Hilly mountainous self-adaptive orchard mowing operation method

Abstract

The application discloses a hilly and mountainous self-adaptive orchard mowing operation method, which comprises the following specific steps: step one, acquiring the image in front of the mower; step two, extracting the rectangular region in front of the mower as the region of interest and taking the region of interest as the target image; step three, establishing a weed identification model and training the weed identification model, and inputting the target image into the weed identification model; step four, combining the corresponding pixel-level mask with the target image for each type of weed and outputting the corresponding binary mask image; calculating the density grade of each type of weed; step five, establishing a control mapping table related to the weed type, weed density grade, flail working radius and flail target rotating speed. The application combines the intelligent weed identification with the flail system with adjustable flail working radius, improves the mowing adaptability and operation efficiency, and realizes efficient, accurate and energy-saving mowing in a complex orchard environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of weeding equipment and weeding technology, and particularly relates to a self-adaptive orchard mowing operation method for hilly areas. BACKGROUND

[0002] To realize the modernization of orchards in hilly areas, developing small agricultural machinery is one of the key measures. The treatment of weeds in orchards is an important part of daily management, which has traditionally relied on manual or chemical weeding. Not only is the labor intensity great, but the use of chemicals can also cause environmental pollution. Meanwhile, excessive removal of weeds can also exacerbate soil water loss. In contrast, mechanical stubble weeding has obvious advantages. It not only reduces the dependence on manpower, but also leaves grass clippings on the ground after mowing, which can be converted into organic fertilizer after decomposition, helping to improve soil fertility. In addition, retaining a certain height of grass stubble can effectively reduce water evaporation and play a role in protecting the soil of the orchard, which is an important means to achieve green and sustainable orchard management.

[0003] The common weeding machinery cutterhead structures currently mainly include disc type, rolling cutter type, reciprocating type, and flail type, etc. Among them, the disc type cutterhead has large power consumption and unsatisfactory grass cutting effect; the rolling cutter type structure is easy to entangle weeds, affecting the operation efficiency; the reciprocating cutterhead has obvious vibration and poor adaptability. Especially in the mountain orchard environment, the terrain is changeable, and higher requirements are put forward for the adaptability of the mowing equipment. At the same time, the types of weeds in the orchard are various, especially some climbing plants are easy to entangle the cutter, causing the equipment to be stuck or even damaged. Among these structures, the flail type cutterhead has better running stability, stronger anti-winding ability, and wider adaptability, and is more suitable for mowing operation in hilly orchards than other types, which is a relatively ideal choice for orchard weeding mechanism.

[0004] Traditional flail-type mowers have low intelligence, and the flail structure is relatively fixed. However, the types of weeds in the orchard are various, and the heights of the weeds are different. Different types of weeds have different requirements for the mowing performance of the mower. The flail system is difficult to adapt to different orchard environments. Specifically, the speed of the flail shaft and the working radius of the flail are important indicators that determine the cutting force. The working radius and rotational speed of the current flail-type mower are fixed values, which are difficult to adapt to the complex and variable weed environment in hilly orchards. The same and single mowing structure parameters are used for both easy-to-mow and difficult-to-mow weeds. Moreover, in order to meet the mowing effect of weeds in hilly orchards, the operating power (rotational speed) of the flail system on the mower is set to be relatively large (rotational speed redundancy). However, this working method causes unnecessary energy waste because the rotational speed of the flail system on the mower is not based on the actual weed environment, such as in an environment with a small weed density.

[0005] In view of the above reasons, it is urgent to provide a kind of can automatically detect weed environment and can adapt to the rotation speed of flail and flail working radius, to improve the adaptability of flail mower in different weed environment in hilly orchard, realize the precision of hilly orchard weed mowing, effectively save the energy consumption of flail system of mower. SUMMARY

[0006] The purpose of the present application is to solve the problems in the prior art, that is, to provide a hilly orchard self-adaptive mowing method.

[0007] The purpose of the present application is achieved by the following technical scheme: a hilly orchard self-adaptive mowing method, comprising the following specific steps:

[0008] Step one, get the image in front of the mower:

[0009] Step two, extract the rectangular area in front of the mower as the region of interest in the obtained image, and take the region of interest as the target image;

[0010] Step three, establish a weed identification model and train the weed identification model, input the target image into the weed identification model; infer the class label, boundary box coordinates and pixel-level mask corresponding to each detection instance in the target image;

[0011] Step four, combine the corresponding pixel-level mask with the target image for each type of weed and output the corresponding binary mask image; calculate the actual coverage area of each type of weed by counting the number of non-zero pixel points in each binary mask image, and calculate the density grade of each type of weed;

[0012] Step five, establish a control mapping table associating weed species, weed density grade, flail working radius and flail target speed; set the flail working radius of the flail system on the mower according to the weed depth, and set the flail target speed according to the control mapping table.

[0013] As a preferred, the specific method of step two is as follows:

[0014] A homography transformation model is established between the ground plane and the image pixel coordinates; select the rectangular ground area in front of the mower as the original image, project the four corner points of the original image to the image plane using the camera intrinsic matrix and extrinsic matrix, and obtain the corresponding pixel coordinates; based on the four corner point pixel coordinates on the original image and the corresponding points of the target normalized rectangle, calculate the perspective transformation matrix; then, the perspective correction is carried out on the original image through the perspective transformation matrix, and the rectangular ground area is mapped to the target image under the orthographic view.

[0015] As preferred, the specific method of step three is as follows:

[0016] In the hilly orchard scene, high-resolution cameras are used to collect ground images of different growth seasons, different light conditions, different ground backgrounds, and cover target weed species. Each type of weed collects no less than 500 valid images, thereby obtaining a weed instance segmentation dataset for the hilly orchard scene; each independent weed individual in the collected images is pixel-by-pixel labeled using a labeling tool;

[0017] The weed recognition model selects YOLO-Seg as the basic model architecture, and trains the model through the weed instance segmentation dataset; after the weed recognition model predicts the class label, the boundary box coordinates of each detection instance in the target image, and the pixel-level mask corresponding to each detection instance, it removes redundant predictions through non-maximum suppression and retains the optimal result.

[0018] As preferred, in step four, the calculation method of the actual coverage area and the density level of each type of weed is as follows:

[0019] The number of non-zero pixel points in the binary mask image is counted , combined with the pixel-physical area conversion coefficient calculated the actual coverage area of the current type of weed The calculation formula of the actual coverage area of the current type of weed

[0020] ;

[0021] And calculate the proportion of the current type of weed in the total weeds ;

[0022] If is less than 30%, the density level of the current type of weed is "low";

[0023] If is between 30% and 60%, the density level of the current type of weed is "medium";

[0024] If is higher than 60%, the density level of the current type of weed is "high".

[0025] As preferred, in step five, the target weed species includes white clover, wormwood, dogtail grass, bittersweet, and hop.

[0026] As preferred, the flail system is provided with a first radius gear and a second radius gear. When the flail system is in the first radius gear, the flail working radius is 130 mm; when the flail system is in the first radius gear, the flail working radius is 150 mm.​

[0027] As preferred, the flail system comprises a cover body, a rotating drive motor, an adjusting gear, an incomplete gear, a flail shaft, the cover body is provided with locking straight groove wheels on both sides, the locking straight groove wheels are rotationally connected with the cover body, and the locking straight groove wheels are provided with straight grooves; the adjusting gear is coaxially arranged with the locking straight groove wheels, and the adjusting gear is provided with an arc-shaped groove; the flail shaft is arranged between the two locking straight groove wheels, and the two ends of the flail shaft pass through the straight grooves and the arc-shaped groove at the same time, and the flail shaft is connected with a flail; the incomplete gear is matched with the adjusting gear, and the incomplete gear is connected with the adjusting motor; at least one locking straight groove wheel is connected with a rotating drive mechanism, and the rotating drive mechanism drives the locking straight groove wheel to rotate; the locking straight groove wheel is provided with a flail shaft retaining mechanism corresponding to the straight groove, and the flail shaft retaining mechanism comprises two flail shaft retaining frames arranged along the straight groove in sequence; in a non-adjusting state of the flail system, the flail shaft is located in one of the flail shaft retaining frames, and the flail shaft retaining frame clamps the flail shaft to keep the relative position between the flail shaft and the locking straight groove wheel stable.

[0028] As preferred, the flail shaft retaining frame comprises two oppositely arranged arc-shaped retaining pieces, and a clamping space for clamping the flail shaft is formed between the two arc-shaped retaining pieces; the arc-shaped retaining piece is connected with the locking straight groove wheel through an elastic piece.

[0029] As preferred, the locking straight groove wheel is provided with a rotating shaft, and the rotating shaft is rotationally matched with the cover body; the rotating drive mechanism comprises a rotating drive motor, the rotating drive motor is provided with a first belt pulley, the rotating shaft of one of the locking straight groove wheels is provided with a second belt pulley, and a belt is connected between the first belt pulley and the second belt pulley.

[0030] As preferred, the flail is in a "Y" shape.

[0031] The present application has the advantages that the present application combines intelligent weed identification with a flail system with adjustable flail working radius, realizes real-time sensing of the types and density of weeds in orchards on hilly terrain through intelligent weed identification means, and provides a basis for self-adaptive adjustment of the flail system, identifies and counts the types of weeds through image acquisition by a visible light camera and in combination with a deep learning target detection algorithm, and realizes real-time calculation of the distribution density and proportion of various types of weeds to provide data support for control decisions; based on the identification result, the flail system can adjust the flail working radius and realize dynamic adjustment of the target rotating speed of the flail, form an intelligent operation mode driven by data, effectively improve the mowing adaptability and operation efficiency, reduce energy consumption, reduce manual intervention, greatly improve the level of automatic operation, and realize efficient, accurate and energy-saving mowing in complex orchard environments. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1Structure diagram of the flail system in one direction.

[0033] Figure 2 Structure diagram of the flail system in another direction.

[0034] Figure 3 Structure diagram of the flail system.

[0035] Figure 4 Structure diagram of the locking straight groove wheel and the incomplete gear.

[0036] Figure 5 Structure diagram of the adjusting gear.

[0037] Figure 6 Structure diagram of the flail system when switching the working radius of the flail.

[0038] Figure 7 Work flow diagram of the mowing method.

[0039] Figure 8 Diagram of extracting the target image in the obtained image.

[0040] Figure 9 Framework diagram of the YOLO-Seg model.

[0041] Figure 10 Diagram of combining the pixel-level mask with the target image and outputting the binary mask image.

[0042] In the figure: 1, cover body, 2, rotary drive motor, 3, first pulley, 4, second pulley, 5, belt, 6, locking straight groove wheel, 6-1, rotating shaft, 6-2, straight groove, 6-3, arc-shaped retaining member, 6-4, elastic member, 7, adjusting gear, 7-1, arc-shaped groove, 7-2, shaft hole, 8, incomplete gear, 9, flail shaft, 10, flail, 11, camera, 12, adjusting motor. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art belong to the scope of protection of the present application.

[0044] Those skilled in the art should understand that in the disclosure of the present application, the terms "longitudinal", "transverse", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship shown in the drawings, which are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore the above terms cannot be understood as a limitation of the present application.

[0045] It can be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of one element can be one, and in another embodiment, the number of the element can be multiple, and the term "one" cannot be understood as a limitation on the number.

[0046] As shown in Figure 7 A hilly mountainous self-adaptive orchard mowing operation method, comprising the following specific steps:

[0047] Step one, acquire the image in front of the mower.

[0048] In this step, the image is acquired by the high-resolution camera 11 of 2048x1088 pixels and 50fps frame rate in front of the mower. The camera 11 is installed on the top of the cover body 1 of the flail system, the camera 11 is equipped with a ring-shaped LED light supplement system with color temperature 5500K and illumination adjustable range 1000-10000lux, the visible light camera 11 is installed on the central axis in front of the mower, the vertical height from the ground is H=80±5cm, the lens optical axis forms a 15±2° depression angle with the horizontal plane, and it is ensured that the imaging plane of the camera 11 is perpendicular to the forward direction of the mower. The installation posture is fixed by a mechanical clamp to ensure long-term operation stability, and stable imaging under different illumination conditions can be realized.

[0049] Step two, extract the rectangular region in front of the mower as the region of interest (ROI) in the acquired image, and take the region of interest as the target image. As shown in Figure 8 Figure 8 A schematic diagram for extracting the target image in the acquired image.

[0050] The specific method of this step is as follows:

[0051] First, a homography transformation model between the ground plane and the image pixel coordinates is established; a rectangular ground area in front of the mower is selected as the original image,

[0052] The rectangular ground area in front of the mower is defined in the real space as:

[0053] ​Physical region: x ∈ [0.5m, 2.0m], y ∈ [ 0.75m, 0.75m];

[0054] This region is a rectangular area with a length of 0.5m to 2.0m from the front edge of the flail system and a transverse range of ±0.75m, corresponding to a mowing width of about 1.5m, covering the work path in the next 1-3 seconds.

[0055] Using the camera intrinsic matrix K and the extrinsic matrix (rotation R, translation T), the four corner points (corner coordinates Pi ( x i , y i ,0) of the original image are projected onto the image plane, and their corresponding pixel coordinates are obtained, which are Ti=(u i ,v i ). The projection formula is:

[0056] ;

[0057] Based on the four corner pixel coordinates on the original image and the corresponding points of the target normalized rectangle, the perspective transformation matrix is calculated.

[0058] Where the camera intrinsic matrix K includes the focal length and the principal point , which are obtained in advance by the chessboard calibration method. The cv2.getPerspectiveTransform() function in the OpenCV software is used to calculate the perspective transformation matrix M based on the four corner pixel coordinates (u i ,v i ) on the original image and the corresponding points of the target normalized rectangle. Then, the perspective correction is performed on the original image through the perspective transformation matrix, mapping the rectangular ground area shot obliquely to the target image in the orthographic view.

[0059] Where, OpenCV is an open-source computer vision and machine learning software library, widely used in image processing and computer vision tasks. It supports multiple programming languages, including C++, Python, Java, etc., and can run on multiple platforms such as Windows, Linux, etc.

[0060] The cv2.getPerspectiveTransform() function is a function in OpenCV for calculating the perspective transformation matrix, which can convert an image from one perspective to another (e.g. correct the oblique plane to the orthographic view).

[0061] The transformed target image only retains a target size region as an input of a subsequent weed recognition model. The region of interest accurately corresponds to the ground to be processed in front of the mower, avoiding interference of irrelevant backgrounds such as distant vegetation, sky, and tree canopy on the recognition algorithm, and reducing the input resolution of the model to improve the inference speed.

[0062] Step three, establishing a weed recognition model and training the weed recognition model, inputting the target image into the weed recognition model; inferring the class label, boundary box coordinate and pixel-level mask corresponding to each detection instance in the target image.

[0063] The method for establishing the weed recognition model is as follows:

[0064] In the hilly orchard scene, high-resolution cameras are used to collect ground images under different growth seasons, different light conditions, different ground backgrounds and covering target weed species. No less than 500 effective images are collected for each type of weed to ensure that the morphological changes from the seedling stage to the mature stage are covered. Thus, a weed instance segmentation dataset for the hilly orchard scene is obtained. Then, each independent weed individual in the collected images is annotated pixel by pixel using a labeling tool. In this process, LabelMe labeling tool can be used to annotate each independent weed individual in each image pixel by pixel. In the labeling process, adjacent or partially overlapping plants are strictly distinguished to avoid instance adhesion and ensure that the mask boundary closely follows the real outline of the weed.

[0065] In this application, the target weed species include white clover, wormwood, dogtail grass, bittersweet, and hop.

[0066] The weed recognition model selects YOLO-Seg as the basic model architecture and is trained through the weed instance segmentation dataset. After training, the model is exported in ONNX format and is quantized and accelerated using TensorRT on the NVIDIA Jetson AGX Orin edge computing platform carried by the mower. Through INT8 precision quantization and layer fusion optimization, the model inference delay is reduced from 110ms under the original FP32 to within 68ms, meeting the real-time requirement.

[0067] After the weed recognition model infers the class label, boundary box coordinate and pixel-level mask corresponding to each detection instance in the target image, non-maximum suppression is used to remove redundant predictions and retain the optimal result.

[0068] The above method realizes high-precision and real-time instance segmentation recognition of multiple weeds in the orchard, solves the problem that traditional target detection cannot accurately obtain the shape boundary of weeds, provides a fine perception basis for adaptive speed and radius control of the flail system, and significantly improves the pertinence and energy efficiency level of intelligent mowing operation.

[0069] wherein, LabelMe is an open-source image annotation tool, mainly used for data annotation in computer vision tasks, especially suitable for tasks such as target detection, semantic segmentation, instance segmentation, etc.

[0070] YOLO-Seg is an instance segmentation model based on YOLO architecture, which combines target detection and instance segmentation functions, and can output the bounding box and accurate pixel-level mask of the target at the same time, realizing efficient real-time instance segmentation task.

[0071] As shown in Figure 9 , the model is composed of a backbone feature extraction network StarNet, a feature pyramid structure and a decoupled segmentation head. The input image is uniformly scaled to 640x640 pixels, and data enhancement strategies such as Mosaic, random affine transformation, color jittering, MixUp are used to improve the generalization ability of the model. The loss function is composed of three parts: classification loss, detection loss and segmentation loss, which together optimize the performance of the model.

[0072] The model optimizer uses AdamW, the initial learning rate is 1x10 -3 , the weight decay is 5x10 -4 , the batch size is set to 16, and the training period is 300 epochs. During the training process, the learning rate cosine annealing strategy is used, and the early stopping mechanism is combined to prevent overfitting. The final model achieves mAP@0.5=91.7% on the independent test set, and the average mask IoU is 86.3%, and it still has good segmentation robustness for complex shape and edge blurred weeds such as Humulus scandens vine.

[0073] Step four, for each type of weed, combine the corresponding pixel-level mask with the target image and output the corresponding binary mask image; calculate the actual coverage area of each type of weed by counting the number of non-zero pixel points in each binary mask image, and calculate the density grade of each type of weed. As shown in Figure 10 , the figure shows the combination of pixel-level mask and target image and the output of binary mask image. Figure 10

[0074] wherein, the calculation method of the actual coverage area and the density grade of each type of weed is as follows:

[0075] Count the number of non-zero pixel points in the binary mask image , combine the pre-calibrated pixel-physical area conversion coefficient , calculate the actual coverage area of the current type of weed , the calculation formula of the actual coverage area of the current type of weed

[0076] ​​ ;

[0077] and calculate the proportion of the current type of weed in the total weeds ; the proportion The calculation formula is as follows:

[0078] .

[0079] If is less than 30%, the density level of the current type of weed is "low";

[0080] If is between 30% and 60%, the density level of the current type of weed is "medium";

[0081] If is higher than 60%, the density level of the current type of weed is "high".

[0082] Step five, establish a control mapping table that associates weed type, weed density level, flail working radius, and flail target speed; set the flail working radius of the flail system on the mower according to the weed depth, and set the flail target speed according to the control mapping table.

[0083] Wherein, the flail working radius is automatically set according to the expected weed depth or growth before mowing, and the system automatically matches the corresponding flail speed according to the set flail working radius value. Higher speed is required in deep grass area to maintain the end line speed and cutting kinetic energy. The final target speed (flail speed) is determined by the weed density level and the flail working radius:

[0084] ;

[0085] is the density level; density_level represents the type of weed; wherein the function f() is a decision logic based on querying the control mapping table, which determines the basic speed of the flail based on the input weed type and density level and by querying the control mapping table. is the radius gain factor, different flail working radii R correspond to different gain factors. According to the difficulty of cutting and the density level of the weed, the system statically sets the flail working radius before mowing starts.

[0086] Wherein, the flail system is provided with a first radius gear and a second radius gear, when the flail system is in the first radius gear, the flail working radius is 130mm; when the flail system is in the first radius gear, the flail working radius is 150mm. The flail working radius can be switched between the first radius gear and the second radius gear.

[0087] After the working radius of the flail is set, the flail is started and rotated, and the system dynamically adjusts the flail rotating speed in real time according to the current weed type and density, to ensure the balance between cutting efficiency and energy consumption.

[0088] The following is the ranking of the cutting difficulty of several weeds:

[0089] White clover < mugwort < foxtail < poke < goosegrass (vines).

[0090] The control mapping table in the application is shown in Table 1:

[0091] Table 1:

[0092]

[0093] It is worth mentioning that when multiple weeds are detected in the working area at the same time, the corresponding target rotating speed is selected according to each weed type and weed density, and the highest target rotating speed is finally used as the actual operating rotating speed of the flail system.

[0094] The intelligent mowing control logic of the application is as follows: the rotating speed is automatically reduced or the machine is stopped in the weed-free area, the economic rotating speed is matched in the low-density area, and the power output is only increased in the high-density or difficult-to-cut weed area, to realize on-demand energy supply and significantly reduce the energy consumption of the whole machine.

[0095] As shown in Figures 1-6 The flail system includes a cover body 1, a rotating drive motor 2, an adjusting gear 7, an incomplete gear 8, and a flail shaft 9. Locking straight grooved wheels 6 are arranged on both sides of the cover body 1 and are rotatably connected between the cover body 1 and the locking straight grooved wheels 6. Straight grooves 6-2 are arranged on the locking straight grooved wheels 6. The adjusting gear 7 is coaxially arranged with the locking straight grooved wheels 6, and arc-shaped grooves 7-1 are arranged on the adjusting gear 7. The flail shaft 9 is arranged between the two locking straight grooved wheels 6, and the two ends of the flail shaft 9 pass through the straight grooves 6-2 and the arc-shaped grooves 7-1 at the same time. The flail shaft 9 is connected with a flail 10. The incomplete gear 8 is engaged with the adjusting gear 7, and the incomplete gear 8 is connected with an adjusting motor 12. At least one locking straight grooved wheel 6 is connected with a rotating drive mechanism to drive the locking straight grooved wheel 6 to rotate. A flail shaft 9 retaining mechanism corresponding to the straight grooves 6-2 is arranged on the locking straight grooved wheel 6. The flail shaft 9 retaining mechanism includes two flail shaft 9 retaining frames arranged along the straight grooves 6-2 in sequence in the axial direction. In the non-adjusting state of the flail system, the flail shaft 9 is located in one of the flail shaft 9 retaining frames, and the flail shaft 9 retaining frame clamps the flail shaft 9 to keep the relative position between the flail shaft 9 and the locking straight grooved wheel 6 stable.

[0096] The flail shaft 9 holder comprises two oppositely arranged arc-shaped holders 6-3, and a clamping space for clamping the flail shaft 9 is formed between the two arc-shaped holders 6-3; the arc-shaped holder 6-3 is connected with the locking straight groove wheel 6 through the elastic element 6-4.

[0097] The locking straight groove wheel 6 is provided with a rotating shaft 6-1, and the rotating shaft 6-1 is rotationally connected with the cover body 1; the rotating driving mechanism comprises a rotating driving motor 2, and the rotating driving motor 2 is provided with a first belt pulley 3; the rotating shaft 6-1 of one of the locking straight groove wheels 6 is provided with a second belt pulley 4, and the first belt pulley 3 and the second belt pulley 4 are connected with a belt 5. The adjusting gear 7 is provided with a shaft hole 7-2 for the rotating shaft 6-1 to pass through.

[0098] The flail 10 in the application is in the shape of "Y". One end of the flail 10 is hinged with the flail shaft 9. The incomplete gear 8 is a sector gear with a sector angle of 40° to 45°.

[0099] It is worth mentioning that the working radius of the flail in the application refers to the distance between the flail shaft 9 and the rotating central axis of the flail 10.

[0100] The flail 10 system is arranged at the front end of the mower. When the mowing operation is performed, the flail 10 system takes the rotating driving motor 2 as a power source, and transmits power to the rotating shaft 6-1 of the locking straight groove wheel 6 through the transmission mechanism composed of the first belt pulley 3, the belt 5 and the second belt pulley 4, so as to drive the locking straight groove wheel 6 to rotate. Since the flail shaft 9 passes through the straight groove 6-2 of the locking straight groove wheel 6 at both ends and is stably clamped by the flail shaft 9 holder in the non-adjusting state, the rotation of the locking straight groove wheel 6 drives the flail shaft 9 to rotate synchronously, thereby realizing the rotating movement of the flail 10. During the rotating process of the flail shaft 9 and the flail 10, the relative position between the flail shaft 9 and the locking straight groove wheel 6 remains unchanged under the stable clamping action of the flail shaft 9 holder, so that the working radius of the flail is stably kept in the current state. When the flail 10 system is in the rotating operation state, the incomplete gear 8 and the adjusting gear 7 are in the disengaged state.

[0101] In the non-adjusting state, the flail shaft 9 is located in the clamping space of one of the flail shaft 9 holders. The holder is composed of two oppositely arranged arc-shaped holders 6-3, and the arc-shaped holder 6-3 is connected with the locking straight groove wheel 6 through the elastic element 6-4; the elastic force of the elastic element 6-4 tightly clamps the flail shaft 9 between the two arc-shaped holders 6-3, so as to ensure the stability of the relative position between the flail shaft 9 and the locking straight groove wheel 6, avoid the deviation of the flail 10 due to vibration or external force during the rotating operation, and ensure the stability of the working radius of the flail.

[0102] When the working radius of the flail 10 needs to be adjusted, the flail 10 system is first in a stop state, the motor 12 drives the incomplete gear 8 to rotate, and the incomplete gear 8 is first engaged with the adjusting gear 7 and drives the adjusting gear 7 to rotate by a certain angle, and then is disengaged from the adjusting gear 7; in the process of engaging with the adjusting gear 7, the adjusting gear 7 rotates by a certain angle, and under the action of the arc-shaped groove 7-1, the two ends of the flail shaft 9 move along the axial direction of the straight groove 6-2 of the locking straight groove wheel 6, the flail shaft 9 is separated from one flail shaft 9 holder and enters another flail shaft 9 holder in the moving process, and the distance between the flail shaft 9 and the rotation central axis of the flail 10 changes in the moving process of the flail shaft 9, so that the working radius of the flail is adjusted.

[0103] When the flail shaft 9 is clamped in one of the flail shaft 9 holders, the distance between the flail shaft 9 and the rotation central axis of the flail 10 is 130mm, that is, the working radius of the flail is 130mm.

[0104] The flail 10 system of the application comprises a flail shaft 9 with a variable working radius, a rotation driving device and the like, compared with the flail 10 device on the existing mower, the flail shaft 9 is designed as three long shaft structures from the traditional cylindrical structure, and the locking straight groove wheel 6 and the adjusting gear 7 are arranged at the two ends of the three flail shafts 9, which are used to adjust the distance between the flail shaft 9 and the rotation central axis of the flail 10, and then realize the adjustment of the working radius of the flail. The structure realizes the rotation angle control of the adjusting gear 7 by adjusting the rotation angle of the motor 12, and then realizes the expansion and contraction of the rotation radius of the flail 10. According to the mathematical relationship between the cutting force and the rotation radius:

[0105] ;

[0106] Wherein F is the cutting force of the flail, m is the mass of the flail, ω is the angular velocity of the flail, and d is the cutting depth. It can be seen from the mathematical relationship that the cutting force of the flail is proportional to the radius, that is, the larger the rotation radius of the flail, the greater the cutting force. Therefore, the working radius of the flail can be effectively increased by the position of the flail shaft, so as to increase the cutting force of the flail, and then realize the differentiated weed management of different weeds, and further realize the saving of the power consumption of the mower.

[0107] The present application combines weed intelligent recognition with a flail system with adjustable flail working radius, realizes real-time sensing of the types and density of weeds in orchards in hilly terrain through weed intelligent recognition means, and provides a basis for adaptive adjustment of the flail system, identifies and counts the types of weeds through visible light camera image acquisition combined with a deep learning target detection algorithm, and calculates the distribution density and proportion of each type of weed in real time to provide data support for control decisions; based on the recognition results, the flail system can adjust the flail working radius and realize dynamic adjustment of the flail target rotating speed, form an intelligent operation mode driven by data, effectively improve the mowing adaptability and operation efficiency, reduce energy consumption, reduce manual intervention, greatly improve the automation operation level, and realize efficient, accurate and energy-saving mowing in a complex orchard environment.

[0108] The present application is not limited to the above-mentioned best mode, and anyone can derive other various forms of products under the inspiration of the present application, but regardless of any changes in shape or structure, any technical solution with the same or similar technical solution as the present application falls within the protection scope of the present application.

Claims

1. A hilly mountainous self-adaptive orchard mowing operation method, characterized in that, The method comprises the following specific steps: Step one, obtaining the image in front of the mower: Step two, extracting the rectangular area in front of the mower as the region of interest and taking the region of interest as the target image; Step three, establishing a weed recognition model and training the weed recognition model, inputting the target image into the weed recognition model, and inferring the class label, boundary box coordinate and pixel-level mask corresponding to each detection instance in the target image; Step four, combining the corresponding pixel-level mask with the target image for each type of weed and outputting the corresponding binary mask image; The actual coverage area of each type of weed is calculated by counting the number of non-zero pixel points in each binary mask image, and the density level of each type of weed is calculated. Step five, establishing a control mapping table related to the weed type, weed density level, flail working radius and flail target speed; setting the flail working radius of the flail system on the mower according to the weed depth, and setting the flail target speed according to the control mapping table; the flail system comprises a cover body, a rotating drive motor, an adjusting gear, an incomplete gear, a flail shaft, two lock straight groove wheels are arranged on the two sides of the cover body and are rotationally connected between the cover body and the lock straight groove wheel, and a straight groove is arranged on the lock straight groove wheel; the adjusting gear is coaxially arranged with the lock straight groove wheel, and an arc-shaped groove is arranged on the adjusting gear; the flail shaft is arranged between the two lock straight groove wheels, and the two ends of the flail shaft pass through the straight groove and the arc-shaped groove at the same time, and a flail is connected to the flail shaft; the incomplete gear is matched with the adjusting gear, and the incomplete gear is connected with the adjusting motor; at least one lock straight groove wheel is connected with a rotating drive mechanism, and the rotating drive mechanism drives the lock straight groove wheel to rotate; a flail shaft retaining mechanism corresponding to the straight groove is arranged on the lock straight groove wheel, and the flail shaft retaining mechanism comprises two flail shaft retaining frames arranged in sequence along the straight groove axis direction; in the non-adjusting state of the flail system, the flail shaft is located in one of the flail shaft retaining frames, and the flail shaft retaining frame clamps the flail shaft to keep the relative position between the flail shaft and the lock straight groove wheel stable.

2. The hilly mountainous self-adaptive orchard mowing operation method according to claim 1, characterized in that, The specific method of step two is as follows: A homography transformation model between the ground plane and the image pixel coordinates is established; a rectangular ground area in front of the mower is selected as the original image, and four corner points of the original image are projected onto the image plane by using the camera intrinsic matrix and the extrinsic matrix to obtain the corresponding pixel coordinates; based on the four corner point pixel coordinates on the original image and the corresponding points of the target normalized rectangle, a perspective transformation matrix is calculated; then, the perspective transformation matrix is used for perspective correction of the original image, and the rectangular ground area shot obliquely is mapped into the target image in the orthographic view.

3. The hilly mountainous self-adaptive orchard mowing operation method according to claim 1, characterized in that, The specific method of step three is as follows: In the hilly orchard scene, high-resolution cameras are used to collect ground images covering target weed species under different growth seasons, different light conditions, different ground backgrounds, and at least 500 effective images for each type of weed, so as to obtain a weed instance segmentation dataset for the hilly orchard scene; each independent weed individual in the collected image is annotated pixel by pixel using a labeling tool; The weed identification model selects YOLO-Seg as a basic model architecture, and is trained through a weed instance segmentation dataset; after the weed identification model predicts the class label, the boundary box coordinate and the pixel-level mask corresponding to each detection instance in the target image, the non-maximum suppression is used to remove redundant prediction and retain the optimal result.

4. The hilly mountainous self-adaptive orchard mowing operation method according to claim 1, characterized in that, In step four, the calculation method of the actual coverage area and the density grade of each type of weed is as follows: Counting the number of non-zero pixel points in a statistical binary mask image , combining the pixel-physical area conversion coefficient calibrated in advance , calculating the actual coverage area of weeds of the current type , the actual coverage area of weeds of the current type The calculation formula is as follows: ； and calculates the proportion of the current type of weed in the total weeds ; If a density level of the current species of weed is "low" if the percentage is below 30%; If between 30% and 60%, the density level of the current species of weeds is "medium"; If If the density of the current species of weed is higher than 60%, the density level of the current species of weed is "high".

5. The hilly mountainous self-adaptive orchard mowing operation method according to claim 3, characterized in that, In step five, the target weed types include white axle grass, wormwood, dogtail grass, bittersweet and hop grass.

6. The hilly mountainous self-adaptive orchard mowing operation method according to claim 1, characterized in that, The flail system is provided with a first radius gear and a second radius gear, when the flail system is in the first radius gear, the flail working radius is 130mm; when the flail system is in the first radius gear, the flail working radius is 150mm.

7. The hilly mountainous self-adaptive orchard mowing operation method according to claim 1, characterized in that, The flail shaft holder includes two oppositely arranged arc-shaped holders, and a clamping space for clamping the flail shaft is formed between the two arc-shaped holders; the arc-shaped holder is connected with the locking straight groove wheel through an elastic member. 8.The hilly mountainous self-adaptive orchard mowing operation method according to claim 1, characterized in that, The locking straight groove wheel is provided with a rotating shaft, and the rotating shaft is rotatably connected with the cover body; the rotating driving mechanism includes a rotating driving motor, the rotating driving motor is provided with a first belt pulley, the rotating shaft of one of the locking straight groove wheels is provided with a second belt pulley, and the first belt pulley and the second belt pulley are connected with a belt. 9.The hilly mountainous self-adaptive orchard mowing operation method according to claim 1, characterized in that, The flail is in the shape of "Y".

Images