the coil 311 inside the shovel box 312 is energized to generate an instantaneous magnetic field, which attracts the shovel rod 313 to move forward quickly. When the main rod of the shovel rod 313 moves to the middle of the coil, the electromagnetic shovel controller 5 disconnects the power supply of the coil 311, and the shovel rod 313 continues to move forward by inertia until the limit cap at the bottom of the shovel rod 313 reaches the limit position.

[0071] like Figure 9 As shown, as an operative embodiment, the electromagnetic shovel 31 in the electronic ejection weeding device also includes a U-shaped rubber head as another shovel type. A cutting shovel is used to sever thick rhizomes, while a pulling shovel with the U-shaped rubber head can be used to pull out smaller weeds by utilizing the lateral impact force and the greater friction of the rubber head.

[0072] like Figure 4 As shown, as an implementable manner, the load bracket 17 is supported from the bottom by a plurality of support studs, the bottoms of the support studs are riveted to the inner bottom of the main load box, and the support studs are fixed with nuts and washers.

[0073] Another embodiment of the electromagnetic shovel 31 movement, wherein the electromagnetic shovel 31 includes multiple sets of coil structures, specifically referring to Figure 6 , Figure 7 The diagram shows the multi-coil electromagnetic shovel 31 in action. During the electromagnetic shovel's motion, the controller controls the current bidirectionally, capturing the position of the permanent magnet core in real time. In this embodiment, the discrete magnetic fields of the independent coils allow the shovel rod 313 to have a greater effective travel, reducing the initial position reserved for the shovel rod 313 in the shovel box 11. In the previous embodiment, the initial position of the shovel rod 313 was outside the integral coil 311, meaning that the reserved space in the shovel box 11 required twice the length of the integral coil 311. In this solution, while the length of the shovel box 312 remains unchanged, the shovel rod 313 can be completely pre-positioned inside the shovel box 312, maintaining the same effective travel. This also eliminates the need for the return spring 315, further reducing costs.

[0074] like Figure 1 、 3 As shown in FIG8 , as another embodiment of the present application, the weeding mechanism 1 further includes a video algorithm processor 6, a main unit capable of receiving video streams from 8-16 fixed-focus cameras with a resolution of no higher than 4K pixels. The processor can then identify the root and stem locations of specific crop varieties and non-target weeds at a speed no higher than 20 km / h. The camera 41 directly captures images of the root and stem locations of crops and weeds at a near-horizontal angle. This effectively reduces leaf obstruction caused by natural plant growth, the required longer shooting distance, and wind interference when shooting vertically.

[0075] The video algorithm processor 6 transmits the coordinate position, speed and time reference information of the target weed to the electromagnetic shovel controller 5. After calculation and processing, the controller 5 controls the corresponding electromagnetic shovel 31 to eject to the left or right to cut the roots of the weeds.

[0076] like Figure 1 、 3 As shown, the weeding mechanism is provided with a camera assembly 4, which is arranged inside the shovel box 11. The camera assembly 4 includes a camera 41, a camera fixing bracket 42, and a shock-absorbing cotton block 43. The shock-absorbing cotton block 43 is clamped between the camera 41 and the load bracket 17. A glass window 44 is provided on the side of the shovel box 11; the camera 41 is in the same direction as the shovel head of the electromagnetic shovel 31, and the camera 41 is used to collect weeding image data.

[0077] When in use, the weeding device of the present application is longitudinally pulled by the locomotive and moves along the ridge groove. The camera faces the side and rear of the moving direction, and the camera path is through the glass window of the main body carrying box, and shoots horizontally and slightly downward towards the root direction of the plant.

[0078] like Figure 10 As shown, the present application also discloses a weeding method, comprising:

[0079] The camera collects image data;

[0080] The video algorithm processor processes the data transmitted by the camera and then transmits the mowing instruction to the electromagnetic shovel controller;

[0081] Used to control the electromagnetic shovel to remove weeds.

[0082] like Figure 11 As shown, the video algorithm of the above video algorithm processor includes the following steps:

[0083] S1: Use the camera on the weeding device to collect images of different weeds, different weed numbers between ridges, and weeds with different leaf occlusions in the field to increase data diversity. At the same time, add data samples under different weather conditions so that the model can adapt to different weather conditions.

[0084] S2: Based on the Pycharm operating platform, Python scripts are used to enhance the collected data, enrich the data set, and improve the robustness of the detection network model. At the same time, image dehazing technology is used based on the Pytorch framework to process the data set images to increase the image contrast;

[0085] S3: Extract SURF features from samples and build a data set;

[0086] S4: Labelimg is used to annotate the dataset so that the specific location coordinates of the target weeds can be read during network training. The dataset is divided into a validation set (val), a test set (text), and a training set (train).

[0087] S5: Replace the original CIoU loss function in the YOLOv8 network with G-IoU and specify the IoU threshold to increase the movement speed of the prediction box, which can effectively improve training speed and inference accuracy;

[0088] S6: Adjust the weight parameters of the three loss functions of the YOLOv8 network to achieve higher recognition accuracy;

[0089] S7: The collected data is transmitted to the network for weed identification.

[0090] Furthermore, The G-IoU function introduces the minimum true box, with a value range of (-1, 1). When the IoU threshold is n, the algorithm is as follows:

[0091]

[0092] G is the minimum bounding box that can be surrounded by n bounding boxes, A is the real A bounding box for target recognition, B is the predicted B bounding box for target recognition, and so on until the predicted box n. During the operation, G is calculated first, then the intersection of n bounding boxes is calculated, and then G minus the intersection is calculated. The ratio of this value to the area of ​​G is calculated, and then the IOU of all target recognition bounding boxes is subtracted from this ratio to get GIOU.

[0093] Figure 12 The figure shows the relationship between A, B and G. A is the real A border for target recognition, and B is the predicted B border for target recognition. During the calculation, G is calculated first, then the intersection of AB is calculated, and then G minus the intersection is calculated. The ratio of this value to the area of ​​G is calculated, and then the IOU of AB is subtracted from this ratio to get GIOU.

[0094] Furthermore, SURF feature extraction includes:

[0095] Image reading, image display;

[0096] Image grayscale conversion, RGB image to grayscale image;

[0097] Feature point detection: Use the Hessian matrix to detect feature points. Use the second-order derivative matrix in the x and y directions to measure the local curvature of the function. The determinant value represents the amount of change around the pixel point. The feature point needs to be the extreme point of the determinant value.

[0098] Feature point positioning: locate feature points by interpolating feature point neighboring information;

[0099] Direction positioning: By calculating the Haar wavelet transform of the pixels around the feature point in the x and y directions, and adding the transformation values ​​in the x and y directions within a certain angle range in the xy plane to form a vector, the longest of all vectors is the direction of the feature point;

[0100] Feature description: After the direction of the feature point is selected, the surrounding pixels are used as a reference to establish a descriptor;

[0101] Feature matching: Extract feature descriptors from multiple images, use the SURF algorithm for feature matching, and obtain matching points.

[0102] Furthermore, the Training Set is split into two sets: Training Data and Validation Data, with a ratio of 9:1. Training is performed using the split Training Data. After each epoch, validation is performed using validation data that the machine has not seen during training. The model is adjusted based on the loss value obtained from the validation set.

[0103] The above description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A weed control method, characterized in that: An electronic ejection weeding device is used, including a weeding mechanism, which is provided with multiple groups. Each group of weeding mechanisms includes: a main body carrying box, including a shovel box, a shaft-moving upper cover connected to the upper edge of the front end of the shovel box, a tail end closing cover connected to the upper edge of the rear end of the shovel box, a traction ring provided at the front end of the shovel box, an electromagnetic shovel opening window provided on the side of the shovel box, a battery slot provided inside the shovel box, and a load bracket fixed to the bottom of the shovel box; a shovel blade, which is an inverted trapezoidal shape and is fixed to the rear of the main body carrying box by shovel blade fixing brackets on both sides; an electromagnetic shovel mechanism, including multiple groups of electromagnetic shovels arranged in sequence at 45 degrees along the X-axis, and a drive battery; an electromagnetic shovel controller for controlling the electromagnetic shovel to remove weeds; a video algorithm processor for processing weed image data and then transmitting a weeding instruction to the electromagnetic shovel controller; The camera collects image data; the video algorithm processor processes the data transmitted by the camera and transmits a weeding instruction to the electromagnetic shovel controller to control the electromagnetic shovel to remove weeds; The video algorithm of the video algorithm processor includes the following steps: S1: using the camera on the weeding device to collect different weeds, different weed numbers between ridges, and weed images with different leaf occlusions in the field to increase data diversity and increase data samples under different weather conditions so that the model can adapt to different weather conditions; S2: based on the Pycharm operating platform, using Python scripts to enhance the collected data, enrich the data set, improve the robustness of the detection network model, and at the same time use image dehazing technology based on the Pytorch framework to process the data set images to increase the image contrast ; S3: SURF feature extraction is performed on the samples to establish a data set; S4: Labelimg is used to annotate the data set so that the specific location coordinates of the target weeds can be read during network training. The data set is divided into a validation set, a test set, and a training set; S5: The original CIoU loss function in the YOLOv8 network is replaced with G-IoU, and the IoU threshold is specified to increase the movement speed of the prediction box, which can effectively improve the training speed and inference accuracy; S6: The weight parameters of the three loss functions of the YOLOv8 network are adjusted to achieve higher recognition accuracy; S7: The collected data is passed into the network for weed recognition; The G-IoU function introduces the minimum true box, with a value range of (-1, 1). When the IoU threshold is n, the algorithm is as follows: , G is the minimum bounding box that can be surrounded by n bounding boxes, A is the target identification A bounding box, B is the target identification B bounding box, and so on until n. During the operation, first calculate G, then calculate the intersection of n bounding boxes, then calculate G minus the intersection, calculate the ratio of this value to the area of ​​G, and then use the IOU of all target identification bounding boxes to subtract this ratio, and the result is GIOU; The SURF feature extraction includes: image reading, image display; image grayscale, RGB image to grayscale image; feature point detection: using the Hessian matrix to detect feature points, using the second-order derivative matrix in the x and y directions to measure the local curvature of the function, the determinant value of which represents the change around the pixel point, and the feature point needs to take the extreme value of the determinant value; feature point positioning: locating the feature point by interpolating the neighboring information of the feature point; direction positioning: by calculating the Haar wavelet transform of the pixels around the feature point in the x and y directions, and adding the transformation values ​​in the x and y directions within a certain angle range in the xy plane to form a vector, the longest of all vectors is the direction of the feature point; feature description: after the direction of the feature point is selected, the surrounding pixels are used to establish a descriptor based on this direction; feature matching: extracting feature descriptors of multiple images, using the SURF algorithm for feature matching, and obtaining matching points.

2. The weeding method according to claim 1, wherein The electromagnetic shovel comprises a coil, a shovel box, a shovel rod, an electromagnetic shovel blade, and a return spring.

3. The weeding method according to claim 1, wherein The electromagnetic shovel includes multiple groups of coils and a permanent magnet core.

4. The weeding method according to claim 1, wherein The weeding mechanism is provided with a camera assembly, which is arranged inside the shovel box. The camera assembly includes a camera, a camera fixing bracket, and a shock-absorbing cotton block. The shock-absorbing cotton block is clamped between the camera and the load bracket. A glass window is provided on the side of the shovel box; the camera and the shovel head of the electromagnetic shovel are in the same direction.

5. The weeding method according to claim 1, wherein The electromagnetic shovel includes a U-shaped rubber head.

6. The weeding method according to claim 1, wherein The load bracket is supported from the bottom by a plurality of support studs, the bottoms of the support studs are riveted to the inner bottom of the main body load box, and the support studs are fixed with nuts and washers.

Images