Automatic and rapid picking method of nut green husk fruit in hilly area

Abstract

This invention provides an automated and rapid harvesting method for green-skinned nuts in hilly areas, belonging to the technical field of green-skinned nut harvesting. The method includes the following steps: setting up a harvesting trolley, driving the trolley to the front of the nut tree, and then detecting whether the trolley's bottom plate is level. If it is not level, it is adjusted to a level position using a lifting device; using a laser radar to initially image and locate the green-skinned nuts; constructing a three-dimensional spatial model for harvesting green-skinned nuts; calculating the harvesting order of all nuts based on the three-dimensional spatial model, and then controlling a robotic arm to harvest them one by one; after harvesting the nuts on one side, the trolley moves to the next side of the nut tree until the entire tree is harvested. Each nut can be harvested from all four directions (north, south, east, and west). This invention fully automates the harvesting process, achieves high harvesting efficiency, has a very low probability of missed harvests, and realizes precise and rapid harvesting.

Description

Technical Field

[0001] This invention relates to the field of nut and green-skinned fruit harvesting technology, and in particular to an automatic and rapid harvesting method for nut and green-skinned fruit in hilly areas. Background Technology

[0002] Macadamia nut harvesting generally involves manual harvesting and mechanical vibration harvesting. Manual harvesting involves using a hook to peel off the involucre one by one when the fruit is ripe. When using mechanical harvesting, the trees are sprayed with ethephon two weeks before harvest to promote ripening, and then the branches are vibrated mechanically to cause the ripe fruit to fall. However, the use of ethephon can cause premature leaf drop, affecting tree growth and the following year's yield. Currently, in China, mechanical vibration harvesting or robotic arms are commonly used for assisted harvesting. However, robotic arms lack the ability to automatically locate and identify unripe macadamia nuts, resulting in low harvesting efficiency, slow speed, and poor results.

[0003] Because the color of unripe nuts is very similar to that of tree leaves, image recognition methods struggle to identify them. The similarity in color further complicates identification, making it difficult to pinpoint the exact location of unripe nuts and causing significant challenges and inefficiency when using robotic arms for harvesting. Therefore, a more precise identification method is needed to enable more efficient control of the robotic arm to select which nuts to harvest. This paper proposes an automated and rapid harvesting method for unripe nuts in hilly areas. Summary of the Invention

[0004] The purpose of this invention is to provide an automatic and rapid harvesting method for green-skinned nuts in hilly areas, solving the technical problems of low harvesting efficiency and easy omissions in existing green-skinned nut harvesting methods.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] An automated and rapid harvesting method for green-skinned nuts in hilly areas, the method comprising the following steps:

[0007] Step 1: Install a balance detection device on the bottom plate of the tracked trolley, and at the same time install a lifting device at the bottom of the bottom plate of the trolley;

[0008] Step 2: Install the lidar detection device, robotic arm harvesting device, and camera on the chassis of the vehicle.

[0009] Step 3: Drive the picking cart to the front of the nut tree, and then check whether the bottom of the cart is level. If it is not level, adjust it to a level position using the lifting device.

[0010] Step 4: Use lidar to perform initial imaging and localization of the green-skinned nuts;

[0011] Step 5: Construct a three-dimensional spatial model for harvesting green-skinned nuts;

[0012] Step 6: Calculate the picking order of all nuts based on the three-dimensional spatial model of the harvesting, and then control the robotic arm to pick them one by one;

[0013] Step 7: After picking the nuts on one side, move the cart to the side of the nut tree and repeat steps 3-6 until the entire nut tree is picked. Each nut can be picked from the four directions of east, west, south, and north.

[0014] Furthermore, in step 1, the tracked vehicle can automatically climb hilly areas. Before harvesting nuts, a balance detection device is used to check the level of the vehicle's bottom plate. The lifting device is set with four adjustable screw rotating rods. The control device is connected to the balance detection device and raises or lowers one or two screw rotating rods according to the horizontal deviation.

[0015] Furthermore, in step 2, the lidar detection device is located at the front end of the trolley. The lidar detection device includes a lidar and a longitudinal and lateral movement device. The longitudinal and lateral movement device includes a lateral slide bar and a longitudinal slide bar, which are arranged in a cross shape. The lidar is equipped with a moving wheel device, which can switch between the lateral slide bar and the longitudinal slide bar. At the same time, the moving wheel device is equipped with a timed movement control device, which is used to control the moving speed, moving position and dwell time of the moving wheel device. The timed movement control device is connected to the lidar and the external harvesting machine respectively, and is used to receive movement control signals and notify the lidar to scan.

[0016] Furthermore, the specific process of step 4 is as follows:

[0017] Step 4.1: The lidar is installed on the harvester and it comes to the front of the nut tree. The lidar acquires an initial three-dimensional image of the nut area for the first time.

[0018] Step 4.2: Define the initial 3D image and then label several feature points as recognition points;

[0019] Step 4.3: Construct a three-dimensional coordinate system for the defined area, set the recognition center point, and then label the coordinates of all points on the nut tree;

[0020] Step 4.4: The lidar moves along the longitudinal and lateral moving device, pausing after each point. The lidar scans to obtain a stereo image in one direction, and then all the three-dimensional images are acquired and summarized to obtain a set of three-dimensional stereo images of the nut tree in different directions.

[0021] Step 4.5: Fuse all the 3D images to obtain the 3D coordinate dataset of the nut tree in different directions;

[0022] Step 4.6: Then find the positions of all nut shapes from the nut tree 3D coordinate dataset and output the 3D position points of the nuts to achieve imaging and localization of the nuts.

[0023] Furthermore, the specific process of step 5 is as follows:

[0024] Step 5.1: Construct a three-dimensional spatial location model of the green-skinned nut based on its location;

[0025] Step 5.2: Install a distance sensor on the harvesting robot arm to measure the distance between the robot arm and the object in front in real time;

[0026] Step 5.3: Construct a three-dimensional spatial model for harvesting based on the fixed position of the harvesting robot, the position of the nut collection box, and the three-dimensional spatial position model of the green nut skin.

[0027] Step 5.4: Use a camera to observe in real time the status of the robotic arm, the status of the nut collection box, and the data on the harvesting environment in the 3D model of the harvesting space;

[0028] Step 5.5: Calculate the real-time location of the next nut to be picked based on the status and position of the robotic arm, so that the robotic arm can pick it.

[0029] Further, the specific process of step 6 is as follows: Since the position coordinates of all nuts are known in the three-dimensional spatial model of harvesting, and the position coordinates of the nut collection box are also known, the distances of all nuts to the nut collection box can be calculated. Each nut is labeled, and all distances are stored in an array. The array data is then sorted from smallest to largest, and the output array sorting is the harvesting order. Nuts are located according to the output distance table, and the path of the robotic arm is planned in advance.

[0030] Further, the specific process of step 4.1 is as follows: When the lidar acquires the initial three-dimensional image, the harvester uses the lidar to identify the height and then determines the height of the nut tree. The harvester aligns the center of the longitudinal and lateral moving device with the middle position of the nut tree. The lidar stops at the intersection of the lateral and longitudinal sliding rods of the longitudinal and lateral moving device. The lidar begins to scan the initial three-dimensional image data of the nut tree and uses the acquired initial three-dimensional image as the nut tree positioning map.

[0031] Further, the specific process of step 4.2 is as follows: when framing the initial three-dimensional image, a box is used for framing, and the four sides of the box are close to the edge of the nut tree. Then, several edge points and three internal midpoints are obtained from the three-dimensional image of the nut tree as feature recognition points. The feature recognition points are used as reference points for the subsequent fusion of several images. During fusion, points in each three-dimensional image that are the same as or similar to the feature recognition points are used as coincident points and attached together, so that the corresponding coordinate points can be guaranteed to correspond when all three-dimensional images are fused.

[0032] Further, the specific process of step 4.3 is as follows: establish a three-dimensional coordinate system, then take the lower left corner of the framed three-dimensional image as the origin, and assign corresponding three-dimensional coordinate data to each point on the three-dimensional image. Take the center point inside the X-axis and Y-axis planes at the foremost edge as the recognition center point, and the recognition center point also serves as the reference coincidence point for image fusion.

[0033] Further, the specific process of step 4.4 is as follows: The LiDAR first moves from the intersection of the horizontal and vertical slide bars, first moving towards the upper part of the vertical slide bar. After moving to a fixed position, it pauses and then instructs the LiDAR to scan a three-dimensional image of the nut-skinned tree. After scanning, it moves upward to a fixed position and scans again until it reaches the top. Then it returns to the intersection of the horizontal and vertical slide bars and moves downward to scan, obtaining several three-dimensional images of vertical movement. The LiDAR then returns to the intersection of the horizontal and vertical slide bars and moves to both sides to scan, obtaining several three-dimensional images of horizontal movement. The several three-dimensional images of vertical movement and the several three-dimensional images of horizontal movement are combined to obtain a three-dimensional image set.

[0034] The present invention, by adopting the above-described technical solution, has the following beneficial effects:

[0035] (1) This invention uses the position of the moving laser radar to image the green nut from different angles, and recognizes the shape of the nut from different angles. This can effectively avoid the inability to identify the nut due to the obstruction of leaves or branches in front, and make the imaging positioning more effective and accurate. After recognition, the obstructing leaves or branches are deleted by image fusion, so that the nuts behind can be directly exposed. This achieves automatic filtering of obstructions, and the position of the nut does not change, making it easier to position when using a robotic arm to pick the nuts.

[0036] (2) The present invention first uses lidar to locate the green nut in multiple directions, avoiding the trouble caused by the green leaves on the image recognition. Furthermore, it constructs a three-dimensional structure space by combining the movement space of the picking robot with the position space of the green nut, realizing the distance between the collection and storage point and each nut. It also realizes the robot control program required to calculate each nut while picking, saving data processing time and improving picking efficiency.

[0037] (3) The harvesting process of this invention is fully automated, with high harvesting effect and a very low probability of missed harvesting, achieving precise and fast harvesting. Attached Figure Description

[0038] Figure 1 This is a flowchart of the automatic harvesting method of the present invention;

[0039] Figure 2 This is a flowchart of the initial imaging and positioning method for green-skinned nuts using lidar according to the present invention;

[0040] Figure 3 This is a flowchart of the method for constructing a three-dimensional spatial model of nut and green-skinned fruit harvesting according to the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and preferred embodiments. However, it should be noted that many details listed in the specification are merely to provide the reader with a thorough understanding of one or more aspects of the present invention, and these aspects of the invention can be implemented even without these specific details.

[0042] like Figure 1 As shown, an automated and rapid harvesting method for green-skinned nuts in hilly areas includes the following steps:

[0043] Step 1: Install a balance detection device on the bottom plate of the tracked trolley, and simultaneously install a lifting device at the bottom of the bottom plate. The tracked trolley can automatically climb hilly areas. Before harvesting nuts, the balance detection device is used to check the level of the trolley's bottom plate. The lifting device consists of four adjustable screw rods. The control device is connected to the balance detection device, and one or two screw rods are raised or lowered according to the horizontal deviation.

[0044] Step 2: Install a lidar detection device, a robotic arm harvesting device, and a camera on the chassis of the trolley. The lidar detection device is located at the front of the trolley and includes a lidar sensor and longitudinal and lateral movement devices. The longitudinal and lateral movement devices include lateral and longitudinal sliding rods arranged in a cross shape. The lidar sensor is equipped with a moving wheel device that can switch between the lateral and longitudinal sliding rods. The moving wheel device is also equipped with a timed movement control device to control the moving speed, position, and dwell time of the moving wheel device. The timed movement control device is connected to both the lidar sensor and the external harvesting machine to receive movement control signals and notify the lidar sensor to scan.

[0045] The moving wheel device has a square base plate with four rotatable pulleys on each of its four sides. For longitudinal movement, the pulleys on the two sides rotate to slide against the back of the horizontal sliding rod, while the two pulleys at the top and bottom retract. For lateral movement, the reverse is applied, enabling both longitudinal and lateral movement. The lidar is a 192-line lidar.

[0046] Step 3: Drive the picking cart to the front of the nut tree, and then check whether the bottom of the cart is level. If it is not level, adjust it to a level position using the lifting device.

[0047] Step 4: Use lidar to perform initial imaging and positioning of the green-skinned nuts.

[0048] Step 4.1: The lidar, mounted on the harvester, approaches the nut tree and acquires an initial 3D image of the nut area. While acquiring this initial 3D image, the harvester uses the lidar for height identification to determine the height of the nut tree. The harvester aligns the center of its longitudinal and lateral movement devices with the center of the nut tree. The lidar stops at the intersection of the lateral and longitudinal sliding bars of the longitudinal and lateral movement devices. The lidar then begins scanning the initial 3D image data of the nut tree and uses this initial 3D image as the location map of the nut tree.

[0049] When harvesting is needed, you can come to one direction in front of the nut tree, first identify and locate it, and then the whole machine will move and pick the nuts using a robotic arm. Basically, to harvest the nuts from a nut tree, you only need to identify it in the four directions of east, west, south, and north, and then pick the nuts from each direction four times to complete the location of the nuts from a tree. After identifying one side, pick the nuts from that side first.

[0050] Step 4.2: Define the initial 3D image and then label several feature points as recognition points. When defining the initial 3D image, use a bounding box, ensuring all four sides of the box closely approximate the edge of the nut-skinned tree. Then, extract several edge points and three interior midpoints from the 3D image of the nut-skinned tree as feature recognition points. These feature recognition points serve as reference points for the subsequent fusion of multiple images. During fusion, points in each 3D image that are identical or similar to the feature recognition points are used as coincident points and fused together, ensuring that corresponding coordinate points correspond during fusion. The feature recognition points serve as a secondary calibration for the coordinate annotations in subsequent image fusion, preventing excessive coordinate position deviations between images.

[0051] Step 4.3: Construct a 3D coordinate system for the defined area, set the recognition center point, and then annotate the coordinates of all points on the nut tree. Establish a 3D coordinate system, then use the lower left corner of the defined 3D image as the origin, and annotate the corresponding 3D coordinate data for each point on the 3D image. Use the center point inside the X-axis and Y-axis planes at the foremost point as the recognition center point, which also serves as the reference point for image fusion. The recognition center point serves as a secondary localization and verification point, making the annotation of all images more accurate.

[0052] Step 4.4: The lidar moves along the longitudinal and lateral movement devices, pausing after each point of movement. The lidar scans to obtain a 3D image in one direction. Then, all the 3D images are acquired and summarized to obtain a set of 3D images of the nut tree in different directions. The lidar first starts moving from the intersection of the lateral and longitudinal sliders, moving towards the upper part of the longitudinal slider. After moving to a fixed position, it pauses and then scans a 3D image of the nut tree. After scanning, it moves upward to a fixed position and scans again until it reaches the top. Then it returns to the intersection of the lateral and longitudinal sliders and moves downward to obtain several 3D images in the vertical direction. The lidar then returns to the intersection of the lateral and longitudinal sliders and moves to both sides to obtain several 3D images in the horizontal direction. The 3D images in the vertical and horizontal directions are then summarized to obtain a set of 3D images.

[0053] When moving, a stereoscopic image is usually scanned and captured after moving 3-5cm. First, it moves upward, then downward, then to the left, and then to the right.

[0054] Step 4.5: Fuse all the 3D images to obtain 3D coordinate datasets of the nut tree in different orientations. Each image in the 3D image set is then laminated into the initial 3D image. During fusion, several layers are created, ensuring that occlusions between images do not affect each other. Then, the 3D coordinate datasets of each layer are annotated. This fusion process does not involve fusion of coordinate data itself, but rather layer stacking. Only the length and width directions are fused, ensuring that the 3D coordinate data pairs of all images correspond correctly and do not exhibit significant deviations.

[0055] Step 4.6: Then, find the positions of all nut shapes from the 3D coordinate dataset of the nut tree and output the 3D position points of the nuts to achieve imaging and localization of the nuts. Identify the green nut model image in each layer of the 3D image, and then collect the position coordinates of the green nut model image. Summarize the position coordinates of the green nut model images in all layers of the image to obtain the set coordinates of the nut shape's position. Then, compare each layer of the image with the set coordinates. If the image in a layer does not contain a green nut model image at the position where the coordinates of the set coordinates overlap, delete the image in that layer. Process all layers of the image in turn to remove obstructing leaves and branches. Then, perform a second complete fusion of all images to obtain the position points of the nut model image. This achieves the filtering out of obstructing leaves or branches, resulting in more accurate nut positioning.

[0056] By fusing coordinate data and image pixels, the occluded parts are deleted, thus filtering out the obstructions. This "perspective" effect allows for better localization and identification of green-skinned nuts, avoiding missed identifications.

[0057] This method moves the position of the lidar to image the green nut from different angles, enabling the identification of the nut's shape from various perspectives. This effectively avoids the inability to identify the nut due to obstruction by leaves or branches directly in front, making imaging and positioning more effective and accurate. After identification, the obstructing leaves or branches are removed through image fusion, allowing the nut behind to be directly exposed. This achieves automatic filtering of obstructions, and the nut's position does not change, making it easier to locate when using a robotic arm for harvesting.

[0058] Step 5: Construct a three-dimensional spatial model for harvesting green-skinned nuts.

[0059] Step 5.1: Construct a three-dimensional spatial location model of the green-skinned nuts based on their positions. Based on the obtained position of each green-skinned nut, construct a three-dimensional spatial model, and mark the three-dimensional coordinates of each nut on this model. Then, summarize all the three-dimensional spatial coordinates of the green-skinned nuts to obtain the three-dimensional spatial location model of the green-skinned nuts.

[0060] Step 5.2: Install distance sensors on the harvesting robot arm to measure the distance between the robot arm and the object in front of it in real time. Three distance sensors are installed on the harvesting robot arm. One sensor is installed at the front of the robot arm to detect the position of the object directly in front of it in real time. The other two sensors are installed on both sides of the harvesting robot arm and are tilted to detect the position of objects on either side of the robot arm.

[0061] Step 5.3: Construct a three-dimensional spatial model for harvesting based on the fixed position of the harvesting robot, the position of the nut collection box, and the three-dimensional spatial position model of the green nut. The last end of the nut collection box is taken as the last edge, the lowest point of the harvesting robot is taken as the bottom edge, the horizontal plane of the highest nut in the three-dimensional spatial position model is taken as the top edge, and the vertical plane of the foremost nut in the three-dimensional spatial position model is taken as the front edge. The three-dimensional spatial box structure is constructed based on the last edge, bottom edge, top edge, and front edge. The position of the harvesting robot within the three-dimensional spatial box structure during its movement is then marked, forming the three-dimensional spatial model for harvesting.

[0062] Step 5.4: Use a camera to observe the status of the robotic arm, the nut collection box, and the harvesting environment in real time within the 3D harvesting model. The camera collects data on whether the nut collection box is full, then notifies relevant personnel to replace it. Simultaneously, it detects whether personnel are within the robotic arm's range of motion to prevent accidental injury. Based on the robotic arm's position, the camera uses image recognition to identify the nut bunches as the robotic arm approaches them, calibrating the gripper's position and improving harvesting efficiency. Furthermore, when the robotic arm passes through leaves and pushes them aside, the camera continuously identifies any unidentified nuts, supplementing the harvesting process and preventing missed nuts.

[0063] Step 5.5: Based on the state and position of the robotic arm, calculate the real-time position of the next nut to be picked, facilitating the robotic arm's picking. A camera determines whether the robotic arm is open or closed and identifies whether the nut was successfully picked. If the picking was unsuccessful, a second picking attempt is performed at the previously picked position. Simultaneously, the straight-line distance between each nut in the 3D picking space model and the nut collection box is calculated, and these distances are sorted to determine the nut picking order. When the robotic arm picks a nut and puts it back into the nut collection box, radar distance detection transmits the position of the next nut to be picked to the robotic arm control device, simultaneously generating the robotic arm's control rotation dimension. This achieves real-time nut positioning and improves picking efficiency.

[0064] By placing all the nuts in a designated space, the harvesting order and path can be better planned, avoiding missing any nuts during the harvesting process. Furthermore, the use of cameras and distance sensors in conjunction with the system makes nut harvesting more efficient and also enables secondary internal identification, allowing even nuts that are heavily covered by leaves to be identified, resulting in better harvesting outcomes.

[0065] Step 6: Calculate the harvesting order of all nuts based on the 3D spatial model, and then control the robotic arm to harvest them one by one. Since the coordinates of all nuts and the nut collection box are known in the 3D spatial model, the distances between each nut and the collection box can be calculated. Each nut is labeled, and all distances are stored in an array. The array data is then sorted from smallest to largest, and the output array represents the harvesting order. Nuts are located based on the output distance table, and the robotic arm's path is planned in advance.

[0066] Step 7: After picking the nuts on one side, move the cart to the side of the nut tree and repeat steps 3-6 until the entire nut tree is picked. Each nut can be picked from the four directions of east, west, south, and north.

[0067] This method enables automated harvesting in hilly areas without human intervention, resulting in higher harvesting efficiency and better results.

[0068] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An automated and rapid harvesting method for green-skinned nuts in hilly areas, characterized in that: The method includes the following steps: Step 1: Install a balance detection device on the bottom plate of the tracked trolley, and at the same time install a lifting device at the bottom of the bottom plate of the trolley; Step 2: Install a lidar detection device, a robotic arm harvesting device, and a camera on the bottom plate of the vehicle; Step 3: Drive the picking cart to the front of the nut tree, and then check whether the bottom of the cart is level. If it is not level, adjust it to a level position using the lifting device. Step 4: Use lidar to perform initial imaging and localization of the green-skinned nuts; Step 5: Construct a three-dimensional spatial model for harvesting green-skinned nuts; Step 6: Calculate the picking order of all nuts based on the three-dimensional spatial model of the harvesting, and then control the robotic arm to pick them one by one; Step 7: After picking the nuts on one side, move the cart to the side of the nut tree and repeat steps 3-6 until the entire tree is harvested. Each nut can be picked from the four directions: east, west, south, and north. In step 1, the tracked trolley automatically climbs hills. Before harvesting nuts, a balance detection device is used to check the level of the trolley's bottom plate. The lifting device is set with four adjustable screw rotating rods. The control device is connected to the balance detection device and raises or lowers one or two screw rotating rods according to the horizontal deviation. In step 2, the lidar detection device is set at the front end of the trolley. The lidar detection device includes a lidar and a longitudinal and lateral movement device. The longitudinal and lateral movement device includes a lateral slide bar and a longitudinal slide bar, which are arranged in a cross shape. The lidar is equipped with a moving wheel device, which can switch between the lateral slide bar and the longitudinal slide bar. At the same time, the moving wheel device is equipped with a timed movement control device, which is used to control the moving speed, moving position and dwell time of the moving wheel device. The timed movement control device is connected to the lidar and the external harvesting machine respectively, and is used to receive movement control signals and notify the lidar to scan. The specific process of step 4 is as follows: Step 4.1: The lidar is installed on the harvester and it comes to the front of the nut tree. The lidar acquires an initial three-dimensional image of the nut area for the first time. Step 4.2: Define the initial 3D image and then label several feature points as recognition points; Step 4.3: Construct a three-dimensional coordinate system for the defined area, set the recognition center point, and then label the coordinates of all points on the nut tree; Step 4.4: The lidar moves along the longitudinal and lateral moving device, pausing after each point. The lidar scans to obtain a stereo image in one direction, and then all the three-dimensional images are acquired and summarized to obtain a set of three-dimensional stereo images of the nut tree in different directions. Step 4.5: Fuse all the 3D images to obtain the 3D coordinate dataset of the nut tree in different directions; Step 4.6: Then find the positions of all nut shapes from the nut tree 3D coordinate dataset and output the 3D position points of the nuts to achieve imaging and localization of the nuts.

2. The automatic and rapid harvesting method for green-skinned nuts in hilly areas according to claim 1, characterized in that: The specific process of step 5 is as follows: Step 5.1: Construct a three-dimensional spatial location model of the green-skinned nut based on its location; Step 5.2: Install a distance sensor on the harvesting robot arm to measure the distance between the robot arm and the object in front in real time; Step 5.3: Construct a three-dimensional spatial model for harvesting based on the fixed position of the harvesting robot, the position of the nut collection box, and the three-dimensional spatial position model of the green nut skin. Step 5.4: Use a camera to observe in real time the status of the robotic arm, the status of the nut collection box, and the data on the harvesting environment in the 3D model of the harvesting space; Step 5.5: Calculate the real-time location of the next nut to be picked based on the status and position of the robotic arm, so that the robotic arm can pick it.

3. The automatic and rapid harvesting method for green-skinned nuts in hilly areas according to claim 1, characterized in that: The specific process of step 6 is as follows: Since the position coordinates of all nuts are known in the three-dimensional spatial model of the harvesting, and the position coordinates of the nut collection box are also known, the distances of all nuts to the nut collection box are calculated, each nut is labeled, all distances are stored in an array, the array data is sorted from smallest to largest, and the output array sorting is the harvesting order. The nuts are found according to the sorting of the output distances, and the path of the robotic arm is planned in advance.

4. The automatic and rapid harvesting method for green-skinned nuts in hilly areas according to claim 1, characterized in that: The specific process of step 4.1 is as follows: When the lidar acquires the initial three-dimensional image, the harvester uses the lidar to identify the height and then determines the height of the nut tree. The harvester aligns the center of the longitudinal and lateral moving device with the middle position of the nut tree. The lidar stops at the intersection of the lateral and longitudinal sliding rods of the longitudinal and lateral moving device. The lidar begins to scan the initial three-dimensional image data of the nut tree and uses the acquired initial three-dimensional image as the nut tree positioning map.

5. The automatic and rapid harvesting method for green-skinned nuts in hilly areas according to claim 1, characterized in that: The specific process of step 4.2 is as follows: When framing the initial 3D image, a box is used for framing. Then, the four sides of the box are close to the edge of the nut tree. Then, several edge points and three internal midpoints are obtained from the 3D image of the nut tree as feature recognition points. The feature recognition points are used as reference points for the subsequent fusion of several images. During fusion, points in each 3D image that are the same as or similar to the feature recognition points are used as coincident points and attached together, so that the corresponding coordinate points can be guaranteed to correspond when all 3D images are fused.

6. The automatic and rapid harvesting method for green-skinned nuts in hilly areas according to claim 1, characterized in that: The specific process of step 4.3 is as follows: establish a three-dimensional coordinate system, then take the lower left corner of the framed three-dimensional image as the origin, and assign corresponding three-dimensional coordinate data to each point on the three-dimensional image. Take the center point inside the X-axis and Y-axis planes at the foremost edge as the recognition center point, which also serves as the reference coincidence point for image fusion.

7. The automatic and rapid harvesting method for green-skinned nuts in hilly areas according to claim 1, characterized in that: The specific process of step 4.4 is as follows: The lidar first moves from the intersection of the horizontal and vertical slide bars, first moving towards the upper part of the vertical slide bar. After moving to a fixed position, it pauses and then instructs the lidar to scan a three-dimensional image of the nut-skinned fruit tree. After scanning, it moves upward to a fixed position and scans again until it reaches the top. Then it returns to the intersection of the horizontal and vertical slide bars and moves downward to scan, obtaining several three-dimensional images of vertical movement. The lidar then returns to the intersection of the horizontal and vertical slide bars and moves to both sides to scan, obtaining several three-dimensional images of horizontal movement. The several three-dimensional images of vertical movement and the several three-dimensional images of horizontal movement are combined to obtain a three-dimensional image set.

Images