An intelligent picking hand based on an improved YOLOv5 model

Abstract

The application relates to an intelligent picking hand based on an improved YOLOv5 model, which comprises a base and a driving controller, the base and the driving controller are connected through a connecting rod mechanism, a picking turntable is arranged on the driving controller, a tray is arranged on the side of the picking turntable away from the driving controller, a plurality of mechanical claws are arranged in the circumferential direction of the tray, each mechanical claw is slidably installed on a sliding rail of the picking turntable through a moving block assembly under the mechanical claw, the mechanical claw is rotationally connected with the moving block assembly, all the mechanical claws and the tray jointly form a picking space, and a camera with the improved YOLOv5 model arranged inside is installed on the driving controller. When picking, the driving controller identifies and positions the fruit to be picked through the camera, controls the mechanical claw to approach the fruit, expands the picking space to cover the fruit, shrinks the picking space to clamp the fruit, and finally controls the picking turntable to rotate to complete the picking. The application can match picking of fruits of various sizes and has higher picking efficiency.

Description

Technical Field

[0001] This invention relates to the field of fruit picking, specifically to an intelligent picking hand based on an improved YOLOv5 model. Background Technology

[0002] Currently, with the development of robotics and intelligent control technologies, robotic harvesters are increasingly widely used in agriculture. However, current intelligent harvesting robotic grippers have limited application scenarios, typically only able to harvest one type of fruit, and their response speed is slow. Therefore, this invention proposes an intelligent harvester based on an improved YOLOv5 model. Summary of the Invention

[0003] To address the limitations of existing mechanical harvesters, which can only harvest a single type of fruit, have a limited scope of application, and suffer from slow target detection algorithms and low harvesting efficiency, this invention proposes an intelligent harvester based on an improved YOLOv5 model. This intelligent harvester can harvest fruits of various sizes while improving work efficiency. The specific technical solution is as follows:

[0004] A smart harvester based on an improved YOLOv5 model includes a base and a drive controller. The base and drive controller are connected by a linkage mechanism. A harvesting turntable is located at the other end of the drive controller. A tray is located on the side of the harvesting turntable away from the drive controller. Multiple mechanical claws are arranged circumferentially on the tray. Each mechanical claw is slidably mounted on a slide rail of the harvesting turntable via a moving block assembly on its lower side. The mechanical claws are rotatably connected to the moving block assembly. All mechanical claws and the tray together form a harvesting space. A camera is mounted on the drive controller, and the improved YOLOv5 model is deployed within the camera. The improved YOLOv5 model uses an improved Shuffling algorithm. LeNetV2 is used as the backbone network, GhostConv is used as the convolution in the neck network, and a fused bidirectional feature pyramid is used as the connection method of the neck network. The EDCA attention mechanism is introduced. The drive controller communicates with the camera, linkage mechanism, each mechanical claw and moving block component. During harvesting, the drive controller identifies and locates the fruit to be harvested through the camera, then controls the linkage mechanism to rotate and move closer to the fruit. Subsequently, it controls the rotation of the mechanical claw on the moving block component and the sliding of the moving block component on the harvesting turntable to expand the harvesting space to cover the fruit and then shrink the harvesting space to clamp the fruit. Finally, it controls the rotation of the harvesting turntable to complete the harvest.

[0005] Furthermore, the linkage mechanism includes a telescopic rod 1 fixedly connected to the base, a telescopic rod 2 rotatably connected to the telescopic rod 1, a telescopic rod 3 rotatably connected to the telescopic rod 2, and a telescopic rod 4 rotatably connected to the telescopic rod 3. The telescopic rod 4 is rotatably connected to the drive controller. The rotation and extension of each telescopic rod, as well as the rotation of the drive controller relative to the telescopic rod 4, are controlled by the drive controller.

[0006] Furthermore, the moving block assembly includes a mechanical claw moving block located on the upper side and rotatably connected to the mechanical claw, and a track moving block located on the lower side and slidably connected to the slide rail. The mechanical claw moving block and the track moving block are rotatably connected. An angle sensor is provided at the rotatable connection between the mechanical claw and the mechanical claw moving block. The slide rail is provided with a track baffle at the edge of the picking turntable, and a distance sensor is installed inside the slide rail to measure the distance between each mechanical claw and its corresponding track baffle.

[0007] Furthermore, a spring is installed between the picking turntable and the tray, and a rubber pad is installed on the mechanical claw. A pressure sensor is installed on the side of the rubber pad that contacts the fruit.

[0008] Furthermore, the improved YOLOv5 model building steps are as follows:

[0009] S1. Constructing the dataset: Use photographic equipment to photograph fruits under natural conditions to construct the original dataset;

[0010] S2. Data preprocessing: Data augmentation is performed on the original dataset, followed by labeling the fruits in the images. Finally, the labeled dataset is randomly divided into training set, validation set and test set according to the proportion.

[0011] S3. Build an improved YOLOv5 model: Replace the backbone network of the original YOLOv5 with an improved ShuffleNetV2, replace the regular convolutions in the neck network with GhostConv, replace the connection method of the neck network with a fused bidirectional feature pyramid, and introduce the EDCA attention mechanism.

[0012] S4. Train and validate the improved YOLOv5 model using the training and validation sets, save the optimal weights during training, and then test it using the test set.

[0013] Furthermore, in the improved ShuffleNetV2, residual connections were introduced into the basic feature extraction module of the original model. Max pooling layers were used to complete downsampling in the residual connections of its downsampling module, and the number of channels was adjusted by 1×1 convolution. A dual-channel feature extraction module was introduced to replace the 3*3 depth convolution in the basic feature extraction module.

[0014] Furthermore, in S1, the sample image is adjusted to a size of 640×640×3.

[0015] Furthermore, in S2, the labeled dataset is randomly divided into training, validation, and test sets in an 8:1:1 ratio.

[0016] Furthermore, in S2, the dataset is divided, and k-means clustering and genetic learning algorithms are used to reset the initial anchor boxes of the labeled dataset.

[0017] Furthermore, the EDCA attention mechanism uses average pooling and standard deviation pooling to process the input features, compressing them into 1×1×C feature maps, where C is the number of channels in the feature map. Weights are generated for each channel through a one-dimensional convolution of size K. The elements of the feature maps obtained from the two paths are added together, and the weights of each channel are obtained through the Sigmoid function. Finally, the weights ω are multiplied by the original input feature map to obtain the output feature map.

[0018] The beneficial effects of this invention are as follows:

[0019] 1. It is equipped with a sliding mechanical claw, which makes the harvesting space large or small, enabling the harvesting and grading of fruits of various sizes and shapes.

[0020] 2. The improved ShuffleNetV2 model is used to replace the backbone network of YOLOv5, reducing model parameters and improving the detection speed. GhostConv is used to replace the regular convolutions in the neck network, further reducing model parameters. A self-designed EDCA attention mechanism is introduced into the C3 module of the neck structure of the YOLOv5 model to improve the model's feature representation ability, thereby enhancing the model's ability to detect targets. The BiFPN structure is used to replace the PANet layer in YOLOv5, thereby reducing the loss between different structural layers and effectively improving the model's ability to extract multi-scale features. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the structure of the intelligent harvesting hand described in this invention;

[0023] Figure 2 This is a schematic diagram of the mechanical claw moving block described in this invention;

[0024] Figure 3 This is a schematic diagram of the mechanical gripper described in this invention;

[0025] Figure 4 This is a network structure diagram of the improved YOLOv5 model described in this invention;

[0026] Figure 5 This is a network structure diagram of the improved ShuffleNetV2 described in this invention;

[0027] Figure 6 This is a network structure diagram of the fused bidirectional feature pyramid described in this invention;

[0028] Figure 7 This is a structural diagram of the EDCA attention mechanism described in this invention;

[0029] Figure 8 This is a schematic diagram of the intelligent harvester described in this invention harvesting oval-shaped fruits.

[0030] In the diagram: 1. Base; 2. Telescopic rod one; 3. Telescopic rod two; 4. Telescopic rod three; 5. Telescopic rod four; 6. Drive controller; 7. Camera; 8. Harvesting turntable; 9. Moving track A; 10. Moving track B; 11. Track moving block; 12. Mechanical claw moving block; 12.1. Rotating shaft; 12.2. Rotating ring; 13. Spring; 14. Tray; 15. Track baffle; 16. Rubber pad; 17. Mechanical claw; 17.1. Shaft hole. Detailed Implementation

[0031] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0032] Taking apple picking as an example, the present invention provides the following specific implementation scheme:

[0033] like Figure 1-3As shown, this invention provides an intelligent harvester based on an improved YOLOv5 model, including a base 1 and a drive controller 6. The base 1 and the drive controller 6 are connected by a linkage mechanism. A harvesting turntable 8 is provided at the other end of the drive controller 6, and the harvesting turntable 8 can rotate according to the drive controller 6. A tray 14 is provided on the side of the harvesting turntable 8 away from the drive controller 6. Multiple mechanical claws 17 are provided on the circumference of the tray 14. Each mechanical claw 17 is slidably mounted on the slide rail of the harvesting turntable 8 via a moving block assembly on its lower side. The mechanical claws 17 are rotatably connected to the moving block assembly. All mechanical claws 17 and the tray 14 together form a harvesting space. A camera 7 is installed on the drive controller 6, and an improved YOLOv5 model is deployed within the camera 7. The improved YOLOv5 model uses an improved ShuffleNetV2. As the backbone network, GhostConv replaces the conventional convolutions in the neck network, and a fused bidirectional feature pyramid is used as the connection method for the neck network, introducing the EDCA attention mechanism. The drive controller is connected to the camera, linkage mechanism, and each mechanical claw. During harvesting, the drive controller 6 identifies and locates the fruit to be harvested through the camera 7, then controls the linkage mechanism to rotate and move closer to the fruit. Subsequently, it controls the mechanical claw 17 to rotate on the moving block assembly and the moving block assembly to slide on the harvesting turntable 8 to expand the harvesting space to cover the fruit, and then moves in the opposite direction to shrink the harvesting space to clamp the fruit. Finally, it controls the harvesting turntable 8 to rotate to complete the harvesting. In order to adapt to harvesting fruits of different sizes and shapes, the mechanical claw 17 can rotate 360 ​​degrees on the mechanical claw moving block 12 to facilitate the harvesting of more types of fruits.

[0034] Preferably, the linkage mechanism is a four-bar linkage, which includes a telescopic rod 1 2 fixedly connected to the base 1, a telescopic rod 2 3 rotatably connected to the telescopic rod 1 2, a telescopic rod 3 4 rotatably connected to the telescopic rod 2 3, and a telescopic rod 4 5 rotatably connected to the telescopic rod 3 4. The telescopic rod 4 5 is rotatably connected to the drive controller 6. The rotation and extension of each telescopic rod, as well as the rotation of the drive controller 6 relative to the telescopic rod 4 5, are controlled by the drive controller 6.

[0035] Preferably, the moving block assembly includes a mechanical claw moving block 12 located on the upper side and rotatably connected to the mechanical claw 17, and a track moving block 11 located on the lower side and slidably connected to the slide rail. The mechanical claw moving block 12 is rotatably connected to the track moving block 11 via a rotating ring 12.2 located at its lower part. Correspondingly, the track moving block is provided with a groove for mounting the rotating ring 12.2. The rotation of the mechanical claw moving block 12 is controlled by a drive controller. The concave surface of the mechanical claw 17 faces the tray 14. The mechanical claw 17 has a shaft hole 17.1. The upper part of the mechanical claw moving block 12 is provided with a rotating shaft 12.1. The mechanical claw 17 and the mechanical claw moving block 12 are rotatably connected via the rotating shaft 12.1 and the shaft hole 17.1. An angle sensor is provided at the rotatable connection point between the mechanical claw 17 and the mechanical claw moving block 12, i.e., at the rotating shaft 12.1. The slide rail is provided with a track baffle 15 located at the edge of the picking turntable 8. The slide rail is equipped with a distance sensor for measuring the distance between each mechanical claw 17 and its corresponding track baffle 15.

[0036] Preferably, the camera 7 is a binocular depth camera, a spring 13 is provided between the picking turntable 8 and the tray 14, and a rubber pad 16 is provided on the mechanical claw 17 to prevent the fruit from being damaged by pressure; the tray 14 is made of rubber; a pressure sensor is provided on the side of the rubber pad 16 that contacts the fruit to determine whether it is in contact with the fruit and the amount of force applied to the fruit.

[0037] In this embodiment, the slide rail on the side of the picking turntable 8 away from the drive controller 6 includes a moving track A9 and a moving track B10 that intersect each other perpendicularly. There are 4 mechanical claws 17, one pair of mechanical claws 17 are installed on the moving track A9, and the other pair of mechanical claws 17 are installed on the moving track B10.

[0038] Preferably, taking Apple as an example, the improved YOLOv5 model building steps are as follows:

[0039] S1. Construct the apple dataset: Use shooting devices such as mobile phones and cameras to take pictures of fruits under natural conditions to construct the original dataset. Natural conditions include cloudy days, rainy days, front lighting and backlighting. To ensure the recognition speed of the model, adjust the images so that the input image size is 640×640×3.

[0040] S2. Data Preprocessing: The original dataset is augmented using mosaic data enhancement and enhancement methods such as brightness, contrast, and Gaussian blur. The fruits in the images are labeled using Labelimg software. During the labeling process, the fruits can be classified into multiple levels based on their color. Finally, the initial anchor boxes of the labeled dataset are reset using a k-means clustering + genetic learning algorithm. The dataset is then randomly divided into training, validation, and test sets in an 8:1:1 ratio. In this embodiment, the number of anchor boxes, input image size, maximum aspect ratio threshold, and number of iterations in the k-means clustering + genetic learning algorithm are set to 9, 640, 4, and 1000, respectively. The most suitable sizes for apple dataset recognition are [12, 15, 17, 23, 27, 35], [46, 55, 92, 97, 199, 146], and [237, 244, 325, 255, 330, 331]. (This process can be followed to preprocess the data and initial anchor frames when harvesting other fruits.)

[0041] S3, such as Figure 4 As shown, an improved YOLOv5 model is constructed: the backbone network of the original YOLOv5 is replaced with an improved ShuffleNetV2 to reduce the number of model parameters and improve the model's recognition speed; GhostConv is used to replace the regular convolutions in the neck network to further reduce the model parameters and facilitate model deployment; the connection method of the neck network is replaced with a bidirectional feature pyramid (BiPFN) to enhance the ability to detect objects at different scales by bidirectionally fusing low-level spatial features and high-level semantic features. The network structure diagram of BiPFN is shown below. Figure 6 As shown, the EDCA attention mechanism is introduced to improve target recognition capabilities. In the improved YOLOv5 architecture, the GBS feature extraction module consists of GhostConv convolutional layers and stacked BatchNormalization layers and activation functions. The C3 module consists of regular convolutions, batch normalization layers, and activation functions. In the improved YOLOv5 architecture, the backbone network performs five downsampling operations, and after passing through the neck network, three detection layers of 160*160, 80*80, and 40*40 are generated in the head network for multi-scale target detection.

[0042] Among them, such as Figure 5As shown, the improved ShuffleNetV2 introduces residual connections into the basic feature extraction module of the original model to solve the gradient vanishing problem, enabling the model to learn more complex feature representations. In its downsampling module, max pooling layers are used in the residual connections to perform downsampling, and 1×1 convolutions are used to adjust the number of channels, thus ensuring consistent output channel counts. A dual-channel feature extraction module is introduced to replace the 3*3 depthwise convolution in the original model's basic feature extraction module, improving the model's ability to extract lesions at different scales. To facilitate the model's use in YOLOv5, the basic feature extraction module of the improved ShuffleNetV2 model is named SFB1, and the downsampling feature extraction module is named SFB2.

[0043] like Figure 7 As shown, in the EDCA attention mechanism, X represents the input feature, and the feature map has a size of H×W×C, where H represents the height of the feature map, W represents the width of the feature map, and C represents the number of channels in the feature map. The EDCA attention mechanism uses average pooling (AvgPool) and standard deviation pooling (StdPool) to process the input, compressing it into a 1×1×C feature map, and generating weights for each channel through a one-dimensional convolution of size K. The calculation of the mean and standard deviation is shown in formulas (1) and (2):

[0044]

[0045]

[0046] In equations (1) and (2), A c and S c This represents the mean and standard deviation of each element in the channel.

[0047] The kernel size K can be adaptively determined by nonlinearly mapping the channel dimension. The fitness function is defined as shown in formula (3):

[0048]

[0049] In equation (3), C represents the dimension of the input feature channels, |x| odd γ represents the odd number closest to x. γ and b are used to change the ratio between the number of channels C and the kernel size, and are set to γ ​​= 2 and b = 1 according to the empirical values ​​taken in the literature. Then, the elements of the feature maps obtained from the two paths are added together, and the weight ω of each channel is obtained through the Sigmoid function. Finally, the weight ω is multiplied by the original input feature map to obtain the output feature map. The calculation of the weight ω is shown in formula (4):

[0050] ω=σ(C1D k (y1)+C1D k(y2)) (4)

[0051] In equation (4), σ represents the Sigmoid activation function, C1D represents one-dimensional convolution, K represents the size of the one-dimensional convolution kernel, y1 represents the feature map output by the average pooling path, and y2 represents the feature map output by the standard deviation pooling path.

[0052] S4. The number of images input to the improved YOLOv5 model in each batch is 32; the loop iterates for 300 rounds; the learning rate is 0.0024. The improved YOLOv5 model is trained and validated using the training and validation sets, the optimal weights during training are saved, the model is tested using the test set, and the model is deployed in the camera after passing the test.

[0053] In this embodiment, the harvesting process is as follows: During the fruit harvesting process, the camera 7 first acquires the target to be harvested and the distance to the target, locks the target, and transmits the information to the drive controller 6. The drive controller 6 then controls the rotation and extension of the telescopic rods 1-2, 2-3, 3-4, and 4-5 to move and extend the mechanical claw 17, thereby bringing the mechanical claw 17 closer to the fruit. Then, the mechanical claw 17 is controlled to rotate and open, and the track moving block 11 slides away from the tray 14. After the tray 14 reaches the fruit, the mechanical claw 17 rotates and retracts, and the track moving block 11 slides towards the tray 14 until the pressure sensor on the mechanical claw 17 senses the pressure, the mechanical claw 17 stops moving, and the harvesting turntable 8 rotates to complete the harvesting.

[0054] The circuit controller required during the fruit picking process is placed on the drive controller 6 to control the shooting of the camera 7, the rotation of the picking turntable 8, the rotation and extension of each telescopic rod in the linkage mechanism, the rotation of the mechanical claw 17 relative to the mechanical claw moving block 12, and the sliding of the track moving block 11.

[0055] When harvesting round fruits, the four robotic claws 17 are positioned in four directions, ensuring even force distribution and easy harvesting. Rubber pads 16 on the claws 17 prevent damage during harvesting, and pressure sensors on the pads transmit the pressure exerted by the claws 17 during gripping to the mobile device in real time for recording, preventing excessive pressure from damaging the fruit. The sliding of each claw 17 on the slide rail is individually controlled by the drive controller 6, allowing for efficient harvesting even of slightly irregularly shaped round fruits. Distance sensors installed on moving tracks A9 and B10 can measure the distance the mechanical claw 17 moves by measuring the relative distance between the track moving block 11 and the track baffle 15. The mechanical claw 17 and the rotating shaft 12.1 and shaft hole 17.1 in the mechanical claw moving block 12 are connected together to enable the rotation of the mechanical claw 17. An angle sensor installed at the rotating shaft 12.1 can sense the tilt angle of the mechanical claw 17. The size of the fruit can be determined by the distance sensor and the angle sensor, thus enabling the fruit to be graded by size during the picking process. Color grading is achieved by the camera 7. After the fruit is picked, one of the mechanical claws is placed horizontally on the ground. The weight of the fruit can be calculated by the pressure sensors on the other three mechanical claws to achieve the weight grading of the fruit, thus realizing the picking and grading of the fruit.

[0056] like Figure 8 As shown, when picking oval fruits, the four robotic arms are connected by the rotating ring 12.2 on the mechanical claw moving block 12 and the rotating connection of the track moving block 11, so as to achieve parallel pairing of the four robotic arms that are not diagonally opposite, which makes it easier to pick oval fruits.

[0057] This application incorporates a sliding mechanical claw 17, allowing the resulting picking space to be adjusted in size to accommodate the picking and grading of fruits of various sizes and shapes. The improved ShuffleNetV2 model replaces the YOLOv5 backbone network, reducing model parameters and improving detection speed. A self-designed EDCA attention mechanism is introduced into the C3 module of the YOLOv5 model's neck structure to enhance feature representation capabilities, thereby strengthening the model's target detection ability. A BiFPN structure replaces the PANet layer in YOLOv5, reducing losses between different structural layers and effectively improving the model's ability to extract multi-scale features.

[0058] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A smart harvester based on an improved YOLOv5 model, characterized in that: The system includes a base (1) and a drive controller (6), which are connected by a linkage mechanism. A picking turntable (8) is located at the other end of the drive controller (6). A tray (14) is located on the side of the picking turntable (8) away from the drive controller (6). Multiple mechanical claws (17) are located on the circumference of the tray (14). Each mechanical claw (17) is slidably mounted on the slide rail of the picking turntable (8) via a moving block assembly on its lower side. The mechanical claws (17) are rotatably connected to the moving block assembly. All mechanical claws (17) and the tray (14) together form a picking space. A camera (7) is mounted on the drive controller (6), and an improved YOLOv5 model is deployed inside the camera (7). The drive controller (6) respectively... The system is connected to the camera (7), linkage mechanism, mechanical claw (17), and moving block assembly. During harvesting, the drive controller (6) identifies and locates the fruit to be harvested through the camera (7), then controls the linkage mechanism to rotate and approach the fruit. Subsequently, it controls the rotation of the mechanical claw (7) on the moving block assembly and the sliding of the moving block assembly on the harvesting turntable (8) to expand the harvesting space to cover the fruit and then shrink the harvesting space to clamp the fruit. Finally, it controls the harvesting turntable (8) to rotate to complete the harvesting. The improved YOLOv5 model construction steps are as follows: S1, Constructing the dataset: Use a shooting device to take pictures of fruits under natural conditions to construct the original dataset; S2, Data preprocessing: Perform data augmentation on the original dataset, and then label the fruits in the pictures. Finally, the labeled dataset is randomly divided into training, validation, and test sets according to the proportions; S3, build an improved YOLOv5 model: replace the backbone network of the original YOLOv5 with an improved ShuffleNetV2, replace the regular convolutions in the neck network with GhostConv, replace the connection method of the neck network with a fused bidirectional feature pyramid, and introduce the EDCA attention mechanism; S4, train and validate the improved YOLOv5 model using the training and validation sets, save the optimal weights during training, and finally test it using the test set; In the improved ShuffleNetV2, residual connections are introduced into the basic feature extraction module of the ShuffleNetV2 model, and max pooling layers are used to complete downsampling in the residual connections of its downsampling module. The number of channels is adjusted by 1×1 convolution, and a dual-channel feature extraction module is introduced to replace the 3×3 depthwise convolution in the basic feature extraction module.

2. The intelligent harvester based on the improved YOLOv5 model according to claim 1, characterized in that: The linkage mechanism includes a telescopic rod 1 (2) fixedly connected to the base (1), a telescopic rod 2 (3) rotatably connected to the telescopic rod 1 (2), a telescopic rod 3 (4) rotatably connected to the telescopic rod 2 (3), and a telescopic rod 4 (5) rotatably connected to the telescopic rod 3 (4). The telescopic rod 4 (5) is rotatably connected to the drive controller (6). The rotation and extension of each telescopic rod, as well as the rotation of the drive controller (6) relative to the telescopic rod 4 (5), are controlled by the drive controller (6).

3. The intelligent harvester based on the improved YOLOv5 model according to claim 1, characterized in that: The moving block assembly includes a mechanical claw moving block (12) located on the upper side and rotatably connected to the mechanical claw (17) and a track moving block (11) located on the lower side and slidably connected to the slide rail. The mechanical claw moving block (12) and the track moving block (11) are rotatably connected. An angle sensor is provided at the rotatable connection between the mechanical claw (17) and the mechanical claw moving block (12). The slide rail is provided with a track baffle (15) at the edge of the picking turntable (8). The slide rail is equipped with a distance sensor for measuring the distance between each mechanical claw (17) and its corresponding track baffle (15).

4. The intelligent harvester based on the improved YOLOv5 model according to claim 1, characterized in that: A spring (13) is provided between the picking turntable (8) and the tray (14), and a rubber pad (16) is provided on the mechanical claw (17). A pressure sensor is provided on the side of the rubber pad (16) that contacts the fruit.

5. The intelligent harvester based on the improved YOLOv5 model according to claim 1, characterized in that: In S1, the sample image is adjusted to a size of 640×640×3.

6. The intelligent harvester based on the improved YOLOv5 model according to claim 1, characterized in that: In S2, the labeled dataset is randomly divided into training, validation and test sets in a ratio of 8:1:

1.

7. The intelligent harvester based on the improved YOLOv5 model according to claim 1, characterized in that: In S2, the dataset is divided, and k-means clustering and genetic learning algorithms are used to reset the initial anchor boxes of the labeled dataset.

8. The intelligent harvester based on the improved YOLOv5 model according to claim 1, characterized in that: The EDCA attention mechanism uses average pooling and standard deviation pooling to process the input features, compressing them into 1×1×C feature maps, where C is the number of channels in the feature map. Weights are generated for each channel through one-dimensional convolution. The elements of the feature maps obtained from the two paths are added together, and the weights of each channel are obtained through the Sigmoid function. Finally, the weights are multiplied by the original input feature map to obtain the output feature map.

Images