Intelligent feeding, packaging and sorting system and method based on high-precision visual identification

Abstract

The invention discloses an intelligent feeding, packaging and sorting system and method based on high-precision visual identification, and the method comprises the steps: a sensor cluster carries out the all-dimensional scanning of a to-be-sorted object, and collects multi-dimensional data; the data is processed in real time through an edge calculation module, and position information, category information and size information of the article are generated; identifying and classifying the articles based on a deep learning model, and if the articles with irregular shapes or damaged surfaces are identified, adjusting an identification strategy through a self-adaptive algorithm; according to the identification result of the articles, a correct path is selected through the intelligent sorting system, and the articles are conveyed to a designated sorting area; according to the size, the shape and the type of the article, a packaging mode is automatically adopted, and the article is packaged through an automatic packaging system; and after being packaged, the articles are conveyed to a delivery area to be distributed or further processed.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent feeding, packaging and sorting based on high-precision visual recognition, and specifically relates to an intelligent feeding, packaging and sorting system and method based on high-precision visual recognition. Background Art

[0002] The intelligent feeding, packaging, and sorting system based on high-precision visual recognition is an intelligent sorting system that combines computer vision, machine learning, and automation technologies. It is widely used in logistics, warehousing, and production lines. Its basic function is to accurately identify and locate items through high-precision visual recognition technology, automatically completing the feeding, packaging, and sorting of items. It uses cameras or sensors to capture image data of items and uses image processing algorithms to analyze and identify item characteristics such as size, shape, color, and label to ensure accurate identification. Automatic feeding devices deliver items from one location to the next, ensuring that items are delivered to the sorting area in order and at the correct speed. Based on the type, size, and shape of the items, the system automatically selects the appropriate packaging method, such as boxes, bags, or other containers, to ensure that the items are not damaged during transportation. Based on the item characteristics, the intelligent sorting system classifies the items according to preset rules and delivers them to the corresponding area or destination. This can be achieved through methods such as robotic arms and conveyor belts.

[0003] However, although the intelligent feeding, packaging and sorting system based on high-precision visual recognition has many advantages, it still has some defects and challenges. The existing system is easily affected by factors such as lighting changes and object occlusion, resulting in reduced recognition accuracy. Objects with complex shapes or surface damage are difficult to accurately identify. In complex environments, the system's processing speed may be limited, affecting sorting efficiency. The hardware cost of the existing system is relatively high, especially the cost of visual sensors and processing units. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the purpose of the present invention is to provide an intelligent feeding, packaging and sorting system and method based on high-precision visual recognition. By combining multimodal sensors and deep learning algorithms, the recognition strategy is automatically adjusted according to the morphological characteristics of the objects. In particular, it has strong fault tolerance for objects with irregular shapes and surface damage. By combining multiple cameras with a 360-degree panoramic vision system, objects can be accurately scanned from multiple angles, overcoming the recognition blind spot problem under a single perspective, and having significant advantages in processing irregular objects.

[0005] The technical solution adopted by the present invention to solve its technical problem is:

[0006] Intelligent feeding, packaging and sorting system based on high-precision visual recognition, including:

[0007] The visual sensor module is used for high-resolution cameras to capture images of objects and perform preliminary recognition through image processing algorithms;

[0008] Multimodal sensor modules, which use lidar, infrared sensors, and ultrasonic sensors to complement visual sensors and ensure recognition in complex environments;

[0009] The computing unit module includes an edge computing module for real-time processing and analysis of sensor data.

[0010] The item handling module includes an automatic feeding device, a robotic arm, and a conveyor belt to transport items from the input port to the designated location and perform sorting operations according to sorting instructions;

[0011] The packaging module is used to adopt relevant packaging methods based on the item identification results and output them to the shipping port.

[0012] The intelligent feeding, packaging and sorting method based on high-precision visual recognition includes the following steps:

[0013] Step 1: The sensor cluster performs an all-round scan of the items to be sorted and collects multi-dimensional data;

[0014] Step 2: The data is processed in real time by the edge computing module to generate the item’s location, category, and size information;

[0015] Step 3: Identify and classify objects based on the deep learning model. If an object with an irregular shape or surface damage is identified, the recognition strategy is adjusted through an adaptive algorithm.

[0016] Step 4: Based on the identification results of the items, the intelligent sorting system selects the correct path and sends the items to the designated sorting area;

[0017] Step 5: Automatically adopt packaging methods based on the size, shape and type of the items and pack them through the automatic packaging system;

[0018] Step 6: After the items are packaged, they are sent to the shipping area for distribution or further processing.

[0019] As a preferred method, the sensor cluster performs an all-round scanning of the items to be sorted and collects multi-dimensional data in the following manner:

[0020] Set up multiple sensors S1, S2, ..., S n , each sensor outputs different observation data D1, D2, ..., D n , then the goal of multidimensional data fusion is to fuse these data;

[0021]

[0022] Among them, D i For the The measurement data of each sensor;

[0023] w i For sensor S i The weight of

[0024] D fused is the fused multidimensional data;

[0025] When performing omnidirectional scanning, a lidar or ultrasonic sensor is used to estimate the distance and position of an object. If the sensor S emits a laser signal and receives a reflected signal, the distance d to the object is calculated using the following formula:

[0026]

[0027] Where d is the distance from the object to the sensor;

[0028] c is the speed of light;

[0029] Δt is the time difference from when the sensor emits the laser to when it receives the reflected signal;

[0030] For visual sensors, the acquired image data is used to calculate the depth information of the object through stereo vision or structured light methods. The depth calculation formula of the stereo vision system is as follows:

[0031]

[0032] Where d is the depth of the object;

[0033] f is the focal length of the camera;

[0034] B is the baseline of the camera;

[0035] x is the disparity corresponding to the object in the image;

[0036] For the three-dimensional spatial positioning of a multimodal sensor cluster, a transformation matrix is used to convert data between different sensors. The sensor coordinate system is set as C1, and the position of the object in this coordinate system is P1. The conversion to the object position P2 in another coordinate system C2 is performed through the following matrix operation:

[0037] P2=T·P1

[0038] Where T is the transformation matrix from C1 to C2;

[0039] P1 is the position of the object in the C1 coordinate system;

[0040] P2 is the position of the object in the C2 coordinate system.

[0041] Preferably, the data is processed in real time by the edge computing module to generate the location information, category information and size information of the item as follows:

[0042] If a laser radar is used to obtain the distance information d and angle information θ of an object, the three-dimensional coordinates (x, y, z) of the object are calculated using the following formula:

[0043] x=d·cos(θ)

[0044] y=d·sin(θ)

[0045] z=f(d,θ)

[0046] Where d is the distance between the object and the sensor;

[0047] θ is the angle of laser scanning;

[0048] For stereo vision systems, the parallax Δx of the binocular cameras is used to calculate the depth of the object:

[0049]

[0050] Where f is the focal length of the camera;

[0051] B is the baseline of the two cameras;

[0052] Δx is the parallax of the object between the two images;

[0053] The edge computing module uses a deep learning model to identify and classify objects. The image data collected by the sensor is processed by the deep learning model to obtain a feature vector f, and the category information of the object is calculated through the softmax layer.

[0054] The calculation formula for category c is:

[0055] c=softmax(W·f+b)

[0056] Among them, f is the feature vector after processing by the convolution layer;

[0057] W is the weight matrix of the classification model;

[0058] b is the bias of the classification model;

[0059] Softmaxsoftmax is an activation function used to output the probability of each category;

[0060] The calculation formula of the softmax function is:

[0061]

[0062] where zi is the first feature vector elements;

[0063] The length, width, and height of an object are calculated by extracting the depth information and object outline. The outline information of the object is obtained by the visual sensor, and the size is estimated by the depth information. For the stereo vision system, the actual size of the object is calculated by the disparity map and depth information.

[0064] If the depth information of the object is d(x, y), the size of the object can be estimated as follows:

[0065]

[0066] Where w is the width of the object, x1 and x2 are the horizontal boundaries of the object in the image;

[0067] h is the height of the object, y1 and y2 are the vertical boundaries of the object on the image;

[0068] d is the depth information of the object;

[0069] f is the focal length of the camera.

[0070] Preferably, objects are identified and classified based on a deep learning model. If an object with an irregular shape or surface damage is identified, the method of adjusting the identification strategy through an adaptive algorithm is as follows:

[0071] A convolutional neural network is used to identify objects. After processing through each layer of the network, the feature vector f of the object is obtained, and the category c of the object is calculated using the softmax function:

[0072] c=softmax(W·f+b)

[0073] Where f is the feature vector of the item;

[0074] W is the weight matrix of the classification model;

[0075] b is the bias of the classification model;

[0076] Softmaxsoftmax is an activation function used to calculate the probability of each category;

[0077] Items with irregular shapes or surface damage have shape characteristics that differ from typical items. These are detected and adaptive adjustments are triggered using the following methods:

[0078] The object detection model is used to measure the bounding box of the object and calculate the contour features of the object. The calculation formula of the contour difference is an error metric based on shape matching, which is:

[0079]

[0080] Among them, C i is the contour point of item i;

[0081] C idcal For the ideal object outline;

[0082] n is the number of contour points;

[0083] The surface of the object is segmented by image segmentation method, and the surface damage is detected. If the area of the damaged area A damage Greater than a threshold T damage , it means the item is damaged;

[0084] The damage area calculation formula is:

[0085] ParseError:KaTeX parse error:Can't usefunction′\(′in math mode atposition 36:...\sum_{i=1}^{n}\(C_{\text{damage...

[0086] Among them, C damagc (i) Pixels representing the damaged area;

[0087] It is the marking function of the damaged area. If the pixel belongs to the damaged area, it returns 1; otherwise, it returns 0.

[0088] T damage is the threshold of the damage area;

[0089] When irregular shapes or surface damage are detected, the recognition strategy will be adjusted. The adaptive algorithm dynamically adjusts the classification model or feature extraction strategy. When irregular shapes or damage are detected, the model is dynamically adjusted through incremental learning. The current model parameter is set to θ, and the new parameter after incremental learning is updated to θ′. The update formula is:

[0090] θ′=θ+△θ

[0091] Where Δθ is the gradient or update amount calculated based on the current new data;

[0092] For damaged or irregular items, dynamically adjust the feature extraction strategy to improve the classification results by weighting the importance of different features. Set each feature f i The contribution to classification is w i , when damage or irregular shapes are detected, the weight w is adaptively adjusted i To improve the sensitivity of the model, the weight update formula is:

[0093] w′i =w i ·(1+λ·△f i )

[0094] Among them, w′ i is the updated feature weight;

[0095] λ is the learning rate or adjustment coefficient;

[0096] Δf i is the difference between current characteristics and ideal characteristics;

[0097] When irregular shapes or damage are identified, more training data with damage or irregular shapes is generated through data augmentation technology;

[0098] Set the original item image to I original , the enhanced image I is obtained by rotating or adding noise augmcntcd :

[0099] I angmcntcd =f angmcntation (I original )

[0100] Among them, f augmentation is the data enhancement function.

[0101] As a preferred method, based on the identification results of the items, the intelligent sorting system selects the correct path and sends the items to the designated sorting area as follows:

[0102] Set the sorting area S corresponding to each item category c , that is, the sorting area to which each category of items should be sent. The sorting area S corresponding to each category c is defined as follows c :

[0103] S c =Mapping(c)

[0104] Where Mapping(c) is a mapping function used to map item category c to a specified sorting area;

[0105] In the intelligent sorting system, the path planning algorithm is used to determine the path from the current sorting position to the target sorting area. The sorting system is set up to include multiple sorting areas and paths. The cost of each path is p ij , that is, the cost from path i to path j;

[0106] According to the item category recognition result c, the intelligent sorting system selects the sorting area according to the optimal path. The mathematical formula of the path selection process is expressed as:

[0107]

[0108] Where P(c) is the optimal path for item category c;

[0109] p ij is the path cost from node i to node j;

[0110] N is the number of all path nodes;

[0111] The sorting system has the ability to adjust dynamically. When some paths are blocked or congested, the system will reselect the path and set the current state of the system to S. t , by evaluating the load of the current path, the path selection is updated through an adjustment function:

[0112]

[0113] Where λ is the adjustment factor;

[0114] load(i) is the load of path i;

[0115] After adopting the path through the path planning algorithm, the system sends the items from the current location to the target sorting area S c , set the current position of the item to P start , the target sorting area is S c , the final path is chosen as P optimal , the delivery process of the items is expressed as:

[0116] P optimal =PathPlanning(P start ,S c ).

[0117] As a preferred method, the packaging method is automatically adopted according to the size, shape and type of the items, and the packaging method is carried out by the automatic packaging system:

[0118] Set the size of the item to D = (l, w, h), the shape to Shape(D), and the type to T;

[0119] Depending on the size, shape and type of items, the automatic packaging system adopts packaging methods, which include box packaging, bag packaging, and customized packaging;

[0120] The packaging method P is decided by the following rules:

[0121] P=ChoosePackaging(D,Shape(D),T)

[0122] The packaging material is based on the weight of the item. itcm , shape Shape (D) and type T are determined, and the selection function of packaging materials is established as:

[0123] M=ChooseMaterial(T,w itcm , Shape(D))

[0124] Where M is the selected packaging material;

[0125] The calculation of packaging size depends on the size and shape of the item and the packaging method used. For rectangular items, the size of the packaging box will be based on the size of the item plus some additional margins. For spherical items, the size of the packaging box will cover the diameter of the sphere.

[0126] If the size of the item is D = (l, w, h), then the size of the packaging box P box =(l+w+h+);

[0127] For irregular items, the package size is determined by calculating the circumscribed rectangle or circle of the item:

[0128] P box =BoundingBox(D)

[0129] The automatic packaging system uses robotic arms and conveyor belt equipment to pack. The packaging process includes the following steps:

[0130] Locates items and places them in the packing area based on their size and shape;

[0131] Automatically cut the packaging material and form it into a suitable shape according to the selected packaging method and material;

[0132] Place the item in the packing material, making sure it is securely wrapped;

[0133] Close the packaging mouth and complete the packaging process;

[0134] The packaging process is expressed by the following control formula:

[0135] Pack(D,P,M)=PositionItem(D)+CutMaterial(M)+SealPackaging(P).

[0136] Preferably, after the items are packaged, they are sent to the shipping area for distribution or further processing as follows:

[0137] After the items are packed, they are marked as ready for processing or delivery, which is represented by P packcd , set up shipping area or processing area as Q ship ;

[0138] The path selection for transferring items from the packaging area to the shipping area usually involves path planning, setting up a sorting system or a transportation system using multiple paths Ppath , each path has a cost of time, distance or resource consumption, expressed as C(P path );

[0139] The calculation formula for path selection is:

[0140]

[0141] Among them, P optimal The path to be adopted;

[0142] C(P path ) is the path cost;

[0143] In the automatic distribution system, the scheduling of items will be based on multiple factors, including inventory, transportation resources and distribution priority. The priority of each item is set as P. priority , there are multiple delivery tasks in the shipping area, and the system will schedule them to minimize the overall delivery time or cost. The scheduling problem is expressed as the following optimization problem:

[0144]

[0145] Among them, C task，i is the cost of the i-th task;

[0146] P priority，i is the priority of task i;

[0147] During the distribution or further processing, the inventory is updated and the inventory quantity of each item is set as I item , when items are moved from the packing area to the shipping area, the inventory quantity is reduced by the quantity of items processed;

[0148] The inventory update formula is:

[0149] I itcm =I itcm -QuantityShipped

[0150] Among them, I itcm is the current inventory quantity of the item;

[0151] QuantityShipped is the number of items shipped from inventory;

[0152] Once the item arrives at the shipping area, depending on the type of item, it will be delivered directly to the customer or enter further processing. The following formula is used to represent the further processing flow:

[0153] NextStage(P packed , Q ship )=ChooseProcess(Ppacked ,Q ship )

[0154] Among them, NextStage(P acked , Q ship ) means after the items are delivered from the packaging area to the shipping area;

[0155] ChooseProcess(P packed , Q ship ) is to select the appropriate processing or delivery method based on the category or demand of the items.

[0156] Another technical problem to be solved by the present invention is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the program, it implements an intelligent feeding, packaging, and sorting system and method based on high-precision visual recognition as described above.

[0157] Another technical problem to be solved by the present invention is to provide a computer-readable storage medium on which a computer program is stored, which, when executed by a processor, implements an intelligent feeding, packaging and sorting system and method based on high-precision visual recognition.

[0158] The beneficial effects of the present invention are:

[0159] Through all-round scanning and real-time data processing, relevant information of items can be instantly acquired and analyzed, ensuring a rapid response to the entire sorting process; through automatic recognition and classification by deep learning models, the system can accurately judge the characteristics of items, especially for items with irregular shapes or damage. The system adjusts the strategy through adaptive algorithms to ensure the accuracy of recognition and reduce human errors and omissions; based on the recognition results of the items, the intelligent sorting system automatically selects the optimal path, which not only improves the accuracy of sorting, but also reduces the misplacement and misdelivery of items, and improves the accuracy and efficiency of sorting; the most suitable packaging method is automatically selected according to the size, shape and type of the items, which improves packaging efficiency and reduces human intervention through automation, thereby reducing human errors and The risk of damage caused by improper packaging; the multi-dimensional data obtained by the sensor cluster provides detailed item information, and real-time processing through edge computing can reduce delays, making the entire processing process smoother, especially when handling large quantities of items, it can maintain efficiency and accuracy; from item scanning, identification, sorting to packaging, each link can be completed automatically, greatly reducing reliance on manual operations; this solution combines deep learning models with adaptive algorithms, enabling the system to cope with various complex item characteristics, including irregular shapes, different sizes and surface damage, greatly enhancing the system's adaptability; intelligent sorting and packaging systems can reduce inventory errors, ensure inventory accuracy, and avoid customer complaints caused by incorrect delivery or improper packaging of items. BRIEF DESCRIPTION OF THE DRAWINGS

[0160] Figure 1 This is a flow chart of the intelligent feeding, packaging and sorting system based on high-precision visual recognition of the present invention. DETAILED DESCRIPTION

[0161] The principles and features of the present invention are described below. The examples provided are intended to illustrate the present invention only and are not intended to limit the scope of the present invention. The following paragraphs describe the present invention in more detail by way of example. The advantages and features of the present invention will become more apparent from the following description and claims.

[0162] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains. The terms used in this specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0163] Example

[0164] The technical solution adopted by the present invention to solve its technical problem is:

[0165] Intelligent feeding, packaging and sorting system based on high-precision visual recognition, including:

[0166] The visual sensor module is used for high-resolution cameras to capture images of objects and perform preliminary recognition through image processing algorithms;

[0167] Multimodal sensor modules, which use lidar, infrared sensors, and ultrasonic sensors to complement visual sensors and ensure recognition in complex environments;

[0168] The computing unit module includes an edge computing module for real-time processing and analysis of sensor data.

[0169] The item handling module includes an automatic feeding device, a robotic arm, and a conveyor belt to transport items from the input port to the designated location and perform sorting operations according to sorting instructions;

[0170] The packaging module is used to adopt relevant packaging methods based on the item identification results and output them to the shipping port.

[0171] The system uses a combination of high-precision visual sensors and multimodal sensors to accurately identify items in complex environments, reduce the chances of misidentification and missorting, and greatly improve sorting efficiency and accuracy; the edge computing module processes sensor data in real time, reducing data transmission and processing delays and ensuring rapid system response; the solution combines a variety of sensors to adapt to different types of items and environmental conditions; the automated feeding, sorting and packaging processes can be flexibly adjusted according to the characteristics of the items, adapting to different item types and processing requirements; the entire process of item processing is achieved through an automated system, greatly reducing manual operations and the probability of human error; robotic arms and automatic sorting The picking equipment can accurately perform sorting tasks, reducing the possibility of item loss or incorrect delivery; the entire sorting and packaging process is highly automated, saving a lot of time and labor costs; the multimodal sensor provides a redundant design. When the visual sensor fails or the environmental conditions are not suitable, the lidar, infrared sensor and ultrasonic sensor can continue to work to ensure the stability and reliability of the system; the edge computing module does not need to rely on the cloud server when processing data, reducing the impact of network delays on system response and enhancing real-time response capabilities; the fast and accurate sorting and packaging process can ensure that items are delivered to customers on time and accurately, thereby improving customer satisfaction.

[0172] The intelligent feeding, packaging and sorting method based on high-precision visual recognition includes the following steps:

[0173] Step 1: The sensor cluster performs an all-round scan of the items to be sorted and collects multi-dimensional data;

[0174] Step 2: The data is processed in real time by the edge computing module to generate the item’s location, category, and size information;

[0175] Step 3: Identify and classify objects based on the deep learning model. If an object with an irregular shape or surface damage is identified, the recognition strategy is adjusted through an adaptive algorithm.

[0176] Step 4: Based on the identification results of the items, the intelligent sorting system selects the correct path and sends the items to the designated sorting area;

[0177] Step 5: Automatically adopt packaging methods based on the size, shape and type of the items and pack them through the automatic packaging system;

[0178] Step 6: After the items are packaged, they are sent to the shipping area for distribution or further processing.

[0179] Edge computing makes data processing more efficient and reduces system latency, thereby improving the response speed of operations such as item identification, sorting, and packaging. Through deep learning models and adaptive algorithms, the system can handle diverse and complex item features and identify irregular items or surface damage. Automatic packaging and automatic sorting further improve operational efficiency and accuracy, reducing the cost and risk of manual operations. The system can automatically select the most appropriate processing method based on the characteristics of different items, including packaging method and sorting path. Through the collaborative work of multiple sensors, the system can obtain more comprehensive and accurate item data. The automated sorting and packaging process reduces dependence on manual labor, thereby reducing labor costs and improving production and processing efficiency. Efficient and accurate sorting and packaging not only speeds up processing, but also ensures that items are delivered to customers intact, thereby improving customer satisfaction. The system has strong scalability and can add new sensor types, improve deep learning models, or upgrade automation equipment as needed to ensure that it can adapt to changing market demands and technological developments.

[0180] The sensor cluster performs an all-around scan of the items to be sorted and collects multi-dimensional data in the following ways:

[0181] Set up multiple sensors S1, S2, ..., S n , each sensor outputs different observation data D1, D2, ..., D n , then the goal of multidimensional data fusion is to fuse these data;

[0182]

[0183] Among them, D i is the measurement data of the i-th sensor;

[0184] w i For sensor S i The weight of

[0185] D fused is the fused multidimensional data;

[0186] When performing omnidirectional scanning, a lidar or ultrasonic sensor is used to estimate the distance and position of an object. If the sensor s emits a laser signal and receives a reflected signal, the distance d to the object is calculated using the following formula:

[0187]

[0188] Where d is the distance from the object to the sensor;

[0189] c is the speed of light;

[0190] Δt is the time difference from when the sensor emits the laser to when it receives the reflected signal;

[0191] For visual sensors, the acquired image data is used to calculate the depth information of the object through stereo vision or structured light methods. The depth calculation formula of the stereo vision system is as follows:

[0192]

[0193] Where d is the depth of the object;

[0194] f is the focal length of the camera;

[0195] B is the baseline of the camera;

[0196] x is the disparity corresponding to the object in the image;

[0197] For the three-dimensional spatial positioning of a multimodal sensor cluster, a transformation matrix is used to convert data between different sensors. The sensor coordinate system is set as C1, and the position of the object in this coordinate system is P1. The conversion to the object position P2 in another coordinate system C2 is performed through the following matrix operation:

[0198] P2=T·P1

[0199] Where T is the transformation matrix from C1 to C2;

[0200] P1 is the position of the object in the C1 coordinate system;

[0201] P2 is the position of the object in the C2 coordinate system.

[0202] By combining multiple sensors, it is possible to obtain information on different dimensions of an object, thereby improving the accuracy of object positioning; the use of multi-sensor clusters for data fusion can greatly improve the robustness of the system; the edge computing module can process sensor data in real time, reducing dependence on cloud computing and ensuring the system's response speed in dynamic environments; through precise three-dimensional spatial positioning and coordinate transformation, the system can accurately grasp the spatial position of objects and thus optimize sorting and packaging paths; the adaptive algorithm enables the system to automatically adjust the recognition strategy according to environmental changes or the special circumstances of the objects, further improving the accuracy of object classification and sorting; the multi-sensor and data fusion characteristics of this solution enable it to be effectively applied in a variety of environments, whether it is complex industrial environments, warehousing and logistics, or dynamic sorting lines, it can provide efficient and accurate solutions.

[0203] The data is processed in real time by the edge computing module to generate the location, category, and size information of the item as follows:

[0204] If a laser radar is used to obtain the distance information d and angle information θ of an object, the three-dimensional coordinates (x, y, z) of the object are calculated using the following formula:

[0205] x=d.cos(θ)

[0206] y=d·sin(θ)

[0207] z=f(d,θ)

[0208] Where d is the distance between the object and the sensor;

[0209] θ is the angle of laser scanning;

[0210] For stereo vision systems, the parallax Δx of the binocular cameras is used to calculate the depth of the object:

[0211]

[0212] Where f is the focal length of the camera;

[0213] B is the baseline of the two cameras;

[0214] Δx is the parallax of the object between the two images;

[0215] The edge computing module uses a deep learning model to identify and classify objects. The image data collected by the sensor is processed by the deep learning model to obtain a feature vector f, and the category information of the object is calculated through the softmax layer.

[0216] The calculation formula for category c is:

[0217] c=softmax((W·f+b)

[0218] Among them, f is the feature vector after processing by the convolution layer;

[0219] W is the weight matrix of the classification model;

[0220] b is the bias of the classification model;

[0221] Softmaxsoftmax is an activation function used to output the probability of each category;

[0222] The calculation formula of the softmax function is:

[0223]

[0224] where z i is the i-th element in the eigenvector;

[0225] The length, width, and height of an object are calculated by extracting the depth information and object outline. The outline information of the object is obtained by the visual sensor, and the size is estimated by the depth information. For the stereo vision system, the actual size of the object is calculated by the disparity map and depth information.

[0226] If the depth information of the object is d(x, y), the size of the object can be estimated as follows:

[0227]

[0228] Where w is the width of the object, x1 and x2 are the horizontal boundaries of the object in the image;

[0229] h is the height of the object, y1 and y2 are the vertical boundaries of the object on the image;

[0230] d is the depth information of the object;

[0231] f is the focal length of the camera.

[0232] Real-time data processing through the edge computing module can reduce data transmission delays and improve the real-time performance of item positioning, identification, classification, and size estimation, making it suitable for scenarios such as item sorting and warehousing in dynamic environments. The combination of LiDAR and stereo vision can provide high-precision location information, depth information, and item size under different environmental conditions. Using deep learning models to classify images can identify and classify various items in complex scenarios. This solution integrates LiDAR, visual sensors, and edge computing technologies, and through multi-sensor data fusion, it can provide more accurate and comprehensive item information. This solution is not only suitable for item identification and sorting in industrial and logistics scenarios, but can also be applied to multiple fields such as autonomous driving and robot navigation. Through precise item size calculation and real-time identification, the efficiency and accuracy of automatic sorting systems can be significantly improved. Accurate item location and classification information can optimize the sorting process and reduce errors and misclassifications.

[0233] Based on the deep learning model, objects are identified and classified. If objects with irregular shapes or surface damage are identified, the recognition strategy is adjusted through the adaptive algorithm as follows:

[0234] A convolutional neural network is used to identify objects. After processing through each layer of the network, the feature vector f of the object is obtained, and the category c of the object is calculated using the softmax function:

[0235] c=softmax(W·f+b)

[0236] Where f is the feature vector of the item;

[0237] W is the weight matrix of the classification model;

[0238] b is the bias of the classification model;

[0239] Softmaxsoftmax is an activation function used to calculate the probability of each category;

[0240] Items with irregular shapes or surface damage have shape characteristics that differ from typical items. These are detected and adaptive adjustments are triggered using the following methods:

[0241] The object detection model is used to measure the bounding box of the object and calculate the contour features of the object. The calculation formula of the contour difference is an error metric based on shape matching, which is:

[0242]

[0243] Among them, C i is the contour point of item i;

[0244] C idcal For the ideal object outline;

[0245] n is the number of contour points;

[0246] The surface of the object is segmented by image segmentation method, and the surface damage is detected. If the area of the damaged area A damagc Greater than a certain threshold T damagc , it means the item is damaged;

[0247] The damage area calculation formula is:

[0248] ParseError:KaTeX parse error:Can't use function(′in math mode atposition 36:...\sum_{i=1}^{n}C_{\text{damage...

[0249] Among them, C damage (i) Pixels representing the damaged area;

[0250] It is the marking function of the damaged area. If the pixel belongs to the damaged area, it returns 1; otherwise, it returns 0.

[0251] T damage is the threshold of the damage area;

[0252] When irregular shapes or surface damage are detected, the recognition strategy will be adjusted. The adaptive algorithm dynamically adjusts the classification model or feature extraction strategy. When irregular shapes or damage are detected, the model is dynamically adjusted through incremental learning. The current model parameter is set to θ, and the new parameter after incremental learning is updated to θ′. The update formula is:

[0253] θ′=θ+Δθ

[0254] Where Δθ is the gradient or update amount calculated based on the current new data;

[0255] For damaged or irregular items, dynamically adjust the feature extraction strategy to improve the classification results by weighting the importance of different features. Set each feature f i The contribution to classification is w i , when damage or irregular shapes are detected, the weight w is adaptively adjusted i To improve the sensitivity of the model, the weight update formula is:

[0256] w′ i =w i ·(1+λ·△f i )

[0257] Among them, w′ i is the updated feature weight;

[0258] λ is the learning rate or adjustment coefficient;

[0259] Δf i is the difference between current characteristics and ideal characteristics;

[0260] When irregular shapes or damage are identified, more training data with damage or irregular shapes is generated through data augmentation technology;

[0261] Set the original item image to I original , the enhanced image I is obtained by rotating or adding noise augmented :

[0262] I auguncntcd =f augncntatiom (I original )

[0263] Among them, f augmentation is the data enhancement function.

[0264] This solution can dynamically adjust the recognition strategy according to the shape and damage of the object. When the object has an irregular shape or surface damage, the model will adaptively adjust the feature extraction method and classification strategy to improve the accuracy and robustness of recognition. Incremental learning can update the model parameters in real time based on new data, so that the model can be continuously optimized in actual applications to improve the accuracy of classification. This solution is particularly suitable for environments that need to deal with irregular shapes, surface damage or other abnormal objects. Through weighted features and dynamic adjustment strategies, the system can effectively distinguish between normal objects and damaged objects, thereby avoiding misclassification. Data enhancement methods help generate more diverse training data, increasing the model's adaptability to different situations. Feature weighting adjustment enables the model to prioritize features that contribute more to classification when faced with damaged or irregular objects, thereby improving recognition accuracy. This solution improves the robustness of the model in complex environments by integrating incremental learning, adaptive algorithms and data enhancement, ensuring that various types of objects, especially damaged or irregularly shaped objects, can be reliably identified in actual applications.

[0265] Based on the identification results of the items, the intelligent sorting system selects the correct path and sends the items to the designated sorting area as follows:

[0266] Set the sorting area S corresponding to each item category c , that is, the sorting area to which each category of items should be sent. The sorting area S corresponding to each category c is defined as follows c :

[0267] S c =Mapping(c)

[0268] Where Mapping(c) is a mapping function used to map item category c to a specified sorting area;

[0269] In the intelligent sorting system, the path planning algorithm is used to determine the path from the current sorting position to the target sorting area. The sorting system is set up to include multiple sorting areas and paths. The cost of each path is p ij , that is, the cost from path i to path j;

[0270] According to the item category recognition result c, the intelligent sorting system selects the sorting area according to the optimal path. The mathematical formula of the path selection process is expressed as:

[0271]

[0272] Where P(c) is the optimal path for item category c;

[0273] p ij is the path cost from node i to node j;

[0274] N is the number of all path nodes;

[0275] The sorting system has the ability to adjust dynamically. When some paths are blocked or congested, the system will reselect the path and set the current state of the system to S. t , by evaluating the load of the current path, the path selection is updated through an adjustment function:

[0276]

[0277] Where λ is the adjustment factor;

[0278] load(i) is the load of path i;

[0279] After adopting the path through the path planning algorithm, the system sends the items from the current location to the target sorting area S c , set the current position of the item to P start , the target sorting area is S c , the final path is chosen as P optimal , the delivery process of the items is expressed as:

[0280] P optimal =PathPlanning(P start ,S c ).

[0281] Through the path planning algorithm, the system can select the shortest or optimal path based on the item category and the actual situation of the sorting area, thereby reducing sorting time and improving sorting efficiency. The system has dynamic adjustment capabilities. In the event of path obstruction or congestion, it can evaluate and adjust the path selection in real time to ensure that items are delivered to the target sorting area on time and avoid delays caused by path problems. Through the mapping of item categories and sorting areas, path planning, and intelligent management of path selection, the system can intelligently select the appropriate path based on the characteristics of the item, making the sorting process more efficient and accurate. The solution can flexibly respond to different sorting needs, especially in dynamic environments, and can adjust the sorting strategy according to real-time status to ensure the continuous and efficient operation of the system. The path planning and dynamic adjustment mechanism enhance the robustness and reliability of the sorting system. In the face of emergencies, the system can automatically adjust to avoid system crashes or failure to complete sorting tasks on time. By ensuring that items are accurately delivered from their current location to the target sorting area, the system avoids misclassification or incorrect delivery of items, further improving the accuracy of the system.

[0282] Depending on the size, shape and type of items, the packaging method is automatically adopted and packaged by the automatic packaging system as follows:

[0283] Set the item's size to D = (l, w, h), shape to Shape(D), and type to T;

[0284] Depending on the size, shape and type of items, the automatic packaging system adopts packaging methods, which include box packaging, bag packaging, and customized packaging;

[0285] The packaging method P is decided by the following rules:

[0286] P=ChoosePackaging(D,Shape(D),T)

[0287] The packaging material is based on the weight of the item. itcm , shape Shape (D) and type T are determined, and the selection function of packaging materials is established as:

[0288] M=ChooseMaterial(T,w itcm , Shape(D))

[0289] Where M is the selected packaging material;

[0290] The calculation of packaging size depends on the size and shape of the item and the packaging method used. For rectangular items, the size of the packaging box will be based on the size of the item plus some additional margins. For spherical items, the size of the packaging box will cover the diameter of the sphere.

[0291] If the size of the item is D = (l, w, h), then the size of the packaging box P box =(l+w+h+);

[0292] For irregular items, the package size is determined by calculating the circumscribed rectangle or circle of the item:

[0293] P box =BoundingBox(D)

[0294] The automatic packaging system uses robotic arms and conveyor belt equipment to pack. The packaging process includes the following steps:

[0295] Locates items and places them in the packing area based on their size and shape;

[0296] Automatically cut the packaging material and form it into a suitable shape according to the selected packaging method and material;

[0297] Place the item in the packing material, making sure it is securely wrapped;

[0298] Close the packaging mouth and complete the packaging process;

[0299] The packaging process is expressed by the following control formula:

[0300] Pack(D,P,M)=PositionItem(D)+CutMaterial(M)+SealPackaging(P).

[0301] This solution achieves a high degree of automation in the packaging process through automatic packaging systems and equipment such as robotic arms and conveyor belts, reducing manual operations and improving packaging efficiency. The system automatically determines the appropriate packaging method and material based on the size, shape and type of the item, avoiding manual judgment errors and ensuring that each item is packaged in the most suitable way. Accurate packaging size calculations based on the characteristics of the item avoid packaging waste, and through precise packaging design, not only space can be saved but also packaging costs can be reduced. By selecting the appropriate packaging method based on the shape, type and other characteristics of the item, the item is ensured to be less susceptible to damage during transportation, thereby enhancing the protection of the item. Whether it is a rectangular, spherical or irregular item, the system can respond flexibly and use appropriate packaging methods and materials to ensure the system's adaptability in the packaging of diverse items. Through precise size calculation and automatic cutting functions, the system can ensure the rational use of packaging materials while avoiding errors caused by human factors and improving packaging accuracy.

[0302] After the items are packaged, they are sent to the shipping area for distribution or further processing as follows:

[0303] After the items are packed, they are marked as ready for processing or delivery, which is represented by P packcd , set up shipping area or processing area as Q ship ;

[0304] The path selection for transferring items from the packaging area to the shipping area usually involves path planning, setting up a sorting system or a transportation system using multiple paths P path , each path has a cost of time, distance or resource consumption, expressed as C(P path );

[0305] The calculation formula for path selection is:

[0306]

[0307] Among them, P optimal The path to be adopted;

[0308] C(P path ) is the path cost;

[0309] In the automatic distribution system, the scheduling of items will be based on multiple factors, including inventory, transportation resources and distribution priority. The priority of each item is set as P. priority, there are multiple delivery tasks in the shipping area, and the system will schedule them to minimize the overall delivery time or cost. The scheduling problem is expressed as the following optimization problem:

[0310]

[0311] Among them, C task，i is the cost of the i-th task;

[0312] P priority，i is the priority of task i;

[0313] During the distribution or further processing, the inventory is updated and the inventory quantity of each item is set as I itcm , when items are moved from the packing area to the shipping area, the inventory quantity is reduced by the quantity of items processed;

[0314] The inventory update formula is:

[0315] I itcm =I tem -QuantityShipped

[0316] Among them, I itcm is the current inventory quantity of the item;

[0317] QuantityShipped is the number of items shipped from inventory;

[0318] Once the item arrives at the shipping area, depending on the type of item, it will be delivered directly to the customer or enter further processing. The following formula is used to represent the further processing flow:

[0319] NextStage(P packed , Q ship )=ChooseProcess(P packed ,Q ship )

[0320] Among them, NextStage(P packed , Q ship ) means after the items are delivered from the packaging area to the shipping area;

[0321] ChooseProcess(P packed , Q ship ) is to select the appropriate processing or delivery method based on the category or demand of the items.

[0322] The system can optimize path selection based on path cost, reduce unnecessary delays or resource waste during transportation, and improve the efficiency of item distribution; the scheduling of items is optimized by considering factors such as priority, inventory, and transportation resources, so that high-priority items can be processed first, ensuring the timeliness of distribution tasks and the rational use of resources; the inventory update formula ensures that the inventory level of items is updated in real time during the transfer process from the packaging area to the shipping area, avoiding inventory errors and inaccurate inventory management, and reducing the risk of inventory backlogs and out-of-stocks; according to the type of item, the system can select the most appropriate subsequent processing method, such as direct delivery to customers or further processing, which improves the adaptability and flexibility of the system; by optimizing task scheduling, path selection and inventory management, the system can effectively reduce distribution costs and time, and improve overall operational efficiency.

[0323] This embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the program, the intelligent feeding, packaging, and sorting system and method based on high-precision visual recognition as described above are implemented.

[0324] This embodiment also provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the intelligent feeding, packaging and sorting system and method based on high-precision visual recognition as described above are implemented.

[0325] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0326] Those skilled in the art will clearly understand that for the sake of convenience and brevity in description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0327] The above embodiments of the present invention are not intended to limit the scope of protection of the present invention, and the implementation methods of the present invention are not limited thereto. All other modifications, replacements or changes made to the above structures of the present invention based on the above contents of the present invention, in accordance with common technical knowledge and customary means in this field, without departing from the above basic technical ideas of the present invention, should fall within the scope of protection of the present invention.

Claims

1. Intelligent feeding, packaging and sorting system based on high-precision visual recognition, characterized by: Includes: The visual sensor module is used for high-resolution cameras to capture images of objects and perform preliminary recognition through image processing algorithms; Multimodal sensor modules, which use lidar, infrared sensors, and ultrasonic sensors to complement visual sensors and ensure recognition in complex environments; The computing unit module includes an edge computing module for real-time processing and analysis of sensor data. The item handling module includes an automatic feeding device, a robotic arm, and a conveyor belt to transport items from the input port to the designated location and perform sorting operations according to sorting instructions; The packaging module is used to adopt relevant packaging methods based on the item identification results and output them to the shipping port.

2. Intelligent feeding, packaging and sorting method based on high-precision visual recognition, characterized in that: The following steps are involved: Step 1: The sensor cluster performs an all-round scan of the items to be sorted and collects multi-dimensional data; Step 2: The data is processed in real time by the edge computing module to generate the item’s location, category, and size information; Step 3: Identify and classify objects based on the deep learning model. If an object with an irregular shape or surface damage is identified, the recognition strategy is adjusted through an adaptive algorithm. Step 4: Based on the identification results of the items, the intelligent sorting system selects the correct path and sends the items to the designated sorting area; Step 5: Automatically adopt packaging methods based on the size, shape and type of the items and pack them through the automatic packaging system; Step 6: After the items are packaged, they are sent to the shipping area for distribution or further processing.

3. The intelligent feeding, packaging and sorting method based on high-precision visual recognition according to claim 2 is characterized in that: The sensor cluster performs an all-around scan of the items to be sorted and collects multi-dimensional data in the following ways: Set up multiple sensors S1, S2, ..., S n , each sensor outputs different observation data D1, D2, ..., D n , then the goal of multidimensional data fusion is to fuse these data; Among them, D i For the The measurement data of each sensor; w i For sensor S i The weight of D fuscd is the fused multidimensional data; When performing omnidirectional scanning, a lidar or ultrasonic sensor is used to estimate the distance and position of an object. If the sensor S emits a laser signal and receives a reflected signal, the distance d to the object is calculated using the following formula: Where d is the distance from the object to the sensor; c is the speed of light; Δt is the time difference from when the sensor emits the laser to when it receives the reflected signal; For visual sensors, the acquired image data is used to calculate the depth information of the object through stereo vision or structured light methods. The depth calculation formula of the stereo vision system is as follows: Where d is the depth of the object; f is the focal length of the camera; B is the baseline of the camera; x is the disparity corresponding to the object in the image; For the three-dimensional spatial positioning of a multimodal sensor cluster, a transformation matrix is used to convert data between different sensors. The sensor coordinate system is set as C1, and the position of the object in this coordinate system is P1. The conversion to the object position P2 in another coordinate system C2 is performed through the following matrix operation: P2=T·P1 Where T is the transformation matrix from C1 to C2; P1 is the position of the object in the C1 coordinate system; P2 is the position of the object in the C2 coordinate system.

4. The intelligent feeding, packaging and sorting method based on high-precision visual recognition according to claim 3 is characterized in that: The data is processed in real time by the edge computing module to generate the location, category, and size information of the item as follows: If a laser radar is used to obtain the distance information d and angle information θ of an object, the three-dimensional coordinates (x, y, z) of the object are calculated using the following formula: x=d·cos(θ) y=d·sin(θ) z=f(d,θ) Where d is the distance between the object and the sensor; θ is the angle of laser scanning; For stereo vision systems, the parallax Δx of the binocular cameras is used to calculate the depth of the object: Where f is the focal length of the camera; B is the baseline of the two cameras; Δx is the parallax of the object between the two images; The edge computing module uses a deep learning model to identify and classify objects. The image data collected by the sensor is processed by the deep learning model to obtain a feature vector f, and the category information of the object is calculated through the softmax layer. The calculation formula for category c is: c=softmax(W·f+b) Among them, f is the feature vector after processing by the convolution layer; w is the weight matrix of the classification model; b is the bias of the classification model; Softmaxsoftmax is an activation function used to output the probability of each category; The calculation formula of the softmax function is: where z i is the i-th element in the eigenvector; The length, width, and height of an object are calculated by extracting the depth information and object outline. The outline information of the object is obtained by the visual sensor, and the size is estimated by the depth information. For the stereo vision system, the actual size of the object is calculated by the disparity map and depth information. If the depth information of the object is d(x, y), the size of the object can be estimated as follows: Where w is the width of the object, x1 and x2 are the horizontal boundaries of the object in the image; h is the height of the object, y1 and y2 are the vertical boundaries of the object on the image; d is the depth information of the object; f is the focal length of the camera.

5. The intelligent feeding, packaging and sorting method based on high-precision visual recognition according to claim 4 is characterized in that: Based on the deep learning model, objects are identified and classified. If objects with irregular shapes or surface damage are identified, the recognition strategy is adjusted through the adaptive algorithm as follows: A convolutional neural network is used to identify objects. After processing through each layer of the network, the feature vector f of the object is obtained, and the category c of the object is calculated using the softmax function: c=softmax(W·f+b) Where f is the feature vector of the item; W is the weight matrix of the classification model; b is the bias of the classification model; Softmaxsoftmax is an activation function used to calculate the probability of each category; Items with irregular shapes or surface damage have shape characteristics that differ from typical items. These are detected and adaptive adjustments are triggered using the following methods: The object detection model is used to measure the bounding box of the object and calculate the contour features of the object. The calculation formula of the contour difference is an error metric based on shape matching, which is: Among them, C i is the contour point of item i; C idceal For the ideal object outline; n is the number of contour points; The surface of the object is segmented by image segmentation method, and the surface damage is detected. If the area of the damaged area A damage Greater than a threshold T damagc , it means the item is damaged; The damage area calculation formula is: ParseError:KaTeX parseerror:Can'tuse function′\(′in math mode at position36:...\sum_{i=1}^{n}\(C_{\text{damage... Among them, C damagc (i) Pixels representing the damaged area; It is the marking function of the damaged area. If the pixel belongs to the damaged area, it returns 1; otherwise, it returns 0. T damagc is the threshold of the damage area; When irregular shapes or surface damage are detected, the recognition strategy will be adjusted. The adaptive algorithm dynamically adjusts the classification model or feature extraction strategy. When irregular shapes or damage are detected, the model is dynamically adjusted through incremental learning. The current model parameter is set to θ, and the new parameter after incremental learning is updated to θ′. The update formula is: θ′=θ+Δθ Where Δθ is the gradient or update amount calculated based on the current new data; For damaged or irregular items, dynamically adjust the feature extraction strategy to improve the classification results by weighting the importance of different features. Set each feature f i The contribution to classification is w i , when damage or irregular shapes are detected, the weight w is adaptively adjusted i To improve the sensitivity of the model, the weight update formula is: ω′ i =w i ·(1+λ·Δf i ) Among them, w′ i is the updated feature weight; λ is the learning rate or adjustment coefficient; Δf i is the difference between current characteristics and ideal characteristics; When irregular shapes or damage are identified, more training data with damage or irregular shapes is generated through data augmentation technology; Set the original item image to I original , the enhanced image I is obtained by rotating or adding noise augmented : I augmented =f augmentation (I original ) Among them, f augmentation is the data enhancement function.

6. The intelligent feeding, packaging and sorting method based on high-precision visual recognition according to claim 5 is characterized in that: Based on the identification results of the items, the intelligent sorting system selects the correct path and sends the items to the designated sorting area as follows: Set the sorting area S corresponding to each item category c , that is, the sorting area to which each category of items should be sent. The sorting area S corresponding to each category c is defined as follows c : S c =Mapping(c) Where Mapping(c) is a mapping function used to map item category c to a specified sorting area; In the intelligent sorting system, the path planning algorithm is used to determine the path from the current sorting position to the target sorting area. The sorting system is set up to include multiple sorting areas and paths. The cost of each path is p ij , that is, the cost from path i to path j; According to the item category recognition result c, the intelligent sorting system selects the sorting area according to the optimal path. The mathematical formula of the path selection process is expressed as: Where P(c) is the optimal path for item category c; p ij is the path cost from node i to node j; N is the number of all path nodes; The sorting system has the ability to adjust dynamically. When some paths are blocked or congested, the system will reselect the path and set the current state of the system to S. t , by evaluating the load of the current path, the path selection is updated through an adjustment function: Where λ is the adjustment factor; load(i) is the load of path i; After adopting the path through the path planning algorithm, the system sends the items from the current location to the target sorting area S c , set the current position of the item to P start , the target sorting area is S c , the final path is chosen as P optimal , the delivery process of the items is expressed as: P optmeal =PathPlanning(P start ,S c )。 7. The intelligent feeding, packaging and sorting method based on high-precision visual recognition according to claim 6 is characterized in that: Depending on the size, shape and type of items, the packaging method is automatically adopted and packaged by the automatic packaging system as follows: Set the item's size to d = (l, w, h), shape to Shape(D), and type to T; Depending on the size, shape and type of items, the automatic packaging system adopts packaging methods, which include box packaging, bag packaging, and customized packaging; The packaging method P is decided by the following rules: P=ChoosePackaging(D,Shape(D),T) The packaging material is based on the weight of the item. itcm , shape Shape (D) and type T are determined, and the selection function of packaging materials is established as: M=ChooseMaterial(T,w itcm ,Shape(D)) Where M is the selected packaging material; The calculation of packaging size depends on the size and shape of the item and the packaging method used. For rectangular items, the size of the packaging box will be based on the size of the item plus some additional margins. For spherical items, the size of the packaging box will cover the diameter of the sphere. If the size of the item is D = (l, w, h), then the size of the packaging box is P box =(l+w+h+); For irregular items, the package size is determined by calculating the item's circumscribed rectangle or circle: P box =BoundingBox(D) The automatic packaging system uses robotic arms and conveyor belt equipment to pack. The packaging process includes the following steps: Locates and places items in the packing area based on their size and shape; Automatically cut the packaging material and form it into a suitable shape according to the selected packaging method and material; Place the item in the packing material, making sure it is securely wrapped; Close the packaging mouth and complete the packaging process; The packaging process is expressed by the following control formula: Pack(D,P,M)=PositionItem(D)+CutMaterial(M)+SealPackaging(P).

8. The intelligent feeding, packaging and sorting method based on high-precision visual recognition according to claim 7 is characterized in that: After the items are packaged, they are sent to the shipping area for distribution or further processing as follows: After the items are packed, they are marked as ready for processing or delivery, which is represented by P packed , set up shipping area or processing area as Q ship ; The path selection for transferring items from the packaging area to the shipping area usually involves path planning, setting up a sorting system or a transportation system using multiple paths P path , each path has a cost of time, distance or resource consumption, expressed as C(P path ); The calculation formula for path selection is: P optimal =argminP ath C(P path ) Among them, P optimal The path adopted; C(P path ) is the path cost; In the automatic distribution system, the scheduling of items will be based on multiple factors, including inventory, transportation resources and distribution priority. The priority of each item is set as P. priority , there are multiple delivery tasks in the shipping area, and the system will schedule them to minimize the overall delivery time or cost. The scheduling problem is expressed as the following optimization problem: Among them, C task，i For the The cost of each task; P priority，i For the task priority; During the distribution or further processing, the inventory is updated and the inventory quantity of each item is set as I item , when items are moved from the packing area to the shipping area, the inventory quantity is reduced by the quantity of items processed; The inventory update formula is: I item =I item -QuantityShipped Among them, I item is the current inventory quantity of the item; QuantityShipped is the number of items shipped from inventory; Once the item arrives at the shipping area, depending on the type of item, it will be delivered directly to the customer or enter further processing. The following formula is used to represent the further processing flow: NextStage(P packed ,Q ship )=ChooseProcess(P packed ,Q ship ) Among them, NextStage(P packed , Q ship ) means after the items are delivered from the packaging area to the shipping area; ChooseProcess(P packed , Q ship ) is to select the appropriate processing or delivery method based on the category or demand of the items.

9. An electronic device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the intelligent feeding packaging sorting method based on high-precision visual recognition as described in any one of claims 2 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the intelligent feeding, packaging and sorting method based on high-precision visual recognition as described in any one of claims 2 to 8 is implemented.

Images