An intelligent immune injection robot based on body shape recognition and a dose control method thereof

Abstract

The present application relates to the technical field of livestock breeding immune injection, and particularly relates to an intelligent immune injection robot based on body shape recognition and a dose control method thereof. The method comprises the following steps: receiving ear tag information through identity recognition, collecting red-green-blue data and point cloud data of the back, side and hip regions of a dairy cow through an image acquisition unit, performing first recognition to obtain a three-dimensional model and a body shape recognition result; combining motion state tracking recognition, to-be-injected vaccine information and historical injection information to form a combination of objects and additional conditions, performing corresponding task processing to obtain dose information; processing a push cylinder, an injector push block and a first feeding mechanism to obtain a target motion amount, and completing second recognition, injection pose calculation, mechanical arm control, automatic injection device processing and uploading and distribution feedback. The present application continuously associates dose information with the target motion amount, and improves the adaptability, automation degree and consistency of historical injection information of the immune injection process.

Description

Technical Field

[0001] This invention relates to the field of immunization injection technology in animal husbandry, and in particular to an intelligent immunization injection robot based on body shape recognition and its dosage control method. Background Technology

[0002] In the field of livestock immunization technology, existing solutions for dairy cow immunization typically employ an identification receiver, image acquisition unit, automated injection device, and robotic arm control structure to identify individual cows, locate injectable areas, and perform injection based on vaccine information. However, these solutions suffer from limitations such as a disconnect between body shape recognition results and dosage information, a disconnect between dosage information and target movement, and a disconnect between historical injection information and current injection processing. Existing methods often rely on red-green-blue data, point cloud data, center position coordinates, and normal vectors for secondary identification and injection pose calculation. In scenarios where the cow is in motion, historical injection information needs to be synchronously accessed, and the automated injection device needs to process based on the current injection volume, the dosage information source becomes singular and the target movement is discontinuous. This makes it difficult to stably generate dosage information and target movement based on body shape recognition results through corresponding task processing and cylinder actuation.

[0003] Regarding the joint processing of body shape recognition results, corresponding task processing, and cylinder pusher processing, existing technologies generally suffer from shortcomings such as fragmented links, inconsistent calling relationships, and difficulty in synchronizing previous and subsequent records in the stages of object and additional condition combination, motion state tracking and recognition, dose information generation, target motion quantity generation, and uploading, distribution, and feedback. It is difficult to form a consistent process in dairy cow immunization injection application scenarios, including image acquisition and processing, first recognition processing, corresponding task processing, robotic arm control processing, automatic injection device processing, and uploading, distribution, and feedback processing. As a result, dose information cannot be continuously fed into the cylinder pusher, syringe pusher, and first feeding mechanism, and historical injection information is also difficult to be fed back into the object and additional condition combination processing. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a smart immune injection dosage control method based on body shape recognition, comprising:

[0005] S100: Acquire ear tag information, RGB data, and point cloud data, perform first recognition processing, and obtain body shape recognition results; wherein, the ear tag information is individual identification information read by the radio frequency identification identity recognition receiver, including ID information used to distinguish individual cows; the RGB data is red, green, and blue data; the point cloud data is a set of spatial points formed by unfolding the depth image in 3D coordinate space;

[0006] S200. Based on the body shape recognition result, perform motion state tracking and recognition, object and additional condition combination processing, and execute corresponding task processing to obtain dose information;

[0007] S300: Based on the dosage information, perform processing of the pushing cylinder, syringe pusher and first feeding mechanism and injection posture calculation to obtain the injection posture;

[0008] S400. Based on the injection pose, perform robotic arm control processing and execute upload, distribution and feedback processing to obtain historical injection information.

[0009] Furthermore, the first identification process includes:

[0010] Based on the RGB data and point cloud data, registration processing, cow body region filtering processing, and shape stitching processing are performed to generate a three-dimensional model;

[0011] The registration process includes aligning the point cloud data of the back, side, and hip regions with the RGB data.

[0012] The cow body area screening process includes removing fences, ground and non-dairy cow areas, and retaining the main area of ​​the dairy cow's torso.

[0013] The shape splicing process includes connecting the upper surface contour of the back, the side profile contour, and the rear boundary of the buttocks into a continuous shape.

[0014] Based on the three-dimensional model, key point information of the back region, side region and buttock region is extracted, a geometric model including back length relationship, side height relationship and buttock width relationship is established, and pose information including cow spatial orientation, torso tilt state and main body position state is generated.

[0015] The key point information, geometric model, and pose information are combined to form the body shape recognition result.

[0016] Furthermore, the process of motion state tracking and recognition includes:

[0017] Retrieve the body shape recognition results of the same ear tag information in the previous batch and compare them with the body shape recognition results of the current batch item by item;

[0018] When the average spatial geometric distance between key points in the back, side and rump regions is less than a preset distance threshold, the cow is determined to be in a stationary state.

[0019] Otherwise, it is determined to be in motion; the stationary state field is output.

[0020] Furthermore, the process of combining objects with additional conditions includes:

[0021] The ear tag information, the body shape recognition result, the static state, and the information on vaccines to be injected and historical injection information retrieved from the cloud platform or external platform are obtained.

[0022] The vaccine information to be injected includes the vaccine information that the dairy cow currently needs to receive;

[0023] The historical injection information includes injection records corresponding to the ear tag information;

[0024] When the static state is a processable state and the ear tag information matches the individual identifier in the vaccine information to be injected and the historical injection information, the body shape recognition result, the vaccine information to be injected, the historical injection information and the static state are merged into the same current recognition record to form a combination of object and additional conditions.

[0025] Furthermore, the process of executing the corresponding task processing includes:

[0026] Read the static state of the object and the additional conditions; if it is a moving state, mark it as a pending state and wait for the next call.

[0027] If the status is static, check if there is an injection record in the historical injection information that is identical to the information of the vaccine to be injected;

[0028] If no corresponding injection record exists, the dosage information for the current cow is generated by matching the vaccine information to be injected and the body shape recognition result according to the pre-stored correspondence in the database.

[0029] If a corresponding injection record already exists, output the dosage information for non-repeated injection and write it to the injection information status.

[0030] The dosage information is bound and saved together with the ear tag information and body shape recognition results.

[0031] Furthermore, the process of handling the cylinder, syringe pusher, and first feed mechanism includes:

[0032] Obtain the dosage information and determine the execution path based on the current structure of the automatic injection device;

[0033] When a structure is adopted in which the push cylinder and the syringe push block cooperate, the dosage information is converted into the pushing amount of the push cylinder and the movement amount of the syringe push block, and the target movement amount is written.

[0034] When the structure of the first feeding mechanism and the injection piece are used, the dosage information is converted into the calling state of the first feeding mechanism and written into the target motion amount;

[0035] Check the historical injection information retrieval status corresponding to the current ear tag information. If the same vaccine information to be injected has been injected, the target motion volume is recorded as closed; otherwise, a valid target motion volume is generated.

[0036] The target motion quantity is output, which includes the propulsion amount of the push cylinder, the motion amount of the syringe push block, and the call status of the first feed mechanism.

[0037] Further, the injection pose calculation and processing to obtain the injection pose specifically includes:

[0038] After generating effective target motion, the 3D model and body shape recognition results are used as the basis for the first-stage shape recognition process;

[0039] The second identification process includes:

[0040] Check if the target motion volume is in an effective state; if it is in a closed state, only generate the recorded results.

[0041] If it is in an effective state, search for injectable areas in local areas of the back, sides and buttocks. Extract local areas from the 3D model, filter candidate areas by local boundary continuity and surface smoothness, and determine injectable areas.

[0042] Perform center extraction on the injectable region to obtain the center position coordinates;

[0043] Perform local plane fitting on the injectable region to obtain the normal vector;

[0044] Based on the center position coordinates, normal vector, and target motion, the injection pose is calculated:

[0045] The center position coordinates are used to form the approach point of the end syringe, and the normal vector is used to form the orientation of the end syringe. The reserved distance between the end syringe and the injectable area is corrected according to the propulsion amount or call state in the target motion to obtain the injection pose.

[0046] The injection posture includes the positional and orientational relationships of the end syringe.

[0047] Furthermore, the robotic arm control process includes:

[0048] The injection pose and the target motion are obtained, and the robotic arm controller executes the approach action, alignment action and contact determination action.

[0049] The approach action is that the six-degree-of-freedom robotic arm moves towards the injectable area according to the orientation relationship of the injection pose;

[0050] The alignment action involves adjusting the orientation of the end syringe according to the normal vector, so that the end syringe is aligned with the direction of the local surface.

[0051] The contact determination action is to continuously read the force information between the end syringe and the injectable area through a six-dimensional force sensor. When the contact force enters the set threshold range, the approach is stopped and the current position is maintained. The contact force is written into the current identification record.

[0052] If the contact force changes too much during the approach process, it is recorded as an abnormal contact record, causing the six-degree-of-freedom robotic arm to return to the initial pose and retain the injection pose and target motion for the next call.

[0053] Once the contact force reaches a set threshold, the automatic injection device is triggered for processing.

[0054] For the push cylinder and syringe push block structure, the target motion amount is called to drive the push cylinder to move according to the push amount, which drives the syringe push block to push the continuous syringe to complete the injection action.

[0055] For the first feeding mechanism and the injection component structure, after the contact force reaches the threshold, the injection component is first moved at the first speed. After the distal end contacts the injectable area, the action continues at the second speed. When the marking ring is flush with the body surface, the injection action is performed.

[0056] After the injection is completed, injection information is generated, which includes ear tag information, injection execution status, and injection action information.

[0057] Furthermore, the process of uploading, distributing, and processing feedback includes:

[0058] The gateway acquires the ear tag information, body shape recognition result, dosage information, and injection information, and sends the current recognition record to the cloud platform.

[0059] The cloud platform locates the corresponding dairy cow's data record based on the ear tag information and writes in the body size recognition results, dosage information, and injection information;

[0060] The cloud platform distributes the updated records to external platforms, receives feedback status, and generates updated historical injection information.

[0061] The historical injection information includes the ear tag information, body shape recognition result, dosage information, and injection information corresponding to this injection, and serves as the source for calling historical injection information in the next combination processing of objects and additional conditions.

[0062] Furthermore, an intelligent immune injection robot based on body shape recognition includes: an identity recognition and image acquisition module, a first recognition module, a motion state tracking, recognition and combined processing module, a dosage information output module, a target motion amount and injection posture generation module, a robotic arm control and automatic injection module, and an upload and distribution feedback module; the modules are connected in sequence to implement the method described in any of the above-mentioned methods.

[0063] The key innovations of this invention include:

[0064] (1) Based on the ear tag information, the body shape recognition result, the vaccine to be injected information, the historical injection information and the static state, first perform the combination processing of the object and the additional conditions, and then perform the corresponding task processing based on the combination of the object and the additional conditions to obtain the dosage information. Then, the body shape recognition result is connected to the dosage information generation link by the first recognition processing link.

[0065] (2) Based on the dosage information, the push cylinder, syringe push block and first feeding mechanism are processed to obtain the target motion amount. Then, based on the target motion amount, the second identification processing and injection pose calculation processing are performed to convert the dosage information into the control amount of the automatic injection device, and the target motion amount is placed in the generation link of the center position coordinate, normal vector and injection pose.

[0066] (3) Upload, distribute and feedback processing is performed on the ear tag information, the body shape recognition result, the dosage information and the injection information to obtain historical injection information, and the historical injection information is sent back to the combined processing link of object and additional conditions, so that the first recognition processing, the execution of corresponding task processing, the automatic injection device processing and the historical injection information update form a continuous processing relationship.

[0067] The following are its main beneficial effects:

[0068] (1) To address the problem of the disconnect between body shape recognition results and dosage information in the existing scheme, the body shape recognition results, vaccine information to be injected, historical injection information and static state are incorporated into the combination processing of object and additional conditions, and the dosage information is output by the corresponding task processing, so that the formation of dosage information no longer depends solely on the vaccine information to be injected, and the dosage information maintains a corresponding relationship with the current cow identification record.

[0069] (2) In view of the problem that the dosage information is separated from the action of the automatic injection device in the existing scheme, the target motion amount is obtained by processing the push cylinder, the syringe push block and the first feeding mechanism, so that the dosage information is directly entered into the automatic injection device processing link. The target motion amount maintains a continuous calling relationship with the subsequent robotic arm control processing and automatic injection device processing.

[0070] (3) In response to the existing solution's processing path of first performing the second identification process and injection pose calculation process, and then determining the injection execution amount, the target motion amount is generated before the second identification process, so that the center position coordinates, normal vector and injection pose are established under the existing execution amount conditions, and the injection pose and the target motion amount are jointly entered into the robotic arm control process.

[0071] (4) To address the problem of the fragmented historical injection information and current injection processing links in the existing scheme, the ear tag information, the body shape recognition result, the dosage information and the injection information are uniformly entered into the upload, distribution and feedback processing to obtain historical injection information, and then sent back to the combined processing of the object and additional conditions, so that the previous record and the current record of the same cow can be called in the same link.

[0072] (5) To address the inconsistency in the calling relationships between image acquisition and processing, first recognition processing, execution of corresponding tasks, robotic arm control processing, automatic injection device processing, and uploading, distribution, and feedback processing in the existing scheme, the static state, the dosage information, the target amount of motion, the injection posture, and the historical injection information are connected in sequence to ensure that the acquisition, judgment, control, and recording of dairy cow immunization injections maintain a consistent process. Attached Figure Description

[0073] Figure 1 A flowchart illustrating an intelligent immune injection dosage control method based on body shape recognition, provided for an embodiment of this application;

[0074] Figure 2 This is a structural block diagram of an intelligent immune injection robot based on body shape recognition, provided as an embodiment of this application. Detailed Implementation

[0075] Example 1: Refer to Figure 1 This is a flowchart illustrating a smart immune injection dosage control method based on body shape recognition provided in an embodiment of the present invention. The process may include at least steps S100-S400:

[0076] S100: Acquire ear tag information, RGB data and point cloud data, perform first recognition processing, and obtain body shape recognition results;

[0077] S200. Based on the body shape recognition result, perform motion state tracking and recognition, object and additional condition combination processing, and execute corresponding task processing to obtain dose information;

[0078] S300: Based on the dosage information, perform processing of the pushing cylinder, syringe pusher and first feeding mechanism and injection posture calculation to obtain the injection posture;

[0079] S400. Based on the injection pose, perform robotic arm control processing and execute upload, distribution and feedback processing to obtain historical injection information.

[0080] Step S100 includes at least steps S110-S130:

[0081] S110. Acquire ear tag information, and the image acquisition unit acquires RGB data and point cloud data of the cow's back, side and rump areas, performs image acquisition processing, and obtains RGB data and point cloud data.

[0082] Specifically, the input sources for this step are individual cows entering a fixed location and RFID identification receivers and image acquisition units set up around the fixed location.

[0083] The ear tag information is individual identification information read by a radio frequency identification (RFID) receiver, and includes at least ID information for distinguishing individual cows; the image acquisition unit is an image acquisition device arranged above and to the side of the fixed position, preferably including one or a combination of RGB-D camera and binocular camera, the RGB data is red, green and blue (RGB) data, and the point cloud data is a set of spatial points formed by unfolding the depth image in 3D coordinate space.

[0084] In practice, after the cow enters the fixed location, the RFID identification receiver first performs a receiving operation, reads the ID information from the ear tag, and writes the ID information into the current acquisition task. Subsequently, the image acquisition unit performs simultaneous acquisition on the cow's back, side, and rump regions. The back region is used to acquire the contour data of the upper surface of the torso, the side region is used to acquire the side profile data of the torso, and the rump region is used to acquire the rear boundary data. The image acquisition processing includes camera activation, acquisition sequence alignment, image frame filtering, and spatial coordinate association.

[0085] The timing alignment of the acquisition here refers to grouping the image frames corresponding to the back, side, and rump regions into the same acquisition batch; the image frame filtering refers to removing image frames that are not fully in the fixed position, whose body areas are blocked by fences, or whose image brightness is abnormal; the spatial coordinate association refers to matching the spatial position in the depth image with the pixel position in the RGB data to obtain the color appearance information and spatial contour information of the same part.

[0086] Furthermore, in an engineering embodiment, a binocular camera can be installed on one side of the cow passage, and an RGB-D camera can be installed above the passage, both facing the fixed position. When the RFID identification receiver receives the ear tag information, the image acquisition unit continuously acquires multiple frames of images within one acquisition cycle, and selects image frames of the cow's back, side, and rump regions that are complete, unobstructed, and have continuous spatial contours to form the current cow's RGB data and point cloud data. If the cow's head or torso is detected to have left the fixed position within the acquisition cycle, the current acquisition task is recorded as incomplete, and the image acquisition unit waits for the next ear tag information trigger.

[0087] After this step is completed, the obtained data is recorded as the output field name "RGB data and point cloud data" and used as the direct input for S120; at the same time, the ear tag information participates in the combination of objects and additional conditions in subsequent S220, and participates in the uploading, distribution and feedback processing in S430.

[0088] S120. Based on the RGB data and point cloud data, perform the first recognition process to obtain a three-dimensional model.

[0089] Specifically, the input source for this step is the "RGB data and point cloud data" output by S110. Here, the first recognition is a preliminary recognition step that is different from the subsequent second recognition. The first recognition does not locate the injectable area, but rather performs recognition around the overall shape of the cow. The three-dimensional model is an overall shape model formed in 3D coordinate space based on the RGB data and point cloud data, which at least includes the outline of the cow's back, side, and rump regions in the same spatial coordinates.

[0090] In practice, the control unit first performs registration processing on the RGB data and point cloud data, placing the back region, side region, and rump region data within the same acquisition batch into the same coordinate system; then, it performs cow body region filtering, removing fences, ground, passage edges, and non-cow regions from the point cloud data, retaining the main cow torso region; then, it performs shape stitching, connecting the upper surface contour of the back region, the side contour of the side region, and the rear boundary of the rump region into a continuous shape.

[0091] The registration process here involves aligning the point cloud data obtained from different image acquisition devices with the RGB data; the cow body region filtering involves selecting the main body region of the cow based on the continuity of spatial position and color region; and the shape stitching involves stitching the aligned data into a complete shape in three-dimensional space.

[0092] Furthermore, the first identification also includes unifying the torso orientation. Specifically, it determines the cow's current orientation based on the relative position of the side and rump regions, and unifies the back, side, and rump regions under the same orientation representation to avoid confusion in the front-to-back orientation of the same cow in different acquisition batches. In cases of partial occlusion, the control unit can select adjacent image frames from the same acquisition cycle to fill in missing areas, ensuring continuity of the 3D model at back corners, side edges, and rump boundaries.

[0093] Understandably, the 3D model is not simply a result of image stitching, but rather a holistic shape carrier containing spatial coordinate relationships. The key point information, geometric model, and pose information in subsequent steps S130 are all built upon this 3D model. In a real-world scenario, when a cow stands in a fixed position, the image acquisition unit first obtains RGB data and point cloud data from multiple perspectives. The control unit then aligns, filters, and stitches the data within the current acquisition batch to form a 3D model corresponding to the current cow's torso shape. If the point cloud data in the current acquisition batch has large-area breaks, the control unit marks that batch as an abnormal acquisition record and re-acquires the next batch of RGB data and point cloud data acquired in step S110.

[0094] After this step is completed, the three-dimensional model is recorded as the output field name "three-dimensional model" and called as the input field "the three-dimensional model" of S130; at the same time, the spatial coordinate relationship carried by the three-dimensional model also provides the initial shape basis for the second recognition in S320 and the injection pose calculation in S330.

[0095] S130. Based on the three-dimensional model, key point information, geometric model and pose information are processed to obtain body shape recognition results.

[0096] Specifically, the input source for this step is the "3D model" output by S120. The key point information is representative spatial point information extracted from the 3D model for the back, side, and rump regions, including at least spatial points at the back contour transitions, side contour boundaries, and rump region boundaries. The geometric model is a torso shape relationship model formed based on the key point information, used to represent the geometric correspondence between the back, side, and rump regions. The pose information is the cow's spatial orientation, torso tilt, and overall position in a fixed location. The body shape recognition result is a first recognition result formed by the key point information, geometric model, and pose information, serving as the initial input for subsequent dose information generation.

[0097] In practice, the control unit first extracts key point information from the three-dimensional model: key points of the upper surface contour in the back region, key points of the side boundary in the side region, and key points of the rear boundary in the rump region. Then, a geometric model is established based on these key point information. The geometric model includes the length relationship of the back region, the height relationship of the side region, and the width relationship of the rump region, and is used to represent the overall composition of the current cow's torso shape. Then, pose information is generated based on the distribution of the geometric model in a fixed position. The pose information at least distinguishes whether the cow has deflected, tilted to the side, or experienced significant forward or backward displacement.

[0098] Furthermore, the body shape recognition result is not a single key point or a single geometric dimension, but rather an overall recognition result obtained by combining the key point information, geometric model, and pose information. This result can be saved using body shape category, body shape range, or body shape classification, but in this step, it is uniformly recorded as the "Body Shape Recognition Result" field.

[0099] When the cow is in a slightly swaying scenario, the control unit first compares adjacent 3D models in the same batch of data acquisition. If the positional changes of key point information are within an acceptable range, geometric model and pose information processing continues. If the changes exceed the recording range, the current 3D model is marked as a model to be reviewed, and the 3D models from the previous batch are retained for the next image acquisition process. This processing path ensures that the body shape recognition result is seamlessly integrated with the motion state tracking and recognition in S210. The former provides the basis for overall shape recognition, while the latter further screens this recognition basis over time.

[0100] Furthermore, in a practical engineering embodiment, after the ear tag information is triggered, the cow in the fixed position channel completes the acquisition of the back, side and rump areas. The control unit continuously extracts key point information based on the three-dimensional model and forms a geometric model and pose information corresponding to the current cow. When the cow's body posture is stable, the corresponding body shape recognition result is output.

[0101] The body shape recognition result is one of the essential minimum sets required for improvement in this specification, as the combination of objects and additional conditions in S220, the dose information in S230, and the target motion in S310 all rely on this field as a prerequisite. The number of acquisition batches, the number of supplementary image frames, and the specific combination of image acquisition devices are preferred extensions. After this step, the body shape recognition result is recorded as the output field name "Body Shape Recognition Result" and is directly input in S210. It also participates in the combination of objects and additional conditions in S220 and in uploading, distribution, and feedback processing in S430. Through continuous processing from S110 to S130, ear tag information, RGB data, and point cloud data enter the same processing chain. The 3D model serves as an intermediate carrier, handling image acquisition processing and processing of key point information, geometric models, and pose information, ultimately forming a body shape recognition result that can be directly called upon in subsequent dose information processing.

[0102] In summary, this step adds a continuous link between the initial image acquisition and processing and the subsequent dose information processing. This link includes 3D models, key point information, geometric models, and pose information. Body shape recognition results no longer remain at the level of local image judgment but become a combination of subsequent objects and additional conditions, and a direct input for dose information processing. Compared to processing methods that only identify the injectable area, the body shape recognition results output by this step have a complete source of the torso shape, clearer subsequent calling relationships, and a more complete recording path during the process.

[0103] Step S200 includes at least steps S210-S230:

[0104] S210. Obtain the body shape recognition result, perform motion state tracking and recognition processing, and obtain the static state.

[0105] Specifically, the input source for this step is the body shape recognition result output by S130. This body shape recognition result already includes key point information, geometric models, and pose information for the cow's back, side, and rump regions in the current acquisition batch. Therefore, this step does not re-acquire images; instead, the data processing unit directly calls the body shape recognition result to perform motion state tracking and recognition. Here, motion state tracking and recognition refers to comparing and recognizing consecutive acquisition batches corresponding to the same ear tag information. The data processing unit reads the body shape recognition result of the current acquisition batch and retrieves the body shape recognition result of the same ear tag information from the database in the previous acquisition batch, then compares the key point spatial location information and pose information at different times item by item.

[0106] In specific processing, the data processing unit first checks whether there is an obvious non-standing posture in the current body shape recognition result. The non-standing posture includes excessive back undulation, obvious tilting of the side area, or abnormal rotation of the buttock area. If the non-standing posture exists, the current cow is recorded as non-stationary and the state is written into the current recognition record.

[0107] If the non-standing posture does not exist, the spatial location information of key points in the current batch and the previous batch is compared. The average spatial geometric distance of key points in the back, side and hip regions is calculated and compared with a preset distance threshold.

[0108] When the average spatial geometric distance is less than the distance threshold, the data processing unit determines that the cow is currently stationary in a fixed position; when the average spatial geometric distance is greater than or equal to the distance threshold, the data processing unit determines that the cow is still in motion and retains the current body shape recognition result in the database, waiting for the next collection batch to participate in the comparison again.

[0109] Furthermore, to ensure the continuous operation of this step in an engineering scenario, the data processing unit establishes a current identification record for the cow each time ear tag information is received. This current identification record includes at least the body shape identification result, the status of the body shape identification result from the previous batch, and the current motion tracking and identification result. When the cow briefly sways within a fixed position, the motion tracking and identification does not immediately interrupt the current identification record but retains the current ear tag information and continues to retrieve it in the next batch. When the cow leaves the fixed position, the current identification record is closed.

[0110] Understandably, the static state is not information manually input, but a state field automatically output by the data processing unit based on the body shape recognition result.

[0111] After this step is completed, the static state is recorded as the output field name "Status State" and serves as the direct input for the combined processing of objects and additional conditions in S220. At the same time, the static state is also closely related to the execution of the corresponding task processing in the subsequent S230. Only when the current cow is in the static state will the subsequent dosage information enter the effective processing link.

[0112] S220. Based on the ear tag information, the body shape recognition result, the vaccine to be injected information, the historical injection information, and the static state, perform combination processing of the object and additional conditions to obtain the combination of the object and additional conditions.

[0113] Specifically, the input sources for this step include the ear tag information received in S110, the body shape recognition result output in S130, the stationary state output in S210, and the vaccine information to be injected and historical injection information retrieved from the cloud platform or an external platform. The vaccine information to be injected is the current injection task information corresponding to the ear tag information, and at least includes the vaccine information that the cow needs to receive at the current time; the historical injection information is the past injection information already saved in the cloud platform or external platform, and at least includes the injection records corresponding to the ear tag information.

[0114] The combination of the object and the additional conditions refers to the combined fields after merging the object information and the additional condition information according to the same cow and the same current identification record; in this step, the object is the individual cow corresponding to the ear tag information, and the additional conditions include the body shape identification result, the vaccine to be injected information, the historical injection information, and the resting state.

[0115] In specific processing, the data processing unit first accesses the cloud platform or external platform based on the ear tag information, reads the vaccine information to be injected for the current cow, and then reads the existing historical injection information of the cow; then, it associates the body shape recognition result with the vaccine information to be injected in the same current recognition record, and checks whether the static state is in a state that can continue to be processed.

[0116] If the static state is static, the data processing unit continues to incorporate the historical injection information into the current identification record; if the static state is not static, the data processing unit can still generate a combination of object and additional conditions, but will retain the static state identifier in the combination field for S230 to make a corresponding judgment when performing the corresponding task processing. Furthermore, the combination processing in this step is not a simple splicing, but rather a one-to-one correspondence processing according to the same individual cow pointed to by the ear tag information.

[0117] The data processing unit first checks the individual identifier in the ear tag information and the vaccine information to be injected, and then checks the individual identifier in the ear tag information and the historical injection information. If the two are inconsistent, the current combination processing is recorded as an abnormal record, and the current identification record of the current cow is not entered into S230. If the two are consistent, the data processing unit writes the body shape recognition result, the vaccine information to be injected, the historical injection information, and the static state into the same combination field.

[0118] In specific project implementation, after a cow enters a fixed position and completes S210, the cloud platform returns the cow's vaccination information for that day, and the external platform returns the cow's recent historical vaccination information. The data processing unit combines this information with the cow's body shape recognition result and stationary state into the same current recognition record, forming a combination of object and additional conditions. After the combination of object and additional conditions is completed, it is recorded as the output field name "Combination of Object and Additional Conditions" and directly input into S230. At the same time, the historical vaccination information will be updated after S430 completes the upload, distribution, and feedback processing and will return to this step to participate in new combination processing, thus forming a cross-step connection between this main step and S400.

[0119] S230. Based on the combination of the object and additional conditions, perform corresponding task processing to obtain dose information.

[0120] Specifically, the input source for this step is the combination of the object output by S220 and the additional conditions. The corresponding task processing is performed by the data processing unit, which calls existing associated information from the database, cloud platform, and external platform to determine each item of the combination of the object and additional conditions, and outputs the dosage information corresponding to the current cow. This dosage information refers to the injection volume that the current cow can receive from the automatic injection device under the current identification record; it is a pre-input field for the subsequent generation of the target movement amount in S310.

[0121] In specific processing, the data processing unit first reads the static state from the combination of the object and additional conditions. If the static state indicates that the cow is in a processable state, it continues to read the vaccine information to be injected and historical injection information. If the static state indicates that the cow is still in motion, the data processing unit retains the current identification record in the database and records the result of the corresponding task as a pending state, waiting to be called again after the next S210 and S220. For the current cow in a processable state, the data processing unit further checks whether there is an injection record in the historical injection information that is the same as the vaccine information to be injected. If there is no corresponding injection record, the data processing unit generates the current dosage information based on the vaccine information to be injected and the body shape identification result. If there is already a corresponding injection record, the data processing unit outputs the dosage information that will not be repeated based on the historical injection information and writes the corresponding injection information status into the current identification record.

[0122] Furthermore, the implementation path of "generating current dosage information based on the vaccine information to be injected and body shape recognition results" in this step includes the following: The data processing unit first calls the current task item in the vaccine information to be injected, and then reads the body shape recognition results in the combination of object and additional conditions; subsequently, according to the pre-stored correspondence in the database, it matches the current task items corresponding to different body shape recognition results to form the dosage information for the current cow. The correspondence in this embodiment is one of the smallest sets for realizing the core improvement of this invention, because the processing chain of this invention does not perform a second identification before direct injection, but rather forms the dosage information first in S230, and then generates the target exercise amount based on the dosage information in S310.

[0123] Understandably, the dosage information is not an incidental field in this step, but a core output field in the current identification record; once the dosage information is generated, it is bound and saved with the ear tag information and body shape recognition results, and is called by "the dosage information" in S310 to continue the processing of the cylinder, syringe push block and first feed mechanism.

[0124] In a specific engineering embodiment, after a cow enters a fixed position and completes S130, the data processing unit determines in S210 that the cow is in a stationary state, and then in S220, it retrieves the vaccine information to be injected that day and the cow's existing historical injection information. If the historical injection information shows that the cow has not yet completed the injection corresponding to the current vaccine information, the data processing unit reads the cow's body shape recognition result and generates the current dosage information. Subsequently, this dosage information is transmitted to S310 to participate in the generation of target exercise volume. If the historical injection information shows that there is already an injection record corresponding to the current vaccine information, the data processing unit outputs dosage information that will not be repeated, and retains the current ear tag information, body shape recognition result, and injection information status for unified uploading in S430.

[0125] After this step is completed, the dosage information is recorded as the output field name "dosage information", and its direct input position is "the dosage information" in S310; at the same time, the dosage information also participates in the historical injection information update as part of the upload, distribution and feedback processing in S430.

[0126] Summary of the technical effects of this step: This step integrates body shape recognition results, static state, information on the vaccine to be injected, and historical injection information into a single processing loop. Dosage information is directly output by executing the corresponding task and is recorded synchronously with ear tag information. Compared to processing methods that directly trigger the injection action based solely on the information on the vaccine to be injected, this step places the dosage information before the second recognition and injection posture calculation, making the source of subsequent target motion more explicit and the field relationships in the current recognition record more complete.

[0127] In one specific embodiment, based on the body shape recognition result, motion state tracking and recognition processing, object and additional condition combination processing, and corresponding task processing are first performed to obtain dosage information. This step first receives the body shape recognition result output by S130, which includes key point information, geometric model, and pose information of the current cow in the back, side, and rump regions. The data processing unit compares and identifies consecutive batches corresponding to the same ear tag information, retrieves the body shape recognition result of the previous batch from the database, and compares the key point spatial location information and pose information at different times item by item. To quantify the degree of spatial displacement of the cow's torso between the current batch and the previous batch, the key point set of the current batch is first defined as... The key points set of the previous batch are Each of the key points These are three-dimensional spatial coordinates. Formula ① calculates the average spatial geometric distance of all corresponding keypoints:

[0128]

[0129] in:

[0130] : Average spatial geometric distance, representing the overall displacement of the cow's torso between two data collections;

[0131] : Total number of key points, ranging from 9 to 15, derived from key point information in the body shape recognition results output by S130;

[0132] : The summation index variable, with a value range of 1 to 1. ;

[0133] The first in the current batch of data collection The X-axis coordinates of each key point are derived from the current body shape recognition result;

[0134] The first in the current batch of data collection The Y-axis coordinates of each key point are derived from the current body shape recognition result;

[0135] The first in the current batch of data collection The Z-axis coordinates of each key point are derived from the current body shape recognition result;

[0136] The first ear tag information corresponding to the previous batch of data collection. The X-axis coordinates of each key point are derived from historical body shape recognition results retrieved from the database.

[0137] The first ear tag information corresponding to the previous batch of data collection. The Y-axis coordinates of each key point are derived from historical body shape recognition results retrieved from the database.

[0138] The first ear tag information corresponding to the previous batch of data collection. The Z-axis coordinates of the key points are derived from historical body shape recognition results retrieved from the database.

[0139] Data Source → Indicator → Variable Mapping: Extract "Spatial Location Information of Key Points in the Current Batch" from the "Body Shape Recognition Result" of S130, denoted as... The "spatial location information of key points in the previous batch" is retrieved from the database and recorded as follows: Together, they form the formula ① .

[0140] Simple numerical example: Let (Simplified illustration) The coordinates of the key points in the current batch are as follows: , , The coordinates of the corresponding points in the previous batch are , , The Euclidean distances at each point are calculated as follows: , , average distance mm.

[0141] The data processing unit is pre-set with a distance threshold. (Typical value is 5mm, derived from field calibration experiments). Formula ② is based on... and The comparison results and the static state flag output by the non-standing posture check:

[0142]

[0143] in:

[0144] : Static state flag, 1 indicates static (processing can continue), 0 indicates dynamic state (pending processing).

[0145] Non-standing posture indicator: Boolean value, 1 indicates the presence of a non-standing posture (excessive back undulation, obvious side tilting or abnormal hip rotation), 0 indicates normal standing, the data source is the geometric model and pose information in the body shape recognition results of S130;

[0146] : The average spatial geometric distance calculated by formula ①;

[0147] : Preset distance threshold.

[0148] Obtained from formula ① Substitute this into the comparison condition of formula ②. Simple numerical example: If the non-standing posture marker = 0, and mm< mm, then It is determined to be in a static state.

[0149] like mm or in a non-standing posture, then After the determination is completed, the data processing unit establishes a current recognition record for the current ear tag information. The record includes the body shape recognition result, the recall status of the previous acquisition batch, and the current motion state tracking and recognition result (i.e., ), and will The output is a field named "Quiet State". This Quiet State serves as the direct input to the "Combined Processing of Object and Additional Conditions" in S220, and is also used for the enable judgment of the dose information processing link in S230.

[0150] Furthermore, following the static state output by S210, this step further acquires ear tag information, body shape recognition results, vaccine information to be injected, and historical injection information. The vaccine information to be injected is returned by the cloud platform based on the ear tag information, while the historical injection information is returned by an external platform. The data processing unit first verifies the consistency between the ear tag information and the individual identifiers in the vaccine information to be injected and the historical injection information; if they match, they are combined. To quantify the completeness and decision-making capability of each additional condition field in the current identification record, a contextual integrity scoring function is defined. Assume the current identification record contains five fields: ear tag information... Body shape recognition results static state Information on vaccines to be administered Historical injection information Formula ③ calculates the integrity score. :

[0151]

[0152] in:

[0153] Context integrity score, with a value range of [0,1];

[0154] : Field validity indicator function. It returns 1 when the input field exists and is valid (not empty, correctly formatted, and logically consistent), and 0 otherwise.

[0155] The ear tag information field, sourced from the S110's RFID receiver, contains a unique ID string;

[0156] The body shape recognition result field, derived from S130, contains key points, geometric models, and pose information;

[0157] The static state field is derived from the output of formula ②. ;

[0158] The vaccine information field, sourced from the cloud platform, includes the vaccine name, standard dosage range, and batch number.

[0159] The historical injection information field, sourced from an external platform, includes the time, vaccine name, and dosage of the injection record.

[0160] Data source → Indicator → Variable mapping: The validity of the field extracted from the "Body Shape Recognition Result" of S130 is denoted as... The validity of the field extracted from the "static state" in formula ② is denoted as follows. The validity of the fields extracted from the "Vaccines to be Injected" information on the cloud platform is recorded as follows: The validity of the fields extracted from the "historical injection information" of the external platform is recorded as follows: The validity of the field extracted from the "ear tag information" of S110 is denoted as Together they form the formula ③ .

[0161] Simple numerical example: If ear tag information exists (1), body shape recognition result exists (1), static state is valid (1), vaccine information to be injected exists (1), historical injection information exists but dosage field is missing (0), then .

[0162] when At this point, the data processing unit determines that the context is complete and performs combined vector construction. Each additional condition field is mapped to a numerical component, forming a combined vector of the object and the additional conditions. Formula ④ defines the construction of combined vectors:

[0163]

[0164] in:

[0165] : A combined vector of objects and additional conditions, in column vector form (superscript T indicates transpose);

[0166] Body shape quantification coefficient, obtained by weighted averaging of geometric models (back length, side height, hip width) in the body shape recognition results, with a value range of [0.5, 1.5], derived from the body shape recognition results of S130;

[0167] The standard dose coefficient for vaccines is obtained by normalizing the recommended dose in the information on vaccines to be injected, with a value range of [0.8, 1.2], and is derived from the cloud platform;

[0168] Historical injection constraint coefficient: 0 if the same vaccine as the current vaccine already exists in the historical injection information, otherwise 1. It is derived from an external platform.

[0169] : The static state value (0 or 1) is derived from formula ②;

[0170] The transpose operator converts a row vector into a column vector.

[0171] Simple numerical example: The result of a cow's body shape recognition was calculated as follows (Large body size), standard dose coefficient of vaccine to be injected Historical vaccination records show that this vaccine has not been administered. static state Then the combined vector If the static state is 0, then even if other fields are complete, the combined vector will contain... Subsequently, S230 will mark this as pending processing. After construction is complete, the ear tag information, body shape recognition result, vaccine information to be injected, historical injection information, and static state are merged into the same current recognition record, and the output field name is "Combination of object and additional conditions". This combination vector serves as the input for "Executing the corresponding task processing" in S230.

[0172] Furthermore, this step receives the combination of the object output by S220 and the additional conditions, and parses out the combination vector from it. The data processing unit first examines the static state component and its components, as well as the vaccine information to be injected and historical injection information in the original fields. If the value is 0, the identification record is marked as pending and the process returns to wait for the next acquisition; if the value is 1, dose generation continues.

[0173] To map the body size quantification coefficient and the standard vaccine dose coefficient to specific dose values, while taking into account historical injection constraints, a dose information generation function is designed.

[0174] Set the baseline dose The recommended dose for standard-sized dairy cows in the vaccine information (sourced from the cloud platform). Formula ⑤ calculates the initial dose. :

[0175]

[0176] in:

[0177] Initial dose;

[0178] The basic dose is directly read from the standard dose field in the vaccine information to be injected, and the value range is generally 1~5mL;

[0179] Body shape influence coefficient, set to 0.5, preset by empirical rules;

[0180] Body size quantification coefficient (same as formula ④);

[0181] : Standard dose coefficient for vaccines (same as formula ④).

[0182] Data Source → Metrics → Variable Mapping: Extraction of vaccine information to be injected from the "Combination of Objects and Additional Conditions" output by S220. , by the same combination vector and Substitute the components into formula ⑤.

[0183] Simple numerical example: Let mL, , , ,but mL.

[0184] Then, the data processing unit reads the historical injection constraint coefficients. .like (i.e., the same vaccine has been administered in historical records), then the final dosage information Setting it directly to 0 indicates that the injection will not be repeated; if The final dose information is equal to the preliminary dose, and can be rounded or adjusted for precision depending on the vaccine type. Formula ⑥ determines the final dose information:

[0185]

[0186] in:

[0187] Final dosage information, ranging from 0 to 5 mL;

[0188] Historical injection constraint coefficient (same as formula ④);

[0189] : Rounding function with specified precision. The first parameter is the value to be rounded, and the second parameter is the rounding precision.

[0190] : The initial dose calculated by formula ⑤;

[0191] Rounding precision: 0.1 mL, corresponding to the smallest graduation on the syringe.

[0192] Simple numerical examples: , mL, then mL; if ,but mL. After generating the dose information, the data processing unit will... The information is bound and saved with ear tag information and body shape recognition results, and the output field is named "dosage information". This dosage information is directly used as input for "processing of pushing cylinder, syringe push block and first feed mechanism" in S310, and is also used to update historical injection information in the upload, distribution and feedback processing of S430.

[0193] Engineering Example: In a smart immunization system at a ranch, a Holstein cow enters a fixed position. RFID reads the ear tag information "CN00123". The body shape recognition result output in S130 shows a back length of 145cm, a side height of 82cm, and a hip width of 55cm, classifying the geometric model as "large". The posture information shows no trunk tilt. In S210, the body shape recognition results from the previous batch are retrieved, and the average spatial geometric distance is calculated to be 2.8mm, less than the threshold of 5mm, and there is no non-standing posture, so the static state is output as 1. In S220, the vaccine information to be injected is obtained from the cloud platform as "foot-and-mouth disease vaccine, standard dose 2.0mL". Historical injection information obtained from an external platform shows that the cow has not been vaccinated with this vaccine in the past three months, and the context integrity score C_ctx=1.0. A combined vector is constructed: body shape quantification coefficient α_size=1.25 (based on body shape lookup table), vaccine coefficient β_vac=1.0, γ_hist=1, S_static=1. In S230, the base dose D_base = 2.0 mL, and the calculated dose D_pre = 2.0 × (1 + 0.5 × 0.25) × 1.0 = 2.25 mL, rounded to a precision of 0.1, yields D_final = 2.3 mL. The final output dose information "2.3 mL" is used by S310 to generate the target exercise volume. If the historical injection information shows that the injection has been completed, then D_final = 0, and the system skips the injection.

[0194] This section summarizes the technical effects: Through three consecutive sub-steps, the body shape recognition results, static state, information on the vaccine to be injected, and historical injection information are integrated into a dynamic decision-making context. By utilizing the average spatial geometric distance threshold and the dose mapping formula, automatic identification of static state and quantitative generation of personalized dose information are achieved, forming a complete closed loop from perception to decision-making.

[0195] Step S300 includes at least steps S310-S330:

[0196] S310. Obtain the dosage information, and process it by pushing the cylinder, injector pusher and first feeding mechanism to obtain the target motion amount.

[0197] Specifically, the input source for this step is the dosage information output by S230. The dosage information is the current cow injection dosage information generated by the data processing unit after performing the corresponding task processing. It has been bound to the ear tag information, body shape recognition results, and vaccine information to be injected. Therefore, this step does not re-determine whether the current cow needs to be injected, but directly processes the execution components of the automatic injection device.

[0198] The push cylinder is a driving component in the automatic injection device, used to provide linear propulsion; the syringe pusher is a transmission component connected to a continuous syringe or a needleless syringe, used to receive the propulsion action of the push cylinder; the first feeding mechanism is a feeding component that drives the injection piece to move forward, used to perform approach and feeding after the injection posture is determined.

[0199] The target motion quantity is an execution quantity field calculated from the dosage information, and includes at least the propulsion amount of the push cylinder, the motion amount of the syringe pusher, and the call status of the first feeding mechanism. In specific implementation, the control unit first reads the current task item corresponding to the dosage information, and then determines the execution path based on the current structure of the automatic injection device. When the automatic injection device uses a structure where the push cylinder and syringe pusher cooperate, the control unit converts the dosage information into the propulsion amount of the push cylinder and simultaneously converts it into the motion amount of the syringe pusher. When the automatic injection device uses a structure where the first feeding mechanism and the injection unit cooperate, the control unit writes the dosage information into the call record of the first feeding mechanism and forms the corresponding target motion quantity. This conversion process is not simply writing an injection volume value, but rather implementing the dosage information into the action field of a specific component.

[0200] Furthermore, when the control unit executes the processes of the pushing cylinder, syringe pusher, and first feeding mechanism, it also checks the historical injection information retrieval status corresponding to the current ear tag information. If the historical injection information indicates that the previous identification record has completed the injection corresponding to the same vaccine information to be injected, the current target motion is recorded as closed, and the current record is retained for unified feedback by S430; if the historical injection information indicates that the cow has not yet completed the corresponding injection, the control unit continues to generate a valid target motion. Understandably, the target motion is a pre-field before the subsequent second identification. This step first implements the dosage information to the execution component of the automatic injection device, and then enters the positioning link of the injectable area. This is different from the processing order of first determining the injection posture and then unifying the injection.

[0201] In the engineering embodiment, after the cow enters the fixed position and completes S230, the control unit reads the cow's dosage information. If a continuous syringe structure is currently used, the propulsion amount of the cylinder is first generated, and then the motion amount of the syringe pusher is generated. If a needleless syringe structure is currently used, the corresponding target motion amount is generated and written into the first feed mechanism call record. After this step is completed, the target motion amount is recorded as the output field name "target motion amount" and used as the direct input of S320. It will also be used as "the target motion amount" in S410 along with the injection posture to participate in the robotic arm control processing.

[0202] S320. Based on the target motion, perform a second identification process to obtain the center position coordinates and normal vector.

[0203] Specifically, the input source for this step is the target motion quantity output by S310, while simultaneously using the 3D model formed by S120 and the body shape recognition result formed by S130 as the basis for the initial shape. The second recognition is a local recognition step distinct from the first recognition. The second recognition no longer focuses on the overall body shape of the cow, but rather on locating the injectable area of ​​the current cow; the center position coordinates are the center spatial coordinates of the current injectable area in the 3D model; the normal vector is the orientation information of the local plane of the current injectable area, used to characterize the orientation when the automatic injection device approaches the injectable area. In specific processing, the control unit first checks whether the target motion quantity is in a valid state.

[0204] If the target motion is off, the second identification only generates a recording result and does not proceed to subsequent injection pose calculation. If the target motion is active, the image acquisition unit or visual positioning camera retrieves data from the cow's back, side, and rump regions, and combines this data with the 3D model and body shape recognition results to re-search for injectable areas within a local range. Here, the injectable area refers to the body surface area that the automated injection device can approach and perform the injection action.

[0205] In an embodiment of the present invention, the second identification preferably revolves around the adjacent positions of the posterior side and buttock regions. First, a local area is extracted from the current 3D model. Then, candidate regions are selected based on the continuity of local boundaries and the smoothness of the surface. Finally, the injectable region is determined from the candidate regions. Further, the control unit performs center extraction and local plane fitting on the injectable region. The center extraction finds the center position from the boundary of the injectable region, forming center position coordinates. The local plane fitting obtains the local surface orientation based on the point cloud data around the center position coordinates, forming a normal vector.

[0206] The center position coordinates and normal vector belong to the minimum output set in this step, because S330 needs to call these two fields simultaneously for injection pose calculation. If the current injectable area is affected by local occlusion, the control unit will call adjacent acquisition frames to supplement point cloud data, and then re-extract the center and perform local plane fitting; if stable center position coordinates and normal vectors still cannot be obtained after supplementation, the current record will be written as a local recognition anomaly, and the target motion data will be retained for the next round of image acquisition and processing.

[0207] In real-world scenarios, after a cow completes S310, the image acquisition unit will briefly acquire data in a local area of ​​the cow. The control unit, combined with the front-end 3D model, quickly locates the injectable area and generates the center position coordinates and normal vector from it.

[0208] After this step is completed, the center position coordinates and normal vector are recorded as the output field name "Center Position Coordinates and Normal Vector", and used as the direct input of S330, while providing the basis for the local approach direction for the robotic arm control processing of S410.

[0209] S330. Based on the center position coordinates and normal vector, perform injection pose calculation to obtain the injection pose.

[0210] Specifically, the input source for this step is the center position coordinates and normal vector output by S320, and the target motion quantity in S310 is used as the execution constraint. The injection pose is the pose field used by the robotic arm controller to drive the movement of the six-degree-of-freedom robotic arm, and at least includes the positional relationship and orientation relationship when the end syringe approaches the injectable area; the injection pose calculation process is the process of combining the positional relationship corresponding to the center position coordinates and the orientation relationship corresponding to the normal vector into the same control field.

[0211] In specific implementation, the control unit first reads the center position coordinates to form the approach point of the automatic injection device's end effector in the 3D model; then it reads the normal vector to form the orientation of the automatic injection device's end effector when approaching the approach point; then, based on the current execution structure corresponding to the target motion amount, it corrects the reserved distance between the end-effector and the injectable area to obtain the final injection pose. This correction process means that when the automatic injection device uses a push cylinder and syringe pusher structure, the control unit writes the motion amount of the syringe pusher as the end-effector propulsion reserved amount into the current injection pose; when the automatic injection device uses a first feed mechanism structure, the control unit writes the call status of the first feed mechanism into the current injection pose for synchronous reading by the S410 during robotic arm control processing.

[0212] Furthermore, this step also includes abnormal pose rejection processing. Specifically, the control unit checks the relative relationship between the current normal vector and the current 3D model surface. If the normal vector direction is significantly reversed or the center position coordinates are located outside the boundary, the current injection pose is recorded as an invalid pose, and the process returns to S320 to perform the second recognition again. If the normal vector direction matches the local surface direction, and the center position coordinates are located in the middle of the injectable area, a valid injection pose is output.

[0213] Understandably, the injection pose is not directly given by local recognition alone, but is a control field completed under the premise that the target motion has been formed. Therefore, the injection pose output in this step is directly connected with the robotic arm control processing in S410 and the automatic injection device processing in S420.

[0214] In an engineering embodiment, after obtaining the center position coordinates and normal vector of a cow, the control unit incorporates the target motion of the cow into the current pose calculation record, and then outputs the corresponding injection pose and sends it to the robotic arm controller; the robotic arm controller then drives the six-degree-of-freedom robotic arm to approach the injectable area based on the injection pose.

[0215] After this step is completed, the injection pose is recorded as the output field name "injection pose", and its direct input position is "the injection pose" in S410. At the same time, the injection pose and the target motion generated in S310 are also entered into the injection information feedback link of S400.

[0216] In summary, this step integrates the target motion, center position coordinates, and normal vector into a single injection pose calculation. The injection pose is no longer a localized positional result independent of dosage information, but rather a control field generated synchronously with the preceding execution quantity field. Compared to the approach of locating first and then uniformly executing the injection, this step allows subsequent robotic arm control to simultaneously read position and execution quantity information, resulting in a tighter relationship between fields passed between steps.

[0217] Step S400 includes at least steps S410-S430:

[0218] S410. Obtain the injection pose and the target motion, perform robotic arm control processing, and obtain the contact force.

[0219] Specifically, the input sources for this step are the injection pose output by S330 and the target motion quantity output by S310. The injection pose is the positional and orientational relationship of the end syringe formed by the center position coordinates and the normal vector; the target motion quantity is an execution quantity field converted from dosage information, which at least includes the propulsion amount of the push cylinder, the motion amount of the syringe pusher, and the call status of the first feed mechanism. The robotic arm control processing is executed by the robotic arm controller, which is connected to the six-degree-of-freedom robotic arm and the automatic injection device; the six-degree-of-freedom robotic arm is responsible for moving the automatic injection device from the return to the initial pose to the injection pose, and the automatic injection device is installed at the end of the six-degree-of-freedom robotic arm.

[0220] In specific processing, the robotic arm controller first reads the positional and orientation relationships in the injection pose, and then reads the current execution structure information in the target motion quantity; if the current record corresponds to the push cylinder and syringe push block structure, the robotic arm controller moves the end syringe to the vicinity of the injection pose and retains the push cylinder in the pending execution state; if the current record corresponds to the first feed mechanism structure, the robotic arm controller writes the calling state of the first feed mechanism into the current control record.

[0221] Furthermore, the robotic arm control processing includes approach actions, alignment actions, and contact determination actions. The approach action involves the six-DOF robotic arm moving towards the injectable area according to the injection pose orientation; the alignment action involves the end-effector adjusting its orientation based on the normal vector, aligning the end-effector with the local surface direction; the contact determination action involves continuously reading the force information between the end-effector and the injectable area using a six-dimensional force sensor and recording this force information as a contact force. This contact force is the force field when the injectable area of ​​the cow comes into contact with the automatic injection device, and it is a direct input for subsequent processing by the automatic injection device.

[0222] To accommodate the slight swaying of cows in engineering scenarios, the robotic arm controller continuously reads the injection pose and normal vector from the current control record during the approach motion. When the six-dimensional force sensor has not yet detected a change in contact force, the six-degree-of-freedom robotic arm continues to approach in the current injection pose. When the six-dimensional force sensor detects that the contact force has entered the set threshold range, the robotic arm controller stops approaching and maintains the current end syringe position, writing the contact force into the current identification record.

[0223] If the contact force changes excessively during the approach process, the robotic arm controller will record this robotic arm control process as an abnormal contact record and return the six-degree-of-freedom robotic arm to its initial pose. At the same time, the injection pose and the target motion will be retained and called again in the next batch of data collection.

[0224] Understandably, this step does not perform the actual injection action. Instead, it translates the injection pose and target motion generated in the previous stage into the control link of the six-DOF robotic arm and the six-dimensional force sensor to first form a stable contact force field. After completing this step, the contact force is recorded as the output field name "Contact Force" and serves as the direct input for the automatic injection device in S420. Simultaneously, the contact force is synchronously saved in the current identification record along with the ear tag information, body shape recognition result, and dosage information, for unified uploading, distribution, and feedback processing by S430.

[0225] S420. Based on the contact force, the automatic injection device processes the data to obtain injection information.

[0226] Specifically, the input source for this step is the contact force output by S410, and it continues to call upon the target motion and injection pose from the currently identified record. The automatic injection device is an actuator installed at the end of a six-degree-of-freedom robotic arm, and includes at least one of the following structures: a push cylinder, a syringe pusher, and a continuous syringe; or it includes one of the following structures: a first feed mechanism and an injection component; or it includes a needleless syringe structure. The injection information is a record field formed after the automatic injection device performs the injection action, and it includes at least the injection execution status corresponding to the current ear tag information and the current injection action information.

[0227] In specific processing, the automatic injection device first reads the contact force to determine whether the end syringe has entered the executable state with the injectable area; when the contact force reaches the set threshold, the automatic injection device enters the injection action stage.

[0228] For the push cylinder and syringe push block structure, the control unit calls the target motion amount generated by S310, drives the push cylinder to move according to the current push amount, and drives the syringe push block to push the continuous syringe to complete the current injection action; for the first feeding mechanism and injection component structure, after the control unit reads the contact force, it first makes the injection component approach the injectable area at a first speed, and after the distal end of the injection component contacts the injectable area, it makes the injection component continue to move at a second speed. When the position of the marking ring is flush with the surface of the injectable area, the current injection action is executed; for the needleless syringe structure, after the contact force reaches the set threshold, the control unit controls the six-degree-of-freedom robotic arm to continue to move forward for a certain period of time, and then the needleless syringe completes the current injection action.

[0229] Furthermore, the automatic injection device in this step does not regenerate dosage information, but directly executes the target motion amount already established in the previous stage. That is, the propulsion amount of the cylinder, the motion amount of the syringe pusher, and the call status of the first feed mechanism are all directly executed in the current injection action. After completing the current injection action, the automatic injection device writes the current action status, the current call structure, and the current execution completion status into the current identification record, forming injection information. If, during the automatic injection device's processing, the contact force deviates from the set threshold range, the control unit stops the current injection action and writes the current record as an injection interruption message.

[0230] Understandably, the injection information is not a simple injection mark, but a record field directly generated by the automatic injection device processing chain, which will subsequently enter S430 along with ear tag information, body shape recognition results and dosage information.

[0231] In a specific engineering embodiment, when the six-degree-of-freedom robotic arm corresponding to a certain cow has moved to the injection position and read the contact force, if the current structure is a continuous syringe structure, the cylinder is pushed forward according to the target motion amount, the syringe pusher pushes the continuous syringe to output the current injection volume, and the completion status is written into the injection information; if the current structure is a needleless syringe structure, the six-degree-of-freedom robotic arm continues to move forward for a certain period of time to complete the current injection action, and the current injection action is written into the injection information.

[0232] After this step is completed, the injection information is recorded as the output field name "Injection Information" and used as the direct input of S430. At the same time, it will be used as a core record field in the uploading, distribution and feedback processing when the historical injection information is updated in the future.

[0233] S430. Based on the ear tag information, the body shape recognition result, the dosage information, and the injection information, upload, distribute, and provide feedback to obtain historical injection information.

[0234] Specifically, the input sources for this step include the ear tag information received in S110, the body shape recognition result output in S130, the dosage information output in S230, and the injection information output in S420. The upload, distribution, and feedback processes are collaboratively executed by the gateway, cloud platform, and external platform. The gateway is responsible for sending the current identification record from the local control unit to the cloud platform; the cloud platform is responsible for receiving the current identification record and writing it into the corresponding individual cow's data record; the external platform is responsible for receiving the distributed injection record and returning an update status. The historical injection information is an updated field formed after the current cow completes the upload, distribution, and feedback processes, and it is subsequently retrieved again as historical injection information in S220.

[0235] In specific processing, the control unit first writes the ear tag information, body shape recognition result, dosage information, and injection information into the same upload record, and then sends the upload record to the cloud platform through the gateway. After receiving the upload record, the cloud platform first locates the data record corresponding to the current cow based on the ear tag information, and then writes the body shape recognition result, dosage information, and injection information into the data record item by item. Subsequently, the cloud platform distributes the updated current record to an external platform, which receives and provides feedback on the current record and returns the feedback status to the cloud platform. The cloud platform generates the current cow's historical injection information based on the feedback status. The historical injection information here includes at least the ear tag information, body shape recognition result, dosage information, and injection information corresponding to this injection. Therefore, when S220 performs the combined processing of objects and additional conditions again, it can directly call this historical injection information.

[0236] Furthermore, when the injection information output by S420 is interrupted injection information, S430 still performs upload, distribution, and feedback processing, but writes the record as incomplete in the cloud platform; when the injection information output by S420 is complete, S430 writes the record as complete. For cases where the same ear tag information has multiple upload records within the same time period, the cloud platform retains the latest record in the upload time order and saves the previous record in the historical injection information for subsequent reading by S220. Understandably, the historical injection information formed in this step is not an additional record independent of the previous steps, but rather a return field in the closed-loop chain of this invention. This field returns from S430 to S220, enabling the next combination processing of the object and additional conditions to directly call the previous injection record of the current cow. During project implementation, after a cow completes its injection, the gateway uploads the ear tag information, body shape recognition results, dosage information, and injection information to the cloud platform. The cloud platform updates the cow's current record and distributes it to an external platform. Subsequently, the external platform returns a feedback status, and the cloud platform generates new historical injection information. When the cow enters the fixed location again, the S220 directly retrieves the historical injection information to participate in the combined processing of new objects and additional conditions.

[0237] Summary of the technical effects of this step: This step integrates ear tag information, body shape recognition results, dosage information, and injection information into the historical injection information. The historical injection information no longer only records whether an injection was administered, but also records the current cow's body shape recognition result and current dosage information. Compared to the processing method that only uploads object identity and vaccine information, the feedback fields generated by this step are more complete. When S220 is called again, the previous identification record and the previous injection record can be used continuously in the same processing chain.

[0238] Example 2: Figure 2 A structural block diagram of an intelligent immune injection robot based on body shape recognition according to an embodiment of the present invention is shown. Figure 2 As shown, the structure may include:

[0239] The identification and image acquisition module 01 is used to receive ear tag information through an identification receiver, and to acquire red, green, and blue data and point cloud data of the cow's back, side, and rump regions by an image acquisition unit. Specifically, the identification and image acquisition module receives individual cows entering a fixed position as input objects. First, the identification receiver reads the ear tag information and writes it into the current acquisition record. Then, the image acquisition unit performs synchronous acquisition according to a preset field of view for the back, side, and rump regions to obtain red, green, and blue data and point cloud data. The image acquisition unit completes frame access, region filtering, and acquisition record association within the same acquisition cycle, removing occluded and missing regions and retaining valid acquisition data corresponding to the ear tag information. The identification and image acquisition module outputs the red, green, and blue data and point cloud data to the first identification module as preliminary input, and keeps the ear tag information in the current acquisition record for use by the motion state tracking, identification, and combination processing module and the upload, distribution, and feedback module.

[0240] The first recognition module 02, connected to the identity recognition and image acquisition module, performs first recognition processing on the red-green-blue data and point cloud data to obtain a three-dimensional model. It then processes the three-dimensional model for key point information, geometric model, and pose information to obtain a body shape recognition result. Specifically, the first recognition module receives the red-green-blue data and point cloud data output by the identity recognition and image acquisition module, registers and stitches the back region, side region, and rump region data from the same acquisition record to form a three-dimensional model. The first recognition module then performs key point information extraction, geometric model construction, and pose information processing around the three-dimensional model. The key point information corresponds to the representative position of the cow's torso outline, the geometric model corresponds to the shape relationship between the back region, side region, and rump region, and the pose information corresponds to the cow's current orientation and posture in a fixed position. The first recognition module outputs the body shape recognition result to the motion state tracking, recognition, and combination processing module for use, and saves the three-dimensional model in the current acquisition record for the target motion amount and injection pose generation module to read in the second recognition processing.

[0241] The motion state tracking, recognition, and combination processing module 03, connected to the first recognition module, is used to perform motion state tracking and recognition processing on the body shape recognition result to obtain a static state. It also receives the ear tag information, the vaccine information to be injected, and historical injection information. The module performs combination processing on the ear tag information, the body shape recognition result, the vaccine information to be injected, the historical injection information, and the static state, obtaining a combination of the object and additional conditions. Specifically, the motion state tracking, recognition, and combination processing module receives the body shape recognition result and reads corresponding key point information and pose information from the current and previous acquisition records. It compares the spatial position changes at different times to obtain a static state. The module also receives the vaccine information to be injected from an external platform and the historical injection information from the upload, distribution, and feedback module. It writes the ear tag information, the body shape recognition result, the vaccine information to be injected, the historical injection information, and the static state into the same combination record, forming a combination of the object and additional conditions. If the static state does not meet the current processing conditions, the combined record is retained in the current acquisition record and awaits recall in the next acquisition cycle; if the static state meets the current processing conditions, the combination of the object and the additional conditions is passed to the dose information output module as the input field name.

[0242] The dosage information output module 04, connected to the motion state tracking, recognition, and combination processing module, is used to perform corresponding task processing on the combination of the object and additional conditions to obtain dosage information. Specifically, the dosage information output module receives the combination of the object and additional conditions, reads the ear tag information, body shape recognition result, vaccine information to be injected, historical injection information, and static state item by item in the combination record, and performs corresponding task processing. The execution of corresponding task processing includes current task item verification, historical record comparison, and dosage field generation. First, it determines whether the vaccine information to be injected for the current cow has a duplicate record with the historical injection information, and then combines the body shape recognition result to form the current dosage information. If the same task item record exists in the historical injection information, the dosage information output module writes the current recognition record as a non-duplicate injection state; if the same task item record does not exist in the historical injection information, the dosage information output module generates the current dosage information and transmits it to the target motion and injection posture generation module, while writing the current dosage information into the current acquisition record for subsequent reading by the upload, distribution, and feedback module.

[0243] The target motion quantity and injection pose generation module 05, connected to the dosage information output module, receives the dosage information, processes the push cylinder, syringe push block, and first feeding mechanism to obtain the target motion quantity, and performs a second recognition process on the target motion quantity to obtain the center position coordinates and normal vector. Then, it performs injection pose calculation processing on the center position coordinates and normal vector to obtain the injection pose. Specifically, the target motion quantity and injection pose generation module receives the dosage information, first performs action field mapping processing around the push cylinder, syringe push block, and first feeding mechanism to convert the dosage information into the target motion quantity, and writes the target motion quantity into the current control record. Subsequently, the target motion quantity and injection pose generation module calls the three-dimensional model saved by the first recognition module, performs a second recognition process on the local surface area of ​​the current cow, determines the injectable area in the candidate area, and obtains the center position coordinates and normal vector. Then, the target motion quantity and injection pose generation module associates the center position coordinates and normal vector with the target motion quantity and performs injection pose calculation processing to form the injection pose. The target motion and the injection pose are jointly transmitted to the robotic arm control and automatic injection module. The center position coordinates and normal vector are retained in the current control record for tracing the correspondence of the current injection pose.

[0244] The robotic arm control and automatic injection module 06, connected to the target motion and injection pose generation module, receives the injection pose and the target motion, performs robotic arm control processing on the six-degree-of-freedom robotic arm to obtain contact force, and processes the contact force using the automatic injection device to obtain injection information. Specifically, the robotic arm control and automatic injection module receives the injection pose and the target motion, and the robotic arm controller first drives the six-degree-of-freedom robotic arm from the return to the initial pose to the injection pose, maintaining the orientation relationship between the end syringe and the injection pose during the movement. When the six-degree-of-freedom robotic arm approaches the injectable area, the robotic arm control and automatic injection module reads the output of the six-dimensional force sensor to form a contact force. When the contact force enters the set range in the current control record, the automatic injection device reads the target motion and executes the corresponding push cylinder action, syringe push block action, or first feed mechanism action to complete the current injection action and obtain injection information. If the contact force does not meet the current control conditions or changes abnormally, the robotic arm control and automatic injection module pauses the automatic injection device processing and writes the abnormal state into the injection information. The injection information is output to the upload and distribution feedback module, and the contact force and the injection posture are synchronously retained in the current control record.

[0245] The upload, distribution, and feedback module 07, connected to the robotic arm control and automatic injection module, receives the ear tag information, body shape recognition result, dosage information, and injection information. It then uploads, distributes, and provides feedback to obtain historical injection information, which is then sent to the motion state tracking, recognition, and combination processing module. Specifically, the upload, distribution, and feedback module receives ear tag information, body shape recognition result, dosage information from the currently collected record, and injection information from the robotic arm control and automatic injection module. It organizes these fields into a single upload record and sends it to the cloud platform via a gateway. After writing to the cloud platform, the upload, distribution, and feedback module distributes the current upload record to an external platform and receives feedback status from the external platform. Subsequently, it updates the historical injection information corresponding to the current cow based on the feedback status. The historical injection information includes the ear tag information, body shape recognition result, dosage information, and injection information corresponding to the current cow, and is sent as a feedback field to the motion state tracking, recognition, and combination processing module, allowing subsequent processing of objects and additional conditions to directly call the currently updated historical injection information. The upload, distribution, and feedback module also registers the upload status, distribution status, and feedback status in the current record, so that the dose information output module and the motion state tracking, identification, and combination processing module can read consistent historical records.

Claims

1. A method for intelligent immune injection dosage control based on body shape recognition, characterized in that, include: S100: Acquire ear tag information, RGB data, and point cloud data, perform first recognition processing, and obtain body shape recognition results; wherein, the ear tag information is individual identification information read by the radio frequency identification identity recognition receiver, including ID information used to distinguish individual cows; the RGB data is red, green, and blue data; the point cloud data is a set of spatial points formed by unfolding the depth image in 3D coordinate space; S200. Based on the body shape recognition result, perform motion state tracking and recognition, object and additional condition combination processing, and execute corresponding task processing to obtain dose information; S300: Based on the dosage information, perform processing of the pushing cylinder, syringe pusher and first feeding mechanism and injection posture calculation to obtain the injection posture; S400. Based on the injection pose, perform robotic arm control processing and execute upload, distribution and feedback processing to obtain historical injection information.

2. The method according to claim 1, characterized in that, The first identification process includes: Based on the RGB data and point cloud data, registration processing, cow body region filtering processing, and shape stitching processing are performed to generate a three-dimensional model; The registration process includes aligning the point cloud data of the back, side, and hip regions with the RGB data. The cow body area screening process includes removing fences, ground and non-dairy cow areas, and retaining the main area of ​​the dairy cow's torso. The shape splicing process includes connecting the upper surface contour of the back, the side profile contour, and the rear boundary of the buttocks into a continuous shape. Based on the three-dimensional model, key point information of the back region, side region and buttock region is extracted, a geometric model including back length relationship, side height relationship and buttock width relationship is established, and pose information including cow spatial orientation, torso tilt state and main body position state is generated. The key point information, geometric model, and pose information are combined to form the body shape recognition result.

3. The method according to claim 1, characterized in that, The process of motion state tracking and recognition includes: Retrieve the body shape recognition results of the same ear tag information in the previous batch and compare them with the body shape recognition results of the current batch item by item; When the average spatial geometric distance between key points in the back, side and rump regions is less than a preset distance threshold, the cow is determined to be in a stationary state. Otherwise, it is determined to be in motion; the stationary state field is output.

4. The method according to claim 3, characterized in that, The process of combining objects with additional conditions includes: The ear tag information, the body shape recognition result, the static state, and the information on vaccines to be injected and historical injection information retrieved from the cloud platform or external platform are obtained. The vaccine information to be injected includes the vaccine information that the dairy cow currently needs to receive; The historical injection information includes injection records corresponding to the ear tag information; When the static state is a processable state and the ear tag information matches the individual identifier in the vaccine information to be injected and the historical injection information, the body shape recognition result, the vaccine information to be injected, the historical injection information and the static state are merged into the same current recognition record to form a combination of object and additional conditions.

5. The method according to claim 4, characterized in that, The process of executing the corresponding task includes: Read the static state of the object and the additional conditions; if it is a moving state, mark it as a pending state and wait for the next call. If the status is static, check if there is an injection record in the historical injection information that is identical to the information of the vaccine to be injected; If no corresponding injection record exists, the dosage information for the current cow is generated by matching the vaccine information to be injected and the body shape recognition result according to the pre-stored correspondence in the database. If a corresponding injection record already exists, output the dosage information for non-repeated injection and write it to the injection information status. The dosage information is bound and saved together with the ear tag information and body shape recognition results.

6. The method according to claim 5, characterized in that, The process of handling the cylinder, syringe pusher, and first feed mechanism includes: Obtain the dosage information and determine the execution path based on the current structure of the automatic injection device; When a structure is adopted in which the push cylinder and the syringe push block cooperate, the dosage information is converted into the pushing amount of the push cylinder and the movement amount of the syringe push block, and the target movement amount is written. When the structure of the first feeding mechanism and the injection piece are used, the dosage information is converted into the calling state of the first feeding mechanism and written into the target motion amount; Check the historical injection information retrieval status corresponding to the current ear tag information. If the same vaccine information to be injected has been injected, the target motion volume is recorded as closed; otherwise, a valid target motion volume is generated. The target motion quantity is output, which includes the propulsion amount of the push cylinder, the motion amount of the syringe push block, and the call status of the first feed mechanism.

7. The method according to claim 6, characterized in that, The injection pose calculation and processing, which yields the injection pose, specifically includes: After generating effective target motion, the 3D model and body shape recognition results are used as the basis for the first-stage shape recognition process; The second identification process includes: Check if the target motion volume is in an effective state; if it is in a closed state, only generate the recorded results. If the condition is effective, search for injectable areas in local areas of the back, sides and buttocks. Extract local areas from the 3D model and filter candidate areas by local boundary continuity and surface smoothness to determine injectable areas. Perform center extraction on the injectable region to obtain the center position coordinates; Perform local plane fitting on the injectable region to obtain the normal vector; Based on the center position coordinates, normal vector, and target motion, the injection pose is calculated: The center position coordinates are used to form the approach point of the end syringe, and the normal vector is used to form the orientation of the end syringe. The reserved distance between the end syringe and the injectable area is corrected according to the propulsion amount or call state in the target motion to obtain the injection pose. The injection posture includes the positional and orientational relationships of the end syringe.

8. The method according to claim 7, characterized in that, The robotic arm control process includes: The injection pose and the target motion are obtained, and the robotic arm controller executes the approach action, alignment action and contact determination action. The approach action is that the six-degree-of-freedom robotic arm moves towards the injectable area according to the orientation relationship of the injection pose; The alignment action involves adjusting the orientation of the end syringe according to the normal vector, so that the end syringe is aligned with the direction of the local surface. The contact determination action is to continuously read the force information between the end syringe and the injectable area through a six-dimensional force sensor. When the contact force enters the set threshold range, the approach is stopped and the current position is maintained. The contact force is written into the current identification record. If the contact force changes too much during the approach process, it is recorded as an abnormal contact record, causing the six-degree-of-freedom robotic arm to return to the initial pose and retain the injection pose and target motion for the next call. Once the contact force reaches a set threshold, the automatic injection device is triggered for processing. For the push cylinder and syringe push block structure, the target motion amount is called to drive the push cylinder to move according to the push amount, which drives the syringe push block to push the continuous syringe to complete the injection action. For the first feeding mechanism and the injection component structure, after the contact force reaches the threshold, the injection component is first moved at the first speed. After the distal end contacts the injectable area, the action continues at the second speed. When the marking ring is flush with the body surface, the injection action is performed. After the injection is completed, injection information is generated, which includes ear tag information, injection execution status, and injection action information.

9. The method according to claim 8, characterized in that, The process of uploading, distributing, and processing feedback includes: The gateway acquires the ear tag information, body shape recognition result, dosage information, and injection information, and sends the current recognition record to the cloud platform. The cloud platform locates the corresponding dairy cow's data record based on the ear tag information and writes in the body size recognition results, dosage information, and injection information; The cloud platform distributes the updated records to external platforms, receives feedback status, and generates updated historical injection information. The historical injection information includes the ear tag information, body shape recognition result, dosage information, and injection information corresponding to this injection, and serves as the source for calling historical injection information in the next combination processing of objects and additional conditions.

10. An intelligent immune injection robot based on body shape recognition, characterized in that, include: The system comprises an identity recognition and image acquisition module, a first recognition module, a motion state tracking and recognition and combined processing module, a dose information output module, a target motion amount and injection posture generation module, a robotic arm control and automatic injection module, and an upload and distribution feedback module; the modules are connected in sequence to implement the method described in any one of claims 1-9.

Images