Intelligent shuttlecock recovery and integrity detection sorting method and system based on artificial intelligence

Abstract

The invention discloses a shuttlecock intelligent recovery and integrity detection sorting method and system based on artificial intelligence, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining collection data, and obtaining target space coordinates and space information according to a YOLO model and a binocular parallax algorithm; inputting the target space coordinates and the space information into an ant colony algorithm to obtain an optimal path; in the process of picking up the shuttlecock along the optimal path, the equipment scans the shuttlecock to be picked up, obtains target data, reconstructs a model of the shuttlecock to be picked up through an infrared scanning and data reverse modeling analysis technology, reconstructs the model, compares the model with a standard model, accurately evaluates the damage degree of the shuttlecock, and performs intelligent classification. The shuttlecocks which cannot be reused are automatically filtered, subsequent arrangement and use of the shuttlecocks are facilitated, meanwhile, the selection and use efficiency of the shuttlecocks in training is improved, the high-standard requirement of a user for the integrity of the shuttlecocks is met, and the training experience is optimized.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and specifically relates to a badminton intelligent recycling, integrity detection and sorting method and system based on artificial intelligence. Background Art

[0002] At present, there have emerged various badminton picking devices and design schemes for transforming badminton and badminton rackets. However, their practicality in actual training scenarios is poor and they are difficult to meet the needs of normalized training. The badminton picking cart that requires manual cooperation, although its device itself has a low manufacturing cost, due to relying on manual operation, it leads to a substantial increase in labor costs, thereby raising the training cost. Moreover, the collected badminton still needs an additional secondary integrity inspection process, consuming time and energy. In addition, although the existing badminton picking robots can complete the picking task, none of them have realized the function of inspecting and classifying the integrity of badminton, and it is impossible to distinguish the integrity degree of badminton synchronously when recycling badminton, which is not conducive to subsequent targeted sorting and use. Therefore, the present invention proposes a badminton intelligent recycling, integrity detection and sorting method and system based on artificial intelligence. Summary of the Invention

[0003] To solve the above technical problems, a badminton intelligent recycling, integrity detection and sorting method and system based on artificial intelligence are provided, which solve the problems of poor practicality in actual training scenarios and the inability to distinguish the integrity degree of badminton synchronously when recycling badminton, which is not conducive to subsequent targeted sorting and use.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A badminton intelligent recycling, integrity detection and sorting method based on artificial intelligence, comprising: S100. Obtain acquisition data, and based on the YOLO model and binocular disparity algorithm, obtain the target space coordinates and spatial information; S200. Input the target space coordinates and spatial information into the ant colony algorithm to obtain the optimal path; S300. During the process of the device picking up badminton along the optimal path, scan the badminton to be picked up, obtain target data, and input the target data into the three-dimensional reconstruction algorithm to obtain a reconstruction model; S400. Obtain a standard model, compare the standard model with the reconstruction model, and classify and save the badminton according to the comparison result.

[0005] Preferably, the obtaining acquisition data, and based on the YOLO model and binocular disparity algorithm, obtaining the target space coordinates and spatial information includes the following steps: S101. Extract the image data in the acquisition data, and construct a two-dimensional coordinate system based on the image data; S102. Based on the image data, combined with the YOLO model, identify the objects in the image, match them with the two-dimensional coordinate system, and obtain the two-dimensional positions of the objects; S103. Obtain historical data, and define the objects in the image based on the historical data to distinguish badminton from obstacles; S104. Based on the two-dimensional position of the badminton, combined with the binocular disparity algorithm, obtain the target space coordinates of the badminton; S105. Based on the two-dimensional position of the obstacle, combined with the binocular disparity algorithm, obtain the space information.

[0006] Preferably, inputting the target space coordinates and the space information into the ant colony algorithm to obtain the optimal path includes the following steps: S201. Obtain the current position information of the device, and mark the current position information as the starting point; S202. Mark the target space coordinates of each badminton as nodes, and construct a matrix chart based on the target space coordinates of the nodes; S203. Based on the space information, obtain the obstacle density of each node; S204. Obtain the distance length from each node to the nearest obstacle; S205. Based on the obstacle density and the distance length, obtain the pheromone of the corresponding node, and fill the matrix chart according to the pheromone to obtain the filled matrix chart; S206. Obtain the heuristic information between the starting point and each node based on the filled matrix chart; S207. Based on the heuristic information between the starting point and each node, combined with the path conversion formula, obtain the transition probability from the starting point to each node; S208. Obtain the optimal path based on the transition probability.

[0007] Preferably, obtaining the optimal path based on the transition probability includes the following steps: S2081. Select the node with the largest transition probability as the starting point of the next path; S2082. Obtain the moving distance of the previous path device, and obtain the path contribution of each device according to the moving distance; S2083. Dynamically adjust the evaporation factor according to the path contribution; S2084. Based on the pheromone, evaporation factor and path contribution of the previous path in the filled matrix chart, update the pheromone of the next path to obtain the new pheromone, and update the filled matrix chart with the new pheromone to obtain the new matrix chart; S2084. Based on the new matrix chart, loop through steps S206 - S2084 until all nodes are involved, and obtain the path after the loop, which is the optimal path.

[0008] Preferably, during the process of the device picking up badminton along the optimal path, the badminton to be picked up is scanned to obtain target data, and the target data is input into a three-dimensional reconstruction algorithm to obtain a reconstruction model, including the following steps: S301. During the process of the device picking up badminton along the optimal path, the badminton to be picked up is scanned layer by layer to obtain point cloud data, and redundant data in the point cloud data is deleted; S302. A three-dimensional coordinate system is constructed with the bottom of the badminton as the origin; S303. The point cloud data is uniformly processed according to the three-dimensional coordinate system to obtain unified data; S304. The unified data is denoised according to Gaussian filtering, and the denoised target data is obtained; S305. According to the target data, combined with the three-dimensional reconstruction algorithm, a reconstruction model of the badminton is reconstructed.

[0009] Preferably, the reconstructing the reconstruction model of the badminton according to the target data and combining with the three-dimensional reconstruction algorithm includes the following steps: S3051. The target data is divided according to the structure of the badminton into two parts: head data and feather data; S3052. The head data is input into the NURBS surface fitting algorithm to obtain a head model; S3053. The feather data is input into a subdivision algorithm to obtain a feather model; S3054. The head model and the feather model are combined to obtain a reconstruction model of the badminton.

[0010] Preferably, the obtaining of the standard model, comparing the standard model with the reconstruction model, and classifying and storing the badminton according to the comparison result include the following steps: S401. Extract multiple feature points to be matched in the reconstruction model according to the SIFT algorithm; S402. Each of the multiple feature points to be matched obtains the nearest neighbor point in the standard model feature point set according to the nearest neighbor search algorithm, obtains the Euclidean distance between the feature point to be matched and the nearest neighbor point, and obtains the average value of the multiple Euclidean distances; S403. A first threshold is constructed according to the quality of the badminton, and the first threshold is compared with the average value; S404. If the average value is greater than the first threshold, the badminton is marked as a non-conforming product; S405. If the average value is less than or equal to the first threshold, the badminton is marked as a possible conforming product and geometric verification is performed; S406. According to the verification result, if the badminton is a non-conforming product, it is conveyed to the scrap storage bin, and if the badminton is a conforming product, it is conveyed to the good product storage bin.

[0011] Preferably, if the average value is less than or equal to the first threshold, the badminton is marked as a possible qualified product, and geometric verification is performed, including the following steps: S4051: Match multiple feature points to be matched with the corresponding nearest neighbor points respectively and construct them into feature point pairs; S4052: Estimate the geometric model parameters based on the feature point pairs in combination with the least squares method; S4053: Project the nearest neighbor points on the standard model onto the reconstructed model according to the geometric model parameters, obtain the position of the projection points on the reconstructed model, and obtain the distance error between the projection points and the feature points to be matched; S4054: Construct a second threshold according to the quality of the badminton, and compare it with the distance error; S4055: If the distance error is greater than the second threshold, it is determined that the badminton fails the geometric verification and is marked as a non-conforming product; S4056: If the distance error is less than or equal to the second threshold, it is determined that the badminton passes the geometric verification and is marked as a qualified product.

[0012] Preferably, an intelligent badminton recycling and integrity detection and sorting system based on artificial intelligence is proposed for implementing the above-mentioned intelligent badminton recycling and integrity detection and sorting method based on artificial intelligence, including: Control module: The control module is used to control the operation of the device and the data transmission within the system; Data acquisition module: The data acquisition module is used to construct a three-dimensional space indoors; Visual recognition module: The visual recognition module is used to recognize indoor objects and match them with the three-dimensional space; Path planning module: The path planning module is used to plan a suitable moving path according to the distribution of badminton indoors; Badminton picking module: The badminton picking module picks up the badminton on the moving path; Model construction module: The model construction module is used to model the picked-up badminton and make a comparison; Badminton storage module: The badminton storage module is used to classify and store the badminton.

[0013] Compared with the prior art, the advantages of the present invention are as follows: The present invention accurately locates the badminton position through infrared detection technology. Combining the badminton position with the ant colony algorithm realizes automatic path planning and ball picking, greatly reducing the manpower requirement, improving the training efficiency, and enhancing the overall practicality of the device; Through infrared scanning and data reverse modeling analysis technology, the model of the badminton to be picked up is reconstructed, and the reconstructed model is compared with the standard model to accurately evaluate the damage degree of the badminton and conduct intelligent classification, automatically filtering out unusable badminton, facilitating the subsequent sorting and use of badminton, and at the same time improving the efficiency of badminton selection and use in training, meeting the high standard requirements of users for the integrity of badminton, and optimizing the training experience. Description of the Drawings

[0014] Figure 1 It is a schematic flow chart of steps S100 - S400 in a badminton intelligent recycling, integrity detection and sorting method and system based on artificial intelligence proposed by the present invention; Figure 2 It is a schematic flow chart of steps S101 - S105 in a badminton intelligent recycling, integrity detection and sorting method and system based on artificial intelligence proposed by the present invention; Figure 3 It is a schematic flow chart of steps S201 - S208 in a badminton intelligent recycling, integrity detection and sorting method and system based on artificial intelligence proposed by the present invention; Figure 4 It is a schematic flow chart of steps S2081 - S2084 in a badminton intelligent recycling, integrity detection and sorting method and system based on artificial intelligence proposed by the present invention; Figure 5 It is a schematic flow chart of steps S301 - S305 in a badminton intelligent recycling, integrity detection and sorting method and system based on artificial intelligence proposed by the present invention; Figure 6 It is a schematic flow chart of steps S3051 - S3054 in a badminton intelligent recycling, integrity detection and sorting method and system based on artificial intelligence proposed by the present invention; Figure 7 It is a schematic flow chart of steps S401 - S406 in a badminton intelligent recycling, integrity detection and sorting method and system based on artificial intelligence proposed by the present invention; Figure 8 It is a schematic flow chart of steps S4051 - S4056 in a badminton intelligent recycling, integrity detection and sorting method and system based on artificial intelligence proposed by the present invention; Figure 9 It is a structural block diagram of a badminton intelligent recycling, integrity detection and sorting method and system based on artificial intelligence proposed by the present invention. Detailed Embodiments

[0015] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments in the following description are only examples, and those skilled in the art can think of other obvious variations.

[0016] Referring to Figures 1-9 as shown, an intelligent badminton recycling and integrity detection and sorting method based on artificial intelligence includes: S100. Obtain the collected data, and based on the YOLO model and the binocular disparity algorithm, obtain the target space coordinates and spatial information; S200. Input the target space coordinates and spatial information into the ant colony algorithm to obtain the optimal path; S300. During the process of the device picking up the badminton along the optimal path, scan the badminton to be picked up, obtain the target data, and input the target data into the three-dimensional reconstruction algorithm to obtain the reconstruction model; S400. Obtain the standard model, compare the standard model with the reconstruction model, and classify and save the badminton according to the comparison result; Those skilled in the art can understand that by performing real-time analysis on the collected image data through the YOLO model, quickly identifying the badminton and obstacles in the image, and determining their positions in the two-dimensional image, combined with the binocular disparity algorithm, converting the two-dimensional image coordinates into three-dimensional space coordinates, obtaining the accurate spatial position information of the badminton and obstacles, constructing an environmental map through the spatial information of the obstacles, providing basic data for subsequent path planning, ensuring that the device can avoid obstacles during the picking process, improving operation safety, generating a three-dimensional spatial information data set including the target (badminton) and obstacles, providing the core input for subsequent path planning, three-dimensional reconstruction and classification decision-making, based on the global search ability of the ant colony algorithm, combined with the spatial coordinates of the badminton and the obstacle distribution, dynamically planning a shortest and collision-free path from the current position to all target badminton, the algorithm can dynamically adjust the path according to the real-time updated environmental information (such as newly added obstacles or target position changes), adapt to the complex and changeable working environment, during the picking process, obtain the point cloud data of the badminton through a high-precision scanning device, extract its surface geometric features, and use a three-dimensional reconstruction algorithm (such as NURBS surface fitting, subdivision algorithm, etc.) to convert the point cloud data into an accurate three-dimensional reconstruction model, the reconstruction model can truly reflect the physical state of the badminton, by comparing the reconstruction model with a preset standard model (such as the geometric model of a defect-free badminton), according to the comparison result, combined with a preset threshold, classify the badminton as "qualified product" or "unqualified product", and automatically convey the badminton to the corresponding storage bin (such as a good product storage bin, a scrap storage bin) according to the classification result, realizing efficient and accurate automatic sorting.

[0017] Such as Figure 2As shown in the figure, the steps of obtaining the acquisition data and obtaining the target space coordinates and space information based on the YOLO model and the binocular disparity algorithm are as follows: S101. Extract the image data from the acquisition data and construct a two-dimensional coordinate system based on the image data; S102. Based on the image data, combined with the YOLO model, identify the objects in the image, match them with the two-dimensional coordinate system, and obtain the two-dimensional positions of the objects; S103. Obtain historical data and define the objects in the image based on the historical data to distinguish badminton from obstacles; S104. Based on the two-dimensional position of the badminton, combined with the binocular disparity algorithm, obtain the target space coordinates of the badminton; S105. Based on the two-dimensional position of the obstacle, combined with the binocular disparity algorithm, obtain the space information; Those skilled in the art can understand that extracting image data from multi-source acquisition data (such as cameras, infrared sensors, etc.), removing noise and redundant information, ensuring the quality of the input data, constructing a two-dimensional coordinate system based on the pixel distribution of the image, and using the high efficiency of the YOLO (You Only Look Once) model to perform real-time detection of the objects in the image, identifying target categories such as badminton and obstacles, matching the detected objects with the two-dimensional coordinate system, and obtaining the two-dimensional coordinates (such as pixel coordinates) of their center points or bounding boxes, providing basic data for subsequent spatial positioning. The high efficiency of the YOLO model ensures the real-time nature of target detection and is suitable for rapid response in dynamic environments. Combining historical data (such as prior knowledge, training samples), performing semantic classification on the detected objects, clearly distinguishing badminton from obstacles (such as court boundaries, other sports equipment, etc.), using the binocular disparity algorithm, calculating the depth information of the badminton through the image differences obtained by two cameras, converting the two-dimensional pixel coordinates into three-dimensional space coordinates, and the space coordinates providing the precise position of the badminton (such as distance, height), providing high-precision data support for subsequent path planning and robotic arm operation. Similarly, based on the binocular disparity algorithm, obtaining the three-dimensional space information of the obstacles (such as size, shape, position), constructing a spatial model of the obstacles, and the space information of the obstacles providing obstacle distribution data for path planning, ensuring that the device can avoid obstacles during movement and improving safety.

[0018] As Figure 3 As shown in the figure, the steps of inputting the target space coordinates and space information into the ant colony algorithm to obtain the optimal path are as follows: S201. Obtain the current position information of the device and mark the current position information as the starting point; S202. Mark the target space coordinates of each badminton as nodes and construct a matrix chart based on the target space coordinates of the nodes; S203. Obtain the obstacle density of each node according to the spatial information; S204. Obtain the distance length from each node to the nearest obstacle; S205. Obtain the pheromone of the corresponding node according to the obstacle density and the distance length, and fill the matrix chart according to the pheromone to obtain the filled matrix chart; S206. Obtain the pheromone of the corresponding node according to the obstacle density and the distance length, and fill the matrix chart according to the pheromone to obtain the filled matrix chart; S207. Obtain the transition probability from the starting point to each node according to the heuristic information between the starting point and each node, combined with the path conversion formula; S208. Obtain the optimal path according to the transition probability; Those skilled in the art can understand that determining the current position of the device as the starting point of path planning ensures that the path planning starts from the actual physical position. The starting point provides a spatial reference for subsequent path calculations, ensuring accurate calculation of the relative positions of all target nodes. Convert the spatial coordinates of each badminton into nodes in path planning to form a target set, providing clear target points for path search. Construct a distance matrix (or adjacency matrix) between nodes to quantify the spatial relationship between nodes, providing a data basis for subsequent path calculations. The matrix chart formally expresses the connection relationship between target nodes, facilitating algorithm processing and optimization. Analyze the spatial information around each node, calculate the obstacle density (such as the number of obstacles per unit area), quantify the environmental complexity of the area where the node is located. The obstacle density is used as a risk indicator for path planning to evaluate the safety and feasibility of different paths. Calculate the distance from each node to the nearest obstacle to quantify the safety margin between the node and the obstacle. At the same time, the distance length is used as a constraint condition for path planning to ensure that the planned path meets the safety distance requirements. Convert the obstacle density and the distance length into pheromone concentration (such as the higher the density and the shorter the distance, the lower the pheromone), quantifying the attraction or repulsion of the node. Extract the heuristic information from the starting point to each node (such as the reciprocal of the distance, the intensity of the pheromone, etc.) from the filled matrix chart to provide directional guidance for path selection. According to the heuristic information and the pheromone concentration, calculate the transition probability from the starting point to each node through the path conversion formula (such as the state transition probability formula of the ant colony algorithm). The probability model combines randomness (exploring new paths) and certainty (utilizing known information), improving the robustness of path search. According to the transition probability, gradually construct the path from the starting point to each node, and finally form a complete path covering all target nodes. Through iterative optimization (such as pheromone update, path evaluation), select the path with the lowest total cost (such as distance, time, risk, etc.) as the optimal path.

[0019] Such as Figure 4As shown in the figure, the process of obtaining the optimal path according to the moving direction probability includes the following steps: S2081. Select the node with the maximum moving direction probability as the starting point of the next path; S2082. Obtain the moving distance of the previous path device, and obtain the path contribution of each device according to the moving distance; S2083. Dynamically adjust the evaporation factor according to the path contribution; S2084. Update the pheromone of the next path according to the pheromone, evaporation factor and path contribution of the previous path in the filling matrix chart, obtain the new pheromone, update the filling matrix chart with the new pheromone, and obtain the new matrix chart; S2084. According to the new matrix chart, loop steps S206 - S2084 until all nodes are involved, and obtain the path after looping, which is the optimal path; Those skilled in the art can understand that, according to the moving direction probability distribution of the current path, select the node with the maximum probability as the next moving target to ensure the optimality of the local path (i.e., most likely to be close to the optimal solution under the current information). By gradually selecting high - probability nodes, dynamically construct a path from the starting point to the target node to form a preliminary path plan. Calculate the moving distance of the device in the previous path to quantify the actual contribution of this path (such as efficiency, cost, etc.) to provide a basis for subsequent pheromone update. The evaporation factor controls the attenuation rate of pheromone. Dynamically adjust the evaporation factor according to the path contribution to avoid excessive accumulation or premature disappearance of pheromone. High - contribution paths correspond to lower evaporation factors (retaining more pheromones), and low - contribution paths correspond to higher evaporation factors (accelerating pheromone attenuation), improving the convergence speed and global search ability of the algorithm. Combine the pheromone residue, evaporation factor and path contribution of the previous path to update the pheromone concentration on the path to form a new pheromone matrix. The pheromone of high - contribution paths is enhanced, and the pheromone of low - contribution paths is weakened. Through positive feedback, guide the ants (path - searching agents) to gather towards better paths. Through the loop iteration of steps S206 (heuristic information acquisition) to S2084 (pheromone update), gradually optimize the path until all target nodes are covered. The iteration process ensures that the algorithm converges to the global optimal or approximate optimal solution through pheromone dynamic adjustment and evaporation factor control.

[0020] As Figure 5 shown, during the process of the device picking up the badminton along the optimal path, scanning the badminton to be picked up, obtaining target data, and inputting the target data into a three - dimensional reconstruction algorithm to obtain a reconstruction model includes the following steps: S301. During the process of the device picking up the badminton along the optimal path, layer - by - layer scan the badminton to be picked up, obtain point cloud data, and delete redundant data in the point cloud data; S302. Construct a three - dimensional coordinate system with the bottom of the badminton as the origin; S303. Uniformly process the point cloud data according to a three-dimensional coordinate system to obtain unified data; S304. Denoise the unified data according to Gaussian filtering and obtain the target data after denoising; S305. Reconstruct the reconstruction model of the badminton by combining the target data with a three-dimensional reconstruction algorithm; Those skilled in the art can understand that by layer-by-layer scanning to obtain the point cloud data of the badminton in real time, it is ensured that the target information can still be stably obtained in complex environments (such as obstacle interference, light changes), redundant data (such as duplicate points, noise points) is deleted, the computational complexity of subsequent processing is reduced, the algorithm efficiency is improved, the point cloud data provides basic geometric information for subsequent three-dimensional reconstruction, the elimination of redundant data can avoid interference and improve the reconstruction accuracy. A coordinate system is constructed with the bottom of the badminton as the origin to unify the reference framework of all point cloud data, ensuring the geometric consistency of three-dimensional reconstruction. The bottom, as a stable feature point of the badminton, is convenient for subsequent algorithm recognition and positioning, reducing errors caused by attitude changes. The point cloud data under different scanning layers or different perspectives is converted into a unified coordinate system to eliminate position deviations and form a complete point cloud model. The unified processing retains the geometric features (such as edges, surfaces) of the point cloud, providing high-quality input for subsequent reconstruction. Gaussian filtering smooths the point cloud data through weighted averaging, effectively removing high-frequency noise (such as scanning errors, environmental interference), improving the data quality. Compared with mean filtering, Gaussian filtering retains edge and detail features while denoising, avoiding geometric distortion caused by over-smoothing. Based on the denoised point cloud data, a geometric model of the badminton is generated through a three-dimensional reconstruction algorithm to restore its three-dimensional shape. The reconstruction model retains the physical characteristics of the badminton (such as feather distribution, sphere contour), providing accurate input for subsequent analysis.

[0021] As Figure 6 shown, the step of reconstructing the reconstruction model of the badminton by combining the target data with a three-dimensional reconstruction algorithm includes the following steps: S3051. Divide the target data according to the structure of the badminton into two parts: head data and feather data; S3052. Input the head data into the NURBS surface fitting algorithm to obtain the head model; S3053. Input the feather data into the subdivision algorithm to obtain the feather model; S3054. Combine the head model and the feather model to obtain the reconstruction model of the badminton; Those skilled in the art can understand that, according to the physical structure of the badminton (the spherical head and the feather branches), the point cloud data is segmented by region to avoid the geometric features of the head and the feathers interfering with each other in subsequent processing. The geometric shapes of the head and the feathers are significantly different (the spherical surface is smooth vs. the feather branches are complex). Separated processing can select the optimal algorithm for different features to improve the reconstruction accuracy. The NURBS (Non-Uniform Rational B-Spline) algorithm is good at fitting smooth surfaces and can accurately restore the spherical shape of the badminton head, avoiding the jaggedness of traditional polygon meshes. The NURBS model is parameterized by control points and weights, facilitating subsequent adjustment of the head shape (such as deformation and scaling) to adapt to different models of badminton. The feather branches are slender, curved, and have many branches. The subdivision algorithm can gradually refine the mesh to accurately capture the topological structure of the feathers. The subdivision algorithm supports multi-resolution modeling and can generate high-precision (for close-range rendering) or low-precision (for long-range display) feather models according to requirements, balancing accuracy and performance. The NURBS model of the head and the subdivision model of the feathers are fused through coordinate alignment, topological connection, etc. to form a complete geometric model of the badminton, ensuring seamless connection of each part. The combined model needs to meet physical constraints (such as the contact relationship between the feather roots and the head surface) to avoid unreasonable phenomena such as penetration or suspension.

[0022] As Figure 7 shown, the steps of obtaining the standard model, comparing the standard model with the reconstructed model, and classifying and storing the badminton according to the comparison result include the following steps: S401. Extract multiple feature points to be matched in the reconstructed model according to the SIFT algorithm; S402. Each of the multiple feature points to be matched obtains the nearest neighbor point in the set of standard model feature points according to the nearest neighbor search algorithm, obtains the Euclidean distance between the feature point to be matched and the nearest neighbor point, and obtains the average value of the multiple Euclidean distances; S403. Construct a first threshold according to the quality of the badminton and compare the first threshold with the average value; S404. If the average value is greater than the first threshold, the badminton is marked as a non-conforming product; S405. If the average value is less than or equal to the first threshold, the badminton is marked as a possible conforming product and geometric verification is performed; S406. According to the verification result, if the badminton is a non-conforming product, it is transported to the non-conforming product storage bin, and if the badminton is a conforming product, it is transported to the conforming product storage bin; Those skilled in the art can understand that the SIFT (Scale-Invariant Feature Transform) algorithm extracts feature points with rotational, scaling, and illumination invariance by detecting key points and calculating their local gradient directions, ensuring the stability of feature points under different postures or environments of the badminton. Each feature point generates a 128-dimensional feature vector to quantitatively describe the geometric and texture information of its surrounding area, providing rich discriminant bases for subsequent matching. Through nearest neighbor search (such as KD-tree acceleration), the nearest neighbor points of the feature points to be matched in the standard model are quickly found, and the Euclidean distance between the two is calculated to quantify the similarity degree of the feature points. The average value of the Euclidean distances of all feature points is calculated to comprehensively reflect the overall deviation between the reconstructed model and the standard model, avoiding the influence of outliers of single feature points on the detection result. According to the industrial standards of badminton (such as dimensional tolerance, shape deviation), a first threshold is constructed to convert the abstract quality requirements into quantifiable numerical indicators. The first threshold can be dynamically adjusted according to the badminton model and detection environment (such as illumination, background) to improve the versatility and adaptability of the detection system. When the average value exceeds the first threshold, it is directly determined that the badminton has significant geometric deviations (such as head deformation, abnormal feather arrangement) without further verification, improving the detection efficiency. When the average value is close to the threshold, there may be minor deviations (such as slightly warped feathers), which are further confirmed through geometric verification to avoid scrapping qualified products due to misjudgment. According to the verification results, the conveying path is automatically controlled to physically separate qualified products from unqualified products, reducing manual intervention.

[0023] As Figure 8 shown, if the average value is less than or equal to the first threshold, the badminton is marked as a possible qualified product, and the geometric verification includes the following steps: S4051: Match multiple feature points to be matched with their corresponding nearest neighbor points respectively and construct them into feature point pairs; S4052: Estimate the geometric model parameters based on the feature point pairs in combination with the least squares method; S4053: Project the nearest neighbor points on the standard model onto the reconstructed model according to the geometric model parameters, obtain the position of the projected points on the reconstructed model, and obtain the distance error between the projected points and the feature points to be matched; S4054: Construct a second threshold according to the quality of the badminton and compare it with the distance error; S4055: If the distance error is greater than the second threshold, it is determined that the badminton fails the geometric verification and is marked as an unqualified product; S4056: If the distance error is less than or equal to the second threshold, it is determined that the badminton passes the geometric verification and is marked as a qualified product; Those skilled in the art can understand that by corresponding the feature points to be matched in the reconstructed model with the nearest neighbor points in the standard model one by one to form feature point pairs, it provides basic data for subsequent geometric transformation. The feature point pairs clarify the corresponding relationship of the local regions between the two models, facilitating the quantification of the geometric differences between them through mathematical methods (such as the least squares method). By fitting the feature point pairs with the least squares method, the geometric transformation parameters (such as rotation matrix, translation vector, scaling factor) between the reconstructed model and the standard model are estimated to quantify the global deviation between the two. The least squares method can minimize the sum of the squares of the geometric errors of all feature point pairs to ensure the robustness and global optimality of the transformation parameters. After projecting the nearest neighbor points of the standard model to the reconstructed model through geometric transformation, the Euclidean distance between the projected points and the original feature points to be matched is calculated, which intuitively reflects the geometric deviation of the two models in the local region. By calculating the distance errors of multiple feature point pairs, the deviation distribution of the reconstructed model in each region can be analyzed to identify potential structural defects (such as uneven feather arrangement, head deformation). According to the industrial standards of badminton (such as dimensional tolerance, shape deviation), a second threshold is constructed to convert the subjective quality requirements into quantifiable numerical indicators, improving the objectivity of decision-making. The second threshold is dynamically constructed based on the quality of badminton, reflecting the industrial-level precision requirements and ensuring a strong correlation between the threshold and the actual quality. When the distance error exceeds the threshold, it is directly determined that the badminton has local geometric deviation (such as bent feathers, asymmetric head), without the need for global analysis. When the distance error is within the threshold, it is determined that the geometric verification of the badminton passes, and the qualified products enter the storage of good products, optimizing resource utilization, reducing manual intervention, and realizing automatic sorting.

[0024] As Figure 9 shown, a badminton intelligent recycling, integrity detection and sorting system based on artificial intelligence is proposed for implementing the above-mentioned badminton intelligent recycling, integrity detection and sorting method based on artificial intelligence, including: Control module: The control module is used to control the operation of the device and the data transmission within the system; Data acquisition module: The data acquisition module is used to construct a three-dimensional space indoors; Visual recognition module: The visual recognition module is used to recognize indoor objects and match them with the three-dimensional space; Path planning module: The path planning module is used to plan a suitable moving path according to the distribution of badminton indoors; Badminton picking module: The badminton picking module picks up the badminton on the moving path; Model construction module: The model construction module is used to model the picked-up badminton and make comparisons; Badminton storage module: The badminton storage module is used to classify and store the badminton.

[0025] In summary, the advantages of the present invention are as follows: By means of infrared scanning and data reverse modeling analysis technology, the model of the badminton to be picked up is reconstructed, the reconstructed model is compared with the standard model, the damage degree of the badminton is accurately evaluated, and intelligent classification is carried out to automatically filter out the unusable badminton, which is convenient for the subsequent sorting and use of badminton. At the same time, the efficiency of badminton selection and use in training is improved, meeting the high-standard requirements of users for the integrity of badminton and optimizing the training experience.

[0026] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and what is described in the above embodiments and the specification is only the principle of the present invention. Without departing from the spirit and scope of the present invention, various changes and improvements will occur to the present invention, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent badminton recycling, integrity detection and sorting method based on artificial intelligence, characterized in that, Including: S100. Obtain the collected data, and based on the YOLO model and the binocular disparity algorithm, obtain the target space coordinates and spatial information; S200. Input the target space coordinates and spatial information into the ant colony algorithm to obtain the optimal path; S300. During the process of the device picking up the badminton along the optimal path, scan the badminton to be picked up to obtain the target data, and input the target data into the three-dimensional reconstruction algorithm to obtain the reconstruction model; S400. Obtain the standard model, compare the standard model with the reconstruction model, and classify and save the badminton according to the comparison result.

2. The intelligent badminton recycling, integrity detection and sorting method based on artificial intelligence according to claim 1, characterized in that: The obtaining of the collected data, and based on the YOLO model and the binocular disparity algorithm, obtaining the target space coordinates and spatial information includes the following steps: S101. Extract the image data from the collected data, and construct a two-dimensional coordinate system based on the image data; S102. Based on the image data, combined with the YOLO model, identify the objects in the image, and match them with the two-dimensional coordinate system to obtain the two-dimensional positions of the objects; S103. Obtain the historical data, and define the objects in the image based on the historical data to distinguish between badminton and obstacles; S104. Based on the two-dimensional position of the badminton, combined with the binocular disparity algorithm, obtain the target space coordinates of the badminton; S105. Based on the two-dimensional position of the obstacle, combined with the binocular disparity algorithm, obtain the spatial information.

3. The intelligent badminton recycling, integrity detection and sorting method based on artificial intelligence according to claim 2, characterized in that: The inputting of the target space coordinates and spatial information into the ant colony algorithm to obtain the optimal path includes the following steps: S201. Obtain the current position information of the device, and mark the current position information as the starting point; S202. Mark the target space coordinates of each badminton as nodes, and construct a matrix chart based on the target space coordinates of the nodes; S203. Based on the spatial information, obtain the obstacle density of each node; S204. Obtain the distance length from each node to the nearest obstacle; S205. Based on the obstacle density and the distance length, obtain the pheromone of the corresponding node, and fill the matrix chart according to the pheromone to obtain the filled matrix chart; S206. Obtain the heuristic information between the starting point and each node based on the filled matrix chart; S207. Based on the heuristic information between the starting point and each node, combined with the path conversion formula, obtain the transition probability between the starting point and each node; S208. Obtain the optimal path based on the transition probability.

4. A badminton intelligent recycling and integrity detection and sorting method based on artificial intelligence according to claim 3, characterized in that: The obtaining of the optimal path based on the transition probability includes the following steps: S2081. Select the node with the largest transition probability as the starting point of the next path; S2082. Obtain the moving distance of the device in the previous path, and obtain the path contribution of each device according to the moving distance; S2083. Dynamically adjust the evaporation factor based on the path contribution; S2084. Based on the pheromone, evaporation factor and path contribution of the previous path in the filled matrix chart, update the pheromone of the next path to obtain the new pheromone, and update the filled matrix chart with the new pheromone to obtain the new matrix chart; S2084. Based on the new matrix chart, loop through steps S206 - S2084 until all nodes are involved, and obtain the path after the loop, which is the optimal path.

5. The intelligent badminton recycling, integrity detection and sorting method based on artificial intelligence according to claim 4, characterized in that: During the process of the device picking up badminton along the optimal path, the badminton to be picked up is scanned to obtain target data, and the target data is input into a 3D reconstruction algorithm to obtain a reconstruction model, which includes the following steps: S301. During the process of the device picking up badminton along the optimal path, the badminton to be picked up is scanned layer by layer to obtain point cloud data, and redundant data in the point cloud data is deleted; S302. A 3D coordinate system is constructed with the bottom of the badminton as the origin; S303. The point cloud data is unified according to the 3D coordinate system to obtain unified data; S304. The unified data is denoised according to Gaussian filtering, and the denoised target data is obtained; S305. According to the target data, combined with the 3D reconstruction algorithm, a reconstruction model of the badminton is rebuilt.

6. The intelligent badminton recycling, integrity detection and sorting method based on artificial intelligence according to claim 5, characterized in that: The step of rebuilding the reconstruction model of the badminton according to the target data and combining with the 3D reconstruction algorithm includes the following steps: S3051. The target data is divided according to the structure of the badminton into two parts: head data and feather data; S3052. The head data is input into the NURBS surface fitting algorithm to obtain a head model; S3053. The feather data is input into the subdivision algorithm to obtain a feather model; S3054. The head model and the feather model are combined to obtain a reconstruction model of the badminton.

7. An intelligent badminton recycling, integrity detection and sorting method based on artificial intelligence according to claim 6, characterized in that: The steps of obtaining a standard model, comparing the standard model with the reconstruction model, and classifying and storing the badminton according to the comparison result include the following steps: S401. Multiple feature points to be matched in the reconstruction model are extracted according to the SIFT algorithm; S402. Each of the multiple feature points to be matched obtains the nearest neighbor point in the set of standard model feature points according to the nearest neighbor search algorithm, obtains the Euclidean distance between the feature point to be matched and the nearest neighbor point, and obtains the average value of the multiple Euclidean distances; S403. A first threshold is constructed according to the quality of the badminton, and the first threshold is compared with the average value; S404. If the average value is greater than the first threshold, the badminton is marked as a non-conforming product; S405. If the average value is less than or equal to the first threshold, the badminton is marked as a possible qualified product and geometric verification is performed; S406. According to the verification result, if the badminton is a non-conforming product, it is conveyed to the waste storage bin, and if the badminton is a qualified product, it is conveyed to the good product storage bin.

8. An intelligent badminton recycling, integrity detection and sorting method based on artificial intelligence according to claim 6, characterized in that: The step of, if the average value is less than or equal to the first threshold, marking the badminton as a possible qualified product and performing geometric verification includes the following steps: S4051. Each of the multiple feature points to be matched is matched with the corresponding nearest neighbor point and constructed into a feature point pair; S4052. According to the feature point pair and combined with the least squares method, the geometric model parameters are estimated; S4053. According to the geometric model parameters, the nearest neighbor points on the standard model are projected onto the reconstruction model to obtain the position of the projection points on the reconstruction model, and the distance error between the projection points and the feature points to be matched is obtained; S4054. A second threshold is constructed according to the quality of the badminton, and the second threshold is compared with the distance error; S4055. If the distance error is greater than the second threshold, it is determined that the badminton fails the geometric verification and is marked as a non-conforming product; S4056. If the distance error is less than or equal to the second threshold, it is determined that the badminton passes the geometric verification and is marked as a qualified product.

9. An intelligent badminton recycling and integrity detection and sorting system based on artificial intelligence, which is used to implement the intelligent badminton recycling and integrity detection and sorting method based on artificial intelligence as described in claims 1-8, and is characterized in that, Including: Control module: The control module is used to control the operation of the device and data transmission within the system; Data acquisition module: The data acquisition module is used to construct a three-dimensional space indoors; Visual recognition module: The visual recognition module is used to recognize indoor objects and match them with the three-dimensional space; Path planning module: The path planning module is used to plan a suitable moving path according to the distribution of badminton indoors; Badminton picking module: The badminton picking module picks up the badminton on the moving path; Model construction module: The model construction module is used to model and compare the picked badminton; Badminton storage module: The badminton storage module is used to classify and store the badminton.

Images