Rice automatic replanting method and system

Abstract

The application discloses a rice automatic replanting method and system, through real-time collection of rice transplanting video data, and fast input of an identification server based on photometric stereo imaging technology and a deep learning algorithm, the photometric stereo imaging technology can obtain multi-dimensional image information of seedlings, the deep learning algorithm can quickly and accurately analyze data, timely identify seedling belt data and seedling shortage positions, realize quick positioning of seedling shortage, shorten problem discovery time, and improve replanting efficiency; according to the identification result, a replanting machine travel path is automatically planned, the replanting machine operates according to the planning, process automation is realized, efficiency is further improved, and large-area replanting can be completed in a short time; meanwhile, most of the work of manual patrol and replanting is replaced by automatic equipment and technology, manual use is reduced, and cost is reduced; moreover, replanting is planned immediately once seedling shortage is detected, and the delay problem caused by untimely patrol or personnel arrangement in manual replanting is avoided.

Description

Technical Field

[0001] This invention relates to the field of rice seedling replanting technology, and in particular to an automatic rice seedling replanting method and system. Background Technology

[0002] In the field of agricultural production, rice is one of the world's most important food crops. Its yield and quality are directly related to food security and the stable development of the agricultural economy. Rice transplanting is a key step in the rice planting process. The quality of transplanting has a profound impact on the subsequent growth, tillering, heading and final yield of rice.

[0003] With the development of agricultural mechanization, mechanical rice transplanting has gradually become widespread, greatly improving transplanting efficiency and reducing labor intensity. However, problems such as missed transplanting and missing seedlings are inevitable during mechanical rice transplanting. These missing seedlings will lead to uneven distribution of rice plants and a reduction in the effective number of plants per unit area, thereby affecting the photosynthetic efficiency of rice and land utilization, ultimately resulting in yield loss. In order to solve the problem of missing seedlings after mechanical transplanting, the traditional method is to manually replant seedlings. However, manual replanting also has problems such as low efficiency, high cost, and untimely replanting, which makes it difficult to meet the needs of large-scale rice planting. Summary of the Invention

[0004] In view of this, the present invention proposes an automatic rice seedling replanting method and system, which can effectively solve the defects of low efficiency, high cost and untimely replanting of existing technologies.

[0005] The technical solution of this invention is implemented as follows:

[0006] An automatic rice seedling replanting method, specifically including:

[0007] Real-time acquisition of video data on rice transplanting;

[0008] The rice transplanting video data is input into the rice transplanting image recognition server, which is built based on photometric stereo imaging technology and deep learning algorithms.

[0009] The rice transplanting image recognition server identifies seedling strip data and missing seedling locations based on rice transplanting video data;

[0010] Based on the seedling strip data and the location of missing seedlings, plan the travel path of the seedling replenishment machine;

[0011] The seedling replenishment machine replenishes seedlings according to the planned travel path;

[0012] The rice transplanting image recognition server identifies seedling strip data and missing seedling locations based on rice transplanting video data, specifically including:

[0013] Analyze the video data of rice transplanting and extract the feature points of rice seedlings;

[0014] Based on the robust regression algorithm, the extracted rice seedling feature points are fitted with row lines to generate the center line of the rice row.

[0015] By combining the fitted center line of the rice row with the spatial coordinates of the feature points, the position information of each rice seedling is calculated;

[0016] Based on the location information of each rice seedling, the seedling strip data and the location of missing seedlings are identified.

[0017] As a further optional solution to the aforementioned automatic rice seedling replanting method, the rice transplanting image recognition server is built based on photometric stereo imaging technology and deep learning algorithms, specifically including:

[0018] Collect multi-angle illumination image data of rice transplanting scenes and use photometric stereo imaging technology to construct a three-dimensional surface model;

[0019] A convolutional neural network model was constructed based on deep learning algorithms. Feature learning and classification training were performed on the three-dimensional surface model to obtain a rice transplanting image recognition model.

[0020] The rice transplanting image recognition model is deployed to the server to obtain the rice transplanting image recognition server.

[0021] As a further optional solution to the aforementioned automatic rice seedling replanting method, the construction of a three-dimensional surface model using photometric stereoscopic imaging technology specifically includes:

[0022] Acquire multi-angle illumination images of the rice transplanting area and collect information on rice seedlings and ground surface reflection under different illumination directions;

[0023] Based on the photometric stereo vision algorithm, combined with the reflection information of rice seedlings and the ground surface under different lighting directions, the normal vector distribution of rice seedlings and the ground surface is calculated.

[0024] Based on the normal vector distribution of rice seedlings and the ground surface, a three-dimensional surface model of the rice transplanting area is reconstructed.

[0025] As a further optional solution to the aforementioned automatic rice seedling replanting method, the step of using a robust regression algorithm to fit the extracted rice seedling feature points to row lines to generate rice row centerlines specifically includes:

[0026] Density clustering analysis was performed on the extracted rice seedling feature points to obtain a set of effective feature points;

[0027] For the effective set of feature points, the line parameters are iteratively calculated using a robust regression algorithm;

[0028] Based on the row line parameters and image boundary range, generate the center line of the rice rows covering the entire image area.

[0029] As a further optional solution to the aforementioned automatic rice seedling replenishment method, the step of planning the travel path of the seedling replenishment machine based on seedling strip data and the location of missing seedlings specifically includes:

[0030] Based on the seedling strip data, extract the spatial orientation, plant spacing, and row spacing parameters of the seedling strip;

[0031] Based on the spatial orientation of the seedling strip, the plant spacing, and the row spacing parameters, the center line of the seedling strip is generated;

[0032] Generate a coordinate system for the seedling strip based on its centerline;

[0033] Map the missing seedling points to the seedling strip coordinate system and determine the lateral offset and longitudinal distance of each missing seedling point relative to the seedling strip;

[0034] The Dijkstra algorithm is used, with the center line of the seedling strip as the reference path, and incorporates the lateral offset and longitudinal distance of each missing seedling point relative to the seedling strip to generate the optimal travel path covering all missing seedling points.

[0035] As a further optional embodiment of the automatic rice replanting method, the method further includes:

[0036] After the replanting is completed, the total number of holes in the monitoring area is calculated based on the row spacing and plant spacing of the rice transplanter. The number of rice plants in each hole is analyzed, and the rice emergence rate is calculated.

[0037] An automatic rice seedling replanting system includes:

[0038] The data acquisition module is used to collect real-time video data of rice transplanting.

[0039] The input module is used to input rice transplanting video data into the rice transplanting image recognition server, which is built based on photometric stereo imaging technology and deep learning algorithms.

[0040] The recognition module is used by the rice transplanting image recognition server to identify the seedling strip data and missing seedling locations based on the rice transplanting video data;

[0041] The path planning module is used to plan the travel path of the seedling replenishment machine based on the seedling strip data and the location of missing seedlings;

[0042] The seedling replenishment module is used by the seedling replenishment machine to replenish seedlings according to the planned travel path.

[0043] A computing device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the above-described automatic rice replanting methods.

[0044] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described automatic rice replanting methods.

[0045] The beneficial effects of this invention are: real-time acquisition of rice transplanting video data, which is then rapidly input into a rice transplanting image recognition server built based on photometric stereo imaging technology and deep learning algorithms. Photometric stereo imaging technology can acquire multi-dimensional image information of rice seedlings, while the deep learning algorithm can quickly and accurately analyze this image data, promptly identifying seedling strip data and missing seedling locations, thus achieving rapid location of missing seedlings and significantly shortening the time required to detect the problem, thereby improving the overall replanting efficiency. Simultaneously, based on the identified seedling strip data and missing seedling locations, the replanting machine's path is automatically planned. The seedling replenishment machine operates according to the planned path, and the entire process requires no manual intervention in the planning of the operation path and the seedling replenishment action, realizing the automation of the seedling replenishment process and further improving the efficiency of seedling replenishment. It can complete the seedling replenishment work of large-area rice fields in a shorter time. In addition, by replacing most of the manual inspection and seedling replenishment work with automated equipment and technology, the amount of manpower used is reduced, thereby reducing labor costs. Furthermore, once a missing seedling is detected, the seedling replenishment machine immediately plans the path and carries out the seedling replenishment operation, avoiding the delays in seedling replenishment that may occur due to untimely inspections or personnel allocation problems that may occur with manual seedling replenishment. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart of an automatic rice seedling replanting method according to the present invention;

[0048] Figure 2 This is a schematic diagram of the composition of an automatic rice seedling replanting system according to the present invention;

[0049] Figure 3 This is a schematic diagram of the composition of a computing device according to the present invention. Detailed Implementation

[0050] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] refer to Figures 1 to 3 An automatic rice seedling replanting method, specifically including:

[0052] Real-time acquisition of video data on rice transplanting;

[0053] The rice transplanting video data is input into a rice transplanting image recognition server. This server is built based on photometric stereoscopic imaging technology and deep learning algorithms. In some embodiments, the rice transplanting image recognition server, specifically built based on photometric stereoscopic imaging technology and deep learning algorithms, includes:

[0054] Collect multi-angle illumination image data of rice transplanting scenes and construct a three-dimensional surface model using photometric stereo imaging technology;

[0055] A convolutional neural network model was constructed based on deep learning algorithms. Feature learning and classification training were performed on the three-dimensional surface model to obtain a rice transplanting image recognition model.

[0056] The rice transplanting image recognition model is deployed to the server to obtain the rice transplanting image recognition server.

[0057] Specifically, collecting multi-angle lighting image data of rice transplanting scenes can obtain rich visual information of rice seedlings under different lighting conditions. Light from different angles can highlight different features of the seedling surface, such as bumps and textures, providing a comprehensive data foundation for the subsequent construction of accurate three-dimensional surface models.

[0058] Using photometric stereo imaging technology, a three-dimensional surface model is constructed based on multi-angle illumination image data. This technology analyzes the brightness changes of images under different illumination directions to accurately calculate the three-dimensional shape and reflection characteristics of the object surface. For rice transplanting scenarios, it can accurately restore the three-dimensional information such as the shape, height, and density of the seedlings. Compared with traditional two-dimensional image recognition, it can more realistically reflect the actual situation of rice transplanting and greatly improve the accuracy of image information acquisition.

[0059] A convolutional neural network model is built based on deep learning algorithms. Convolutional neural networks have powerful feature extraction and learning capabilities. They can automatically learn various features of rice seedlings from three-dimensional surface models, including shape, color, and texture, without the need for manual design of feature extraction rules.

[0060] By performing feature learning and classification training on the 3D surface model, the model can continuously optimize its parameters and improve its ability to recognize rice transplanting images. After training with a large amount of data, the model can accurately distinguish between seedlings, missing seedling areas, and other interference objects, resulting in a high-precision rice transplanting image recognition model that effectively reduces the false recognition rate.

[0061] In some embodiments, the construction of a three-dimensional surface model using photometric stereo imaging technology specifically includes:

[0062] Acquire multi-angle illumination images of the rice transplanting area and collect information on rice seedlings and ground surface reflection under different illumination directions;

[0063] Based on the photometric stereo vision algorithm, combined with the reflection information of rice seedlings and the ground surface under different lighting directions, the normal vector distribution of rice seedlings and the ground surface is calculated.

[0064] Based on the normal vector distribution of rice seedlings and the ground surface, a three-dimensional surface model of the rice transplanting area is reconstructed.

[0065] Specifically, based on the photometric stereo vision algorithm, the normal vector distribution is calculated by combining the reflection information of rice seedlings and the ground surface under different lighting directions. This algorithm can accurately derive the normal vectors of rice seedlings and ground surface points by utilizing the relationship between lighting and object surface reflection. The normal vector is an important parameter describing the orientation of the object surface. Accurate calculation of the normal vector can better understand the three-dimensional shape of rice seedlings and ground surface. The accurate calculation of the normal vector distribution provides key data support for the subsequent reconstruction of the three-dimensional surface model, enabling the model to more realistically restore the terrain and distribution of rice seedlings in the rice planting area and reduce the error between the model and the actual scene.

[0066] Based on the calculated distribution of normal vectors of rice seedlings and the ground surface, a three-dimensional surface model of the rice transplanting area can be reconstructed. This model can accurately reproduce the height, shape, and surface undulations of the rice seedlings. This is of great significance for analyzing the quality of rice transplanting and the growth environment of rice seedlings. For example, it can clearly show whether the rice seedlings are transplanted neatly and at a consistent depth, and whether there are pits or bumps on the ground surface that affect rice growth. The realistic three-dimensional surface model can provide a reliable basis for the subsequent management of rice planting. For example, in an automatic seedling replenishment system, the model can accurately determine the location of missing seedlings.

[0067] It should be noted that, based on the photometric stereo vision algorithm, and combined with the reflection information of rice seedlings and the ground surface under different lighting directions, the normal vector distribution of the rice seedlings and the ground surface is calculated, specifically including:

[0068] Preprocessing operations are performed on the acquired multi-angle image data, including but not limited to image denoising, contrast enhancement, and geometric correction; image denoising algorithms, such as median filtering or Gaussian filtering, are used to remove noise interference in the image; histogram equalization and other methods are used to enhance image contrast, making the features of rice seedlings and the ground surface more obvious; geometric correction is performed on the image using known camera parameters and shooting angle information to eliminate image distortion and ensure image accuracy.

[0069] The reflection information of rice seedlings and ground surface is extracted from the preprocessed image. For each pixel, a reflection function is constructed based on its brightness value under different lighting directions. This reflection function describes the relationship between the brightness of the pixel and the lighting direction. By analyzing the characteristics of the reflection function, key parameters that can reflect the reflection characteristics of rice seedlings and ground surface are extracted.

[0070] Based on the photometric stereo vision algorithm, the normal vector distribution of rice seedlings and the ground surface is calculated by combining the extracted reflection information. Specifically, according to the reflection model, the mathematical relationship between the illumination direction, reflection information and normal vector is established. Using multiple equations provided by multi-angle image data, the mathematical relationship is solved by the least squares method or other optimization algorithms to obtain the normal vector corresponding to each pixel. For the rice seedling area and the ground surface area, the normal vector is calculated separately to accurately reflect the three-dimensional morphological characteristics of both.

[0071] The rice transplanting image recognition server identifies seedling strip data and missing seedling locations based on rice transplanting video data; in some embodiments, the rice transplanting image recognition server identifies seedling strip data and missing seedling locations based on rice transplanting video data, specifically including:

[0072] Analyze the video data of rice transplanting and extract the feature points of rice seedlings;

[0073] Based on the robust regression algorithm, the extracted rice seedling feature points are fitted with row lines to generate the center line of the rice row.

[0074] By combining the fitted center line of the rice row with the spatial coordinates of the feature points, the position information of each rice seedling is calculated;

[0075] Based on the location information of each rice seedling, the seedling strip data and the location of missing seedlings are identified.

[0076] Specifically, robust regression algorithms are used to fit the extracted rice seedling feature points to row lines to generate rice row centerlines. Robust regression algorithms are highly resistant to outliers in the data. In rice transplanting scenarios, there may be individual rice seedlings with abnormal growth or errors in feature point extraction. Robust regression algorithms can reduce the impact of these abnormal factors on row line fitting, making the generated rice row centerlines more accurately reflect the actual distribution of rice seedling rows, providing a reliable basis for subsequent calculation of rice seedling location information. The generated rice row centerlines can accurately describe the direction and position of rice seedling rows, helping to clearly delineate seedling strip areas and providing important spatial references for identifying seedling strip data and missing seedling locations.

[0077] By combining the fitted centerline of the rice row with the spatial coordinates of feature points, the position information of each rice seedling is calculated. This method makes full use of the key data obtained in the previous steps. In this way, the specific position of each rice seedling in the seedling strip can be determined more accurately. It can not only clarify the horizontal and vertical position of the rice seedlings in the row, but also take into account the overall orientation of the rice row, so that the position calculation results are more consistent with the actual situation. Accurate rice seedling position information is the basis for identifying missing seedlings. Only by accurately knowing the position that each rice seedling should be in can we find out in time which positions are missing rice seedlings, thereby improving the accuracy of missing seedling identification.

[0078] Based on the location information of each rice seedling, the identification of seedling strip data and missing seedling locations makes the identification process more logical and accurate. It can clearly determine the boundaries and width of the seedling strip, and at the same time, it can quickly and accurately find the location of missing seedlings, providing timely and accurate information for subsequent replanting operations.

[0079] In some embodiments, the step of fitting row lines to the extracted rice seedling feature points based on a robust regression algorithm to generate rice row center lines specifically includes:

[0080] Density clustering analysis was performed on the extracted rice seedling feature points to obtain a set of effective feature points;

[0081] For the effective set of feature points, the line parameters are iteratively calculated using a robust regression algorithm;

[0082] Based on the row line parameters and image boundary range, generate the center line of the rice rows covering the entire image area.

[0083] Specifically, density clustering analysis is performed on the extracted rice seedling feature points to obtain an effective set of feature points. This step can divide the feature points into different clusters according to their spatial distribution density. In rice transplanting images, the distribution of rice seedling feature points may be quite complex, with some isolated or anomalous points. Through density clustering analysis, feature points belonging to the same rice seedling row or similar areas can be accurately grouped together, effectively removing noise points and outliers, thereby obtaining an effective set of feature points that better reflects the actual distribution of rice seedling rows, thus improving the quality and reliability of the feature points.

[0084] Robust regression algorithms are used to iteratively calculate row line parameters on the effective feature point set. Robust regression algorithms are robust to outliers in the data. In rice transplanting images, even if there are some feature point deviations caused by image acquisition errors, abnormal rice seedling growth, etc., the robust regression algorithm can reduce the impact of these outliers on parameter calculation, and more accurately fit the trend of rice seedling rows, thereby obtaining accurate row line parameters, such as the slope and intercept of the row line. Through iterative calculation, the robust regression algorithm can continuously optimize the row line parameters, making the fitted row line closer to the actual direction of rice seedling rows. Compared with ordinary regression algorithms, it can better handle the uncertainty and noise in the data, and improve the accuracy and stability of row line fitting.

[0085] By generating center lines of rice rows covering the entire image area based on row line parameters and image boundary range, we can comprehensively describe the distribution of rice rows in the entire image. This not only helps to accurately identify the location and range of seedling strips, but also provides comprehensive spatial information for subsequent agricultural operations such as replanting, fertilization, and irrigation.

[0086] Based on the seedling strip data and the location of missing seedlings, the travel path of the seedling replanting machine is planned; in some embodiments, the planning of the travel path of the seedling replanting machine based on the seedling strip data and the location of missing seedlings specifically includes:

[0087] Based on the seedling strip data, extract the spatial orientation, plant spacing, and row spacing parameters of the seedling strip;

[0088] Based on the spatial orientation of the seedling strip, the plant spacing, and the row spacing parameters, the center line of the seedling strip is generated;

[0089] Generate a coordinate system for the seedling strip based on its centerline;

[0090] Map the missing seedling points to the seedling strip coordinate system and determine the lateral offset and longitudinal distance of each missing seedling point relative to the seedling strip;

[0091] The Dijkstra algorithm is used, with the center line of the seedling strip as the reference path, and incorporates the lateral offset and longitudinal distance of each missing seedling point relative to the seedling strip to generate the optimal travel path covering all missing seedling points.

[0092] Specifically, by extracting the spatial orientation, plant spacing, and row spacing parameters from the seedling strip data, the basic characteristics of the seedling strip can be fully and accurately grasped. The spatial orientation information clarifies the extension direction of the seedling strip, which is crucial for the seedling replenishment machine to determine the approximate direction of travel. The plant spacing and row spacing parameters reflect the distribution density and arrangement pattern of rice seedlings, providing key data for the subsequent generation of the seedling strip centerline and coordinate system, ensuring the accurate grasp of the characteristics of the seedling strip.

[0093] Based on the spatial orientation, plant spacing, and row spacing parameters of the seedling strip, a center line of the seedling strip is generated, which can accurately describe the position and shape of the seedling strip in the field. The center line of the seedling strip serves as the core reference line of the seedling strip, providing a benchmark for determining the positional relationship of missing seedlings. On this basis, a seedling strip coordinate system is generated, which converts the spatial position information of the seedling strip into specific coordinate values, so that the position of missing seedlings can be represented by quantified coordinates, greatly improving the accuracy of position description.

[0094] By mapping the missing seedlings to the seedling belt coordinate system and determining the lateral offset and longitudinal distance of each missing seedling relative to the seedling belt, the specific location of the missing seedlings around the seedling belt can be accurately determined. The lateral offset reflects the degree to which the missing seedling is deviated from the center line of the seedling belt, while the longitudinal distance determines the position of the missing seedling in the direction of the seedling belt extension. This precise positional relationship provides key information for the replanting machine to accurately reach the missing seedlings for replanting. By accurately determining the positional relationship of the missing seedlings, the replanting machine can more accurately locate the missing seedlings during its movement, avoiding the problem of inaccurate replanting caused by positional errors, and improving the quality and efficiency of replanting operations.

[0095] Using Dijkstra's algorithm, with the center line of the seedling strip as the reference path, and incorporating the lateral offset and longitudinal distance of the missing seedling points, the optimal travel path covering all missing seedling points can be generated. Dijkstra's algorithm is a classic shortest path algorithm, which can ensure that the total length of the path traversed by the seedling replenishment machine during the journey is minimized, thereby reducing the travel time and energy consumption of the seedling replenishment machine.

[0096] In some embodiments, the method further includes:

[0097] After replanting, the total number of holes in the monitoring area is calculated based on the row spacing and plant spacing of the rice transplanter. The number of rice plants in each hole is analyzed, and the rice emergence rate is calculated.

[0098] Specifically, the total number of holes in the monitoring area is calculated based on the row spacing and plant spacing of the rice transplanter. This method makes full use of the regular parameters of rice transplanting operations. The row spacing and plant spacing are fixed parameters set by the rice transplanter during operation. By using these two parameters to calculate the total number of holes, the theoretically required number of holes in the monitoring area can be obtained quickly and relatively accurately.

[0099] Analyzing the number of rice plants in each hole allows for a detailed understanding of the rice's growth in each hole. By counting the number of rice plants in each hole, it is possible to identify any abnormalities such as missing seedlings or excessive seedlings. This provides detailed data support for evaluating the effectiveness of replanting and the growth status of the rice. For example, if the number of rice plants in a certain hole is significantly less than the normal number, it may indicate that the replanting was incomplete or that the growth of the rice in that hole was affected by other factors.

[0100] By analyzing the total number of planting holes and the number of rice plants in each hole, the rice emergence rate can be calculated, which can accurately assess the effectiveness of replanting operations. The emergence rate is an important indicator for measuring the success of replanting. If the emergence rate is high, it means that the replanting operation has effectively made up for the missing seedlings and ensured the planting density of rice. If the emergence rate is low, it is necessary to further analyze the reasons, such as whether the replanting machine is working properly or whether the quality of the seedlings is up to standard, so as to take timely improvement measures.

[0101] An automatic rice seedling replanting system includes:

[0102] The data acquisition module is used to collect real-time video data of rice transplanting.

[0103] The input module is used to input rice transplanting video data into the rice transplanting image recognition server, which is built based on photometric stereo imaging technology and deep learning algorithms.

[0104] The recognition module is used by the rice transplanting image recognition server to identify the seedling strip data and missing seedling locations based on the rice transplanting video data;

[0105] The path planning module is used to plan the travel path of the seedling replenishment machine based on the seedling strip data and the location of missing seedlings;

[0106] The seedling replenishment module is used by the seedling replenishment machine to replenish seedlings according to the planned travel path.

[0107] A computing device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the above-described automatic rice replanting methods.

[0108] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the above-described automatic rice replanting methods.

[0109] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for automatic rice seedling replanting, characterized in that, Specifically, it includes: Real-time acquisition of video data on rice transplanting; The rice transplanting video data is input into the rice transplanting image recognition server, which is built based on photometric stereo imaging technology and deep learning algorithms. The rice transplanting image recognition server identifies seedling strip data and missing seedling locations based on rice transplanting video data; Based on the seedling strip data and the location of missing seedlings, plan the travel path of the seedling replenishment machine; The seedling replenishment machine replenishes seedlings according to the planned travel path; The rice transplanting image recognition server identifies seedling strip data and missing seedling locations based on rice transplanting video data, specifically including: Analyze the video data of rice transplanting and extract the feature points of rice seedlings; Based on the robust regression algorithm, the extracted rice seedling feature points are fitted with row lines to generate the center line of the rice row. By combining the fitted center line of the rice row with the spatial coordinates of the feature points, the position information of each rice seedling is calculated; Based on the location information of each rice seedling, the seedling strip data and the location of missing seedlings are identified; The rice transplanting image recognition server is built based on photometric stereo imaging technology and deep learning algorithms, specifically including: Collect multi-angle illumination image data of rice transplanting scenes and use photometric stereo imaging technology to construct a three-dimensional surface model; A convolutional neural network model was constructed based on deep learning algorithms. Feature learning and classification training were performed on the three-dimensional surface model to obtain a rice transplanting image recognition model. The rice transplanting image recognition model is deployed to the server to obtain the rice transplanting image recognition server; The construction of a three-dimensional surface model using photometric stereo imaging technology specifically includes: Acquire multi-angle illumination images of the rice transplanting area and collect information on rice seedlings and ground surface reflection under different illumination directions; Based on the photometric stereo vision algorithm, combined with the reflection information of rice seedlings and the ground surface under different lighting directions, the normal vector distribution of rice seedlings and the ground surface is calculated. Based on the normal vector distribution of rice seedlings and the ground surface, a three-dimensional surface model of the rice transplanting area is reconstructed.

2. The automatic rice replanting method according to claim 1, characterized in that, The process of fitting row lines to the extracted rice seedling feature points using a robust regression algorithm to generate rice row center lines specifically includes: Density clustering analysis was performed on the extracted rice seedling feature points to obtain a set of effective feature points; For the effective set of feature points, the line parameters are iteratively calculated using a robust regression algorithm; Based on the row line parameters and image boundary range, generate the center line of the rice rows covering the entire image area.

3. The automatic rice seedling replanting method according to claim 2, characterized in that, The process of planning the replanting machine's path based on seedling strip data and missing seedling locations specifically includes: Based on the seedling strip data, extract the spatial orientation, plant spacing, and row spacing parameters of the seedling strip; Based on the spatial orientation, plant spacing, and row spacing parameters of the seedling strip, the center line of the seedling strip is generated; Generate a coordinate system for the seedling strip based on its centerline; Map the missing seedling points to the seedling strip coordinate system and determine the lateral offset and longitudinal distance of each missing seedling point relative to the seedling strip; The Dijkstra algorithm is used, with the center line of the seedling strip as the reference path, and incorporates the lateral offset and longitudinal distance of each missing seedling point relative to the seedling strip to generate the optimal travel path covering all missing seedling points.

4. The automatic rice replanting method according to claim 3, characterized in that, The method further includes: After the replanting is completed, the total number of holes in the monitoring area is calculated based on the row spacing and plant spacing of the rice transplanter. The number of rice plants in each hole is analyzed, and the rice emergence rate is calculated.

5. An automatic rice seedling replanting system, characterized in that, include: The data acquisition module is used to collect real-time video data of rice transplanting. The input module is used to input rice transplanting video data into the rice transplanting image recognition server, which is built based on photometric stereo imaging technology and deep learning algorithms. The recognition module is used by the rice transplanting image recognition server to identify the seedling strip data and missing seedling locations based on the rice transplanting video data; The path planning module is used to plan the travel path of the seedling replenishment machine based on the seedling strip data and the location of missing seedlings; The seedling replenishment module is used by the seedling replenishment machine to replenish seedlings according to the planned travel path; The rice transplanting image recognition server identifies seedling strip data and missing seedling locations based on rice transplanting video data, specifically including: Analyze the video data of rice transplanting and extract the feature points of rice seedlings; Based on the robust regression algorithm, the extracted rice seedling feature points are fitted with row lines to generate the center line of the rice row. By combining the fitted center line of the rice row with the spatial coordinates of the feature points, the position information of each rice seedling is calculated; Based on the location information of each rice seedling, the seedling strip data and the location of missing seedlings are identified; The rice transplanting image recognition server is built based on photometric stereo imaging technology and deep learning algorithms, specifically including: Collect multi-angle illumination image data of rice transplanting scenes and use photometric stereo imaging technology to construct a three-dimensional surface model; A convolutional neural network model was constructed based on deep learning algorithms. Feature learning and classification training were performed on the three-dimensional surface model to obtain a rice transplanting image recognition model. The rice transplanting image recognition model is deployed to the server to obtain the rice transplanting image recognition server; The construction of a three-dimensional surface model using photometric stereo imaging technology specifically includes: Acquire multi-angle illumination images of the rice transplanting area and collect information on rice seedlings and ground surface reflection under different illumination directions; Based on the photometric stereo vision algorithm, combined with the reflection information of rice seedlings and the ground surface under different lighting directions, the normal vector distribution of rice seedlings and the ground surface is calculated. Based on the normal vector distribution of rice seedlings and the ground surface, a three-dimensional surface model of the rice transplanting area is reconstructed.

6. A computing device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the automatic rice replanting method according to any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steps of the automatic rice replanting method according to any one of claims 1-4.

Images