Conveying system and method based on ai item recognition

Abstract

The application discloses a kind of based on AI commodity identification's conveying system and method, it is related to conveying identification technical field, comprising: acquisition module, for real-time acquisition conveying equipment in the image data and sensor data of the goods to be conveyed;AI feature combination module is used to the image data and sensor data of the goods to be conveyed based on AI model feature extraction, and the extracted features are combined, obtain first feature;Commodity identification module is used to the first feature is input into pre-trained commodity identification model and is identified, determines identification result;Adjustment module is used to the conveying information of the goods to be conveyed based on identification result is adjusted;Optimize the collaborative mechanism of conveying mechanism equipment and identification system, improve the conveying efficiency of conveying mechanism equipment.

Description

Technical Field

[0001] This invention relates to the field of transmission and recognition technology, and in particular to a transmission system and method based on AI-based product recognition. Background Technology

[0002] With the rapid development of the logistics industry, automation and intelligence have become key to improving efficiency. Traditional conveying equipment mostly relies on mechanical structures or simple sensors to sort goods, which has problems such as low recognition accuracy and poor adaptability. In recent years, with the maturity of AI technology, intelligent recognition systems that combine computer vision, deep learning and other technologies have been gradually applied to the logistics field. However, how to achieve seamless coordination between goods recognition and goods conveying is still the focus and difficulty of current technological development.

[0003] Among the existing publicly available related technologies, some technical solutions focus more on the recognition algorithm itself and do not fully consider the collaborative optimization between the transmission device and the recognition system, resulting in problems such as low recognition efficiency, poor transmission efficiency, and poor system stability in practical applications.

[0004] Therefore, there is an urgent need for a delivery system and method based on AI product recognition to solve the above problems. Summary of the Invention

[0005] The present invention aims to at least partially solve one of the technical problems in the aforementioned technologies. Therefore, a first aspect of the present invention aims to provide a conveying system based on AI product recognition, which optimizes the collaborative mechanism between the conveying device and the recognition system, thereby improving the conveying efficiency of the conveying device.

[0006] A second objective of this invention is to provide a delivery method based on AI-based product recognition.

[0007] To achieve the above objectives, a first aspect of the present invention provides a delivery system based on AI product recognition, comprising:

[0008] The acquisition module is used to acquire image data and sensor data of the goods to be transported in the conveying equipment in real time;

[0009] The AI ​​feature combination module is used to extract features from the image data and sensor data of the goods to be delivered based on the AI ​​model, and combine the extracted features to obtain the first feature;

[0010] The product recognition module is used to input the first feature into a pre-trained product recognition model for recognition and to determine the recognition result;

[0011] The adjustment module is used to adjust the delivery information of the goods to be delivered based on the recognition results.

[0012] Preferably, the acquisition module includes:

[0013] The first acquisition submodule is used to acquire images of the goods to be transported in the conveying equipment based on the industrial camera, and obtain image data of the goods to be transported.

[0014] The second acquisition submodule is used to acquire the weight data of the goods to be transported in the conveying device based on the weight sensor, and to acquire the size data of the goods to be transported based on the laser volume measuring instrument, and to acquire the size data of the goods to be transported; and to use the weight data and size data of the goods to be transported as sensor data of the goods to be transported.

[0015] Preferably, the AI ​​feature combination module includes:

[0016] The AI ​​feature extraction submodule is used to extract features from the image data and sensor data of the goods to be delivered based on the AI ​​model, so as to obtain the target features.

[0017] The normalization submodule is used to normalize the target features to obtain the normalized target features.

[0018] The first combination submodule is used to combine the normalized target features to generate the first feature.

[0019] Preferred methods for training product recognition models include:

[0020] Acquire image data, weight data, and dimension data of goods historically transported by the conveying equipment;

[0021] The image data, weight data, and size data of historically delivered goods are preprocessed and labeled to obtain a labeled dataset of historically delivered goods.

[0022] The labeled historical transmitted product dataset is divided into a product recognition training dataset and a product recognition test dataset based on a preset ratio.

[0023] The training image data in the commodity recognition training dataset is input into the convolutional layer of the neural network model to extract features from the training image data and obtain the local features corresponding to the training image data. The generated local features are then input into the pooling layer, which performs dimensionality reduction on the local features output by the convolutional layer to obtain the visual feature vector.

[0024] The preprocessed weight and size data of the products are concatenated to generate a structured feature vector;

[0025] Visual feature vectors and structured feature vectors are concatenated to form a fused feature vector; the fused feature vector is input into a fully connected layer, and deep correlations between features are learned through nonlinear transformation; finally, the output layer outputs N probability values ​​for product recognition.

[0026] The training samples in the product recognition training dataset are sequentially input into the neural network model for training. When the training results meet the requirements, the initial product recognition model is obtained.

[0027] The initial product recognition model was tested using a product recognition test dataset. When the test results met the requirements, a well-trained product recognition model was obtained.

[0028] Preferably, a submodule is constructed to obtain historical transmission data from the transmission device and construct a path database based on the historical transmission data;

[0029] The second combination submodule is used to extract features from other related dimensions of the classified goods to be transmitted based on the goods classification information in the recognition results, to obtain supplementary features of related dimensions, and to combine the supplementary features of related dimensions with the first feature to obtain the second feature;

[0030] The determination submodule is used to calculate the feature-weighted cosine similarity between the second feature and the path database, compare the feature-weighted cosine similarity with a preset similarity threshold, and take the path with the feature-weighted cosine similarity greater than or equal to the preset similarity threshold as the initial target path;

[0031] The filtering submodule is used to filter the initial target path and determine the final target path;

[0032] The adjustment submodule is used to dynamically adjust the transmission speed of goods to be delivered along the final target path.

[0033] Preferably, the sub-modules include:

[0034] The first acquisition unit is used to acquire historical transmission data of the transmission device;

[0035] The first extraction unit is used to extract the transmission nodes in the historical transmission path included in the historical transmission data to obtain the target node;

[0036] The second extraction unit is used to extract the spatiotemporal features of the target node in the historical transmission path and determine the spatiotemporal feature information corresponding to the target node.

[0037] The first construction unit is used to construct a four-dimensional grouped feature based on the target node and spatiotemporal feature information, consisting of a starting point, an end point, spatiotemporal features, and a commodity type; and to construct a path database based on the four-dimensional grouped feature.

[0038] Preferably, the second combined submodule includes:

[0039] The classification unit is used to classify the goods to be delivered based on the commodity classification information included in the identification results;

[0040] The third extraction unit is used to extract other related dimensions of the classified goods to be delivered, including goods type features, environmental related features, operation adaptation features, and time constraint features.

[0041] The second construction unit is used to construct supplementary features for the association dimension based on the extracted product type features, environmental association features, operation adaptation features, and time constraint features.

[0042] The combination unit is used to combine the supplementary features of the related dimension with the first feature to obtain the second feature.

[0043] Preferred filtering submodules include:

[0044] The graph construction unit is used to extract the dependencies between nodes in the historical transmission path included in the historical transmission data, and to construct a node state association graph.

[0045] The first filtering unit is used to acquire the current task data and pre-allocated task volume of each node in the initial target path in real time, and calculate the conflict probability of each node in each initial target path. If the maximum conflict probability is greater than the preset conflict probability threshold, the downstream associated nodes of the node with the maximum conflict probability in the initial target path are queried based on the node state association graph to determine several downstream associated nodes. The propagation impact value of the node with the maximum conflict probability in the initial target path and several downstream associated nodes are calculated respectively. The sum of the maximum conflict probability and the maximum propagation impact value is calculated, and the initial target path with a sum greater than or equal to the preset conflict sum value threshold is deleted.

[0046] The second filtering unit is used to obtain the average conflict probability, spatiotemporal load entropy and total path distance of the remaining initial target paths; based on the average conflict probability, spatiotemporal load entropy and total path distance of the remaining initial target paths, each remaining initial target path is evaluated, and the remaining initial target path corresponding to the maximum evaluation value is taken as the final target path.

[0047] Preferably, the adjustment submodule includes:

[0048] The determination unit is used to segment the final target path into different segment areas and query the commodity type-transmission speed table to determine the initial transmission speed of different segment areas.

[0049] The second acquisition unit is used to acquire the load rate of each segment of the final target path in real time.

[0050] The dynamic adjustment unit is used to dynamically adjust the initial conveying speed of goods in each segment based on the load rate of each segment.

[0051] To achieve the above objectives, a second aspect of the present invention proposes a transmission method based on AI product recognition, comprising:

[0052] Real-time acquisition of image data and sensor data of goods to be transported in the conveying equipment;

[0053] Based on the AI ​​model, feature extraction is performed on the image data and sensor data of the goods to be delivered, and the extracted features are combined to obtain the first feature;

[0054] The first feature is input into a pre-trained product recognition model for recognition, and the recognition result is determined.

[0055] The delivery information of the goods to be delivered is adjusted based on the recognition results.

[0056] This invention provides a delivery system and method based on AI-based product recognition. It combines image data and sensor data to improve the accuracy of image recognition. Based on the accurate image recognition results, it adjusts the delivery information of the product, thereby improving the synergy between image recognition and product delivery, effectively increasing the speed and accuracy of product delivery, and improving the efficiency of product delivery.

[0057] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0058] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0059] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0060] Figure 1 This is a block diagram of a delivery system based on AI product recognition according to an embodiment of the present invention;

[0061] Figure 2 This is a block diagram of a data acquisition module according to an embodiment of the present invention;

[0062] Figure 3 This is a flowchart of an AI-based product recognition transmission method according to an embodiment of the present invention. Detailed Implementation

[0063] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0064] Example 1: As Figure 1 As shown, a delivery system based on AI product recognition includes:

[0065] The acquisition module is used to acquire image data and sensor data of the goods to be transported in the conveying equipment in real time;

[0066] The AI ​​feature combination module is used to extract features from the image data and sensor data of the goods to be delivered based on the AI ​​model, and combine the extracted features to obtain the first feature;

[0067] The product recognition module is used to input the first feature into a pre-trained product recognition model for recognition and to determine the recognition result;

[0068] The adjustment module is used to adjust the delivery information of the goods to be delivered based on the recognition results.

[0069] In this embodiment, the image data includes an image of the product and the product code, destination, priority, and associated order demand information corresponding to the QR code or barcode identified in the product image.

[0070] In this embodiment, the extracted features include the weight and size (length, width and height) of the product, as well as the texture and shape of the product image.

[0071] In this embodiment, the AI ​​model includes a feature extraction model trained based on convolutional neural networks, graph neural networks, etc.

[0072] In this embodiment, the first feature includes:

[0073]

[0074] in, The first characteristic representing the goods to be delivered; Indicates the weight of the goods to be delivered; Indicates the length of the goods to be delivered; Indicates the width of the goods to be delivered; Indicates the height of the goods to be delivered; This represents the texture feature value of the item to be delivered; Indicates the regularity of the shape of the goods to be delivered; This indicates the order priority of the goods to be delivered.

[0075] In this embodiment, the transmitted information includes the transmission path and transmission speed.

[0076] The working principle and beneficial effects of the above technical solution are as follows: by combining image data and sensor data, the accuracy of image recognition is improved; based on the accurate image recognition results, the transmission information of goods in the transmission device is adjusted, which improves the coordination between image recognition and goods transmission, effectively improves the speed and accuracy of goods transmission, and improves the efficiency of goods transmission.

[0077] Example 2: As Figure 2 As shown, the acquisition module includes:

[0078] The first acquisition submodule is used to acquire images of the goods to be transported in the conveying equipment based on the industrial camera, and obtain image data of the goods to be transported.

[0079] The second acquisition submodule is used to acquire the weight data of the goods to be transported in the conveying device based on the weight sensor, and to acquire the size data of the goods to be transported based on the laser volume measuring instrument, and to acquire the size data of the goods to be transported; and to use the weight data and size data of the goods to be transported as sensor data of the goods to be transported.

[0080] The working principle and beneficial effects of the above technical solution are as follows: image data acquired by the first acquisition submodule can intuitively reflect the appearance characteristics of the product, such as shape, color, and markings; while the weight and size data acquired by the second acquisition submodule describe the product from the perspective of physical attributes; in this way, multi-dimensional information about the product can be obtained, and a more comprehensive understanding of the product to be transmitted can be achieved, which helps to make more accurate decisions in subsequent operations such as identification, classification, and transmission adjustment.

[0081] Example 3: AI feature combination module, including:

[0082] The AI ​​feature extraction submodule is used to extract features from the image data and sensor data of the goods to be delivered based on the AI ​​model, so as to obtain the target features.

[0083] The normalization submodule is used to normalize the target features to obtain the normalized target features.

[0084] The first combination submodule is used to combine the normalized target features to generate the first feature.

[0085] In this embodiment, feature extraction of the image data and sensor data of the goods to be transmitted can be performed based on a pre-trained feature extraction model.

[0086] In this embodiment, the normalized target features are combined to generate the first feature, namely:

[0087] in, The first characteristic representing the goods to be delivered; Indicates the weight of the goods to be delivered; Indicates the length of the goods to be delivered; Indicates the width of the goods to be delivered; Indicates the height of the goods to be delivered; Represents the texture feature value of the product to be delivered; Indicates the regularity of the shape of the goods to be delivered; Indicates the order priority of the goods to be delivered; This indicates the type and feature values ​​of the goods to be transmitted, which can be implemented by pre-setting a type-feature value table.

[0088] The working principle and beneficial effects of the above technical solution are as follows: Based on the AI ​​model, feature extraction is performed on the image data and sensor data of the goods to be transmitted, which can filter out the most useful information for goods recognition and remove redundant data. This helps to reduce the amount of data to be processed in the subsequent process, improve the efficiency of data processing, and make the target features more representative and more accurately reflect the characteristics of the goods. The first feature obtained by first extracting, then normalizing, and finally combining is more in line with the input requirements of the goods recognition model. It can improve the recognition performance of the goods recognition model.

[0089] Example 4: Training method for a product recognition model, including:

[0090] Acquire image data, weight data, and dimension data of goods historically transported by the conveying equipment;

[0091] The image data, weight data, and size data of historically delivered goods are preprocessed and labeled to obtain a labeled dataset of historically delivered goods.

[0092] The labeled historical transmitted product dataset is divided into a product recognition training dataset and a product recognition test dataset based on a preset ratio.

[0093] The training image data in the commodity recognition training dataset is input into the convolutional layer of the neural network model to extract features from the training image data and obtain the local features corresponding to the training image data. The generated local features are then input into the pooling layer, which performs dimensionality reduction on the local features output by the convolutional layer to obtain the visual feature vector.

[0094] The preprocessed weight and size data of the products are concatenated to generate a structured feature vector;

[0095] Visual feature vectors and structured feature vectors are concatenated to form a fused feature vector; the fused feature vector is input into a fully connected layer, and deep correlations between features are learned through nonlinear transformation; finally, the output layer outputs N probability values ​​for product recognition.

[0096] The training samples in the product recognition training dataset are sequentially input into the neural network model for training. When the training results meet the requirements, the initial product recognition model is obtained.

[0097] The initial product recognition model was tested using a product recognition test dataset. When the test results met the requirements, a well-trained product recognition model was obtained.

[0098] In this embodiment, the visual feature vector and the structured feature vector are concatenated to form a fused feature vector. For example, "red packaging + weight 0.5kg + size 5×5×20cm" may correspond to "canned beverage".

[0099] In this embodiment, deep associations between features are learned through nonlinear transformations. For example, the mapping relationship between "circular outline + small size + low weight" and the "ping-pong ball" category is illustrated.

[0100] The working principle and beneficial effects of the above technical solution are as follows: it is trained based on a large amount of historical transported commodity data; since the historical data covers a variety of commodities, the model has good versatility and can adapt to commodity recognition tasks of different types and appearances, rather than being limited to a few specific commodities, which is conducive to its application in different transport scenarios and environments with a wide variety of commodities.

[0101] Example 5: Adjustment module, including:

[0102] A submodule is built to obtain historical transmission data from the transmission device and to build a path database based on the historical transmission data.

[0103] The second combination submodule is used to extract features from other related dimensions of the classified goods to be transmitted based on the goods classification information in the recognition results, to obtain supplementary features of related dimensions, and to combine the supplementary features of related dimensions with the first feature to obtain the second feature;

[0104] The determination submodule is used to calculate the feature-weighted cosine similarity between the second feature and the path database, compare the feature-weighted cosine similarity with a preset similarity threshold, and take the path with the feature-weighted cosine similarity greater than or equal to the preset similarity threshold as the initial target path;

[0105] The filtering submodule is used to filter the initial target path and determine the final target path;

[0106] The adjustment submodule is used to dynamically adjust the transmission speed of goods to be delivered along the final target path.

[0107] In this embodiment, the supplementary features of the association dimension include, but are not limited to, environmental association features, operation adaptation features, and time constraint features;

[0108]

[0109] in, Additional features indicating the goods to be delivered; This indicates the environmental characteristics of the goods to be delivered; This indicates the operational adaptation features of the goods to be delivered; This indicates the time-sensitive nature of the goods to be delivered; This indicates the type characteristic value of the goods to be delivered.

[0110] The working principle and beneficial effects of the above technical solution are as follows: By acquiring historical transmission data from the transmission equipment to construct a path database, the system can fully utilize past transmission experience; the historical transmission data contains relevant information about various commodities during the transmission process, and the path database can provide rich reference for subsequent path planning, improving the rationality and accuracy of path planning; the supplementary features of the association dimension are combined with the first feature to obtain the second feature, and the feature combination method comprehensively considers various information about the commodity; by extracting features from the association dimension of the classified commodities based on the commodity classification information in the recognition results, supplementary features of the association dimension are obtained, making the features more comprehensively reflect the characteristics of the commodity; when calculating the weighted cosine similarity between the features and the path database, a suitable path is matched more accurately, improving the accuracy of the initial target path determination; the transmission speed of the commodities to be transmitted in the final target path is dynamically adjusted; adjusting the transmission speed according to different path conditions and commodity characteristics can avoid congestion during the transmission process and improve the overall operating efficiency of the transmission equipment.

[0111] Example 6: Constructing a submodule, including:

[0112] The first acquisition unit is used to acquire historical transmission data of the transmission device;

[0113] The first extraction unit is used to extract the transmission nodes in the historical transmission path included in the historical transmission data to obtain the target node;

[0114] The second extraction unit is used to extract the spatiotemporal features of the target node in the historical transmission path and determine the spatiotemporal feature information corresponding to the target node.

[0115] The first construction unit is used to construct a four-dimensional grouped feature based on the target node and spatiotemporal feature information, consisting of a starting point, an end point, spatiotemporal features, and a commodity type; and to construct a path database based on the four-dimensional grouped feature.

[0116] In this embodiment, the transmission node includes the starting point, the ending point, and each turning point in the transmission path.

[0117] In this embodiment, the spatiotemporal characteristics of the target node are specifically exemplified by the following: Node A has an average load rate of 40% between 10:00 and 12:00.

[0118] The working principle and beneficial effects of the above technical solution are as follows: Historical transmission data contains a large amount of information about commodity transmission. By mining this data, a basis can be provided for subsequent route planning and optimization, making full use of past experience data, avoiding resource waste, and improving the intelligence level of the entire transmission system; extracting transmission nodes in historical transmission paths to obtain target nodes allows focusing on key nodes in the transmission process; the second extraction unit further extracts the spatiotemporal characteristics of the target nodes to determine the corresponding spatiotemporal feature information. This targeted extraction method can simplify complex historical transmission data into nodes and feature information closely related to route planning, which helps to improve the accuracy and effectiveness of subsequent route database construction; the first construction unit constructs a four-dimensional grouped feature based on the target node and spatiotemporal feature information, namely, starting point-end point-spatiotemporal feature-commodity type, and constructs a route database based on this; this four-dimensional grouped feature covers key information such as the starting point, end point, spatiotemporal characteristics, and commodity type of transmission, making the constructed route database very comprehensive; it can take into account the transmission needs of different commodities in different times and spaces, providing rich reference materials for accurately determining the transmission path.

[0119] Example 7: Second combined submodule, including:

[0120] The classification unit is used to classify the goods to be delivered based on the commodity classification information included in the identification results;

[0121] The third extraction unit is used to extract other related dimensions of the classified goods to be delivered, including goods type features, environmental related features, operation adaptation features, and time constraint features.

[0122] The second construction unit is used to construct supplementary features for the association dimension based on the extracted product type features, environmental association features, operation adaptation features, and time constraint features.

[0123] The combination unit is used to combine the supplementary features of the related dimension with the first feature to obtain the second feature.

[0124] In this embodiment, environmental characteristics, such as the humidity sensitivity coefficient of the goods to be transported (0.8 for fresh produce and 0.2 for dry goods), reflect the influence of the environment on path selection.

[0125] In this embodiment, the operation adaptation features, such as the difficulty coefficient of grasping the goods to be conveyed, are set to 1.2 for smooth surface goods and 0.8 for rough surface goods, which are associated with the operation stability of the robotic arm.

[0126] In this embodiment, time-sensitive features include, for example, a countdown timer for the latest delivery of goods to be delivered, with expedited orders set at 30 minutes and regular orders set at 120 minutes.

[0127] In this embodiment, the supplementary features are normalized before being combined with the first feature.

[0128] In this embodiment, the second feature includes, but is not limited to:

[0129]

[0130] in, This indicates the second characteristic of the goods to be delivered; Indicates the weight of the goods to be delivered; Indicates the length of the goods to be delivered; Indicates the width of the goods to be delivered; Indicates the height of the goods to be delivered; Represents the texture feature value of the product to be delivered; Indicates the regularity of the shape of the goods to be delivered; Indicates the order priority of the goods to be delivered; Indicates the type characteristic value of the goods to be delivered; This indicates the environmental characteristics of the goods to be delivered; This indicates the operational adaptation features of the goods to be delivered; This indicates the time-sensitive characteristics of the goods to be delivered.

[0131] The working principle and beneficial effects of the above technical solution are as follows: It categorizes the goods to be transported based on the product type identified in the recognition results; it can classify the goods to be transported into different categories according to their own characteristics, which helps to provide personalized processing for different types of goods. For example, it can classify fragile items, fresh produce, and general daily necessities to plan more suitable transport paths and strategies for different types of goods; it extracts the product type features, environmental association features, operation adaptation features, and time constraint features of the categorized goods to be transported; the multi-dimensional feature extraction covers important aspects such as the type attributes of the goods themselves, their association with the surrounding environment, suitable operation methods, and time constraints; comprehensive feature extraction enables a deeper understanding of the characteristics of the goods.

[0132] Example 8: Filtering submodule, including:

[0133] The graph construction unit is used to extract the dependencies between nodes in the historical transmission path included in the historical transmission data, and to construct a node state association graph.

[0134] The first filtering unit is used to acquire the current task data and pre-allocated task volume of each node in the initial target path in real time, and calculate the conflict probability of each node in each initial target path. If the maximum conflict probability is greater than the preset conflict probability threshold, the downstream associated nodes of the node with the maximum conflict probability in the initial target path are queried based on the node state association graph to determine several downstream associated nodes. The propagation impact value of the node with the maximum conflict probability in the initial target path and several downstream associated nodes are calculated respectively. The sum of the maximum conflict probability and the maximum propagation impact value is calculated, and the initial target path with a sum greater than or equal to the preset conflict sum value threshold is deleted.

[0135] The second filtering unit is used to obtain the average conflict probability, spatiotemporal load entropy and total path distance of the remaining initial target paths; based on the average conflict probability, spatiotemporal load entropy and total path distance of the remaining initial target paths, each remaining initial target path is evaluated, and the remaining initial target path corresponding to the maximum evaluation value is taken as the final target path.

[0136] In this embodiment, the dependency relationship between nodes mainly refers to the influence relationship between nodes, that is, the degree of impact on downstream nodes when a node becomes congested.

[0137] In this embodiment, the collision probability of a node includes:

[0138]

[0139] in, This represents the collision probability of the i-th node; Indicates the i-th node. The number of tasks corresponding to each task; This represents the total number of tasks at the i-th node; This represents the maximum task capacity of the i-th node; The health correction factor for the transmission device at the location of the i-th node;

[0140] In this embodiment, the propagation impact values ​​of the node corresponding to the maximum conflict probability in the initial target path and several downstream associated nodes are calculated, including:

[0141]

[0142] in, This represents the propagation impact value between node i, which corresponds to the node with the maximum conflict probability in the initial target path, and the j-th downstream associated node. This represents the maximum probability of conflict among nodes in the initial target path; This represents the ratio of the amount of tasks transmitted from the i-th node to the j-th node to the total amount of tasks transmitted from the i-th node.

[0143] In this embodiment, the average conflict probability of the initial target path is the average of the conflict probabilities of each node in the initial target path.

[0144] In this embodiment, the spatiotemporal load entropy is the load balance of each node in the initial target path; the lower the entropy value, the more balanced the load.

[0145] In this embodiment, each remaining initial target path is evaluated based on the average conflict probability, spatiotemporal load entropy, and total path distance of the remaining initial target paths, including:

[0146]

[0147] in, This represents the evaluation value of the initial target path; , , This indicates a preset weight, with a sum of 1. This represents the average conflict probability corresponding to the g-th initial target path; This represents the spatiotemporal load entropy corresponding to the g-th initial target path; This represents the total path distance corresponding to the g-th initial target path, ignoring the dimensions.

[0148] The working principle and beneficial effects of the above technical solution are as follows: Relationships are extracted from the nodes of historical transmission paths in historical transmission data, and a node state association graph is constructed. This helps to present complex node relationships in a visual way, while deeply exploring the potential associations between nodes, providing a comprehensive basis for subsequent path selection; Real-time acquisition of the current task data and pre-allocated task volume of nodes in the initial target path within a preset time period is used to calculate the conflict probability; This method can predict potential task conflicts in the path in advance; When the maximum conflict probability exceeds a preset conflict probability threshold, downstream related nodes are determined by querying the node state association graph, and the propagation impact value is calculated, effectively avoiding the selection of paths that may cause transmission delays or failures due to task conflicts, thus improving the stability and reliability of the entire transmission system; The sum of the maximum conflict probability and the maximum propagation impact value is calculated, and initial target paths with a sum greater than or equal to a preset conflict sum value threshold are deleted; This can accurately screen out high-risk initial target paths, ensuring that the final remaining paths have a low risk of task conflicts and guaranteeing the smooth progress of the transmission task.

[0149] Example 9: Adjustment submodule, including:

[0150] The determination unit is used to segment the final target path into different segment areas and query the commodity type-transmission speed table to determine the initial transmission speed of different segment areas.

[0151] The second acquisition unit is used to acquire the load rate of each segment of the final target path in real time.

[0152] The dynamic adjustment unit is used to dynamically adjust the initial conveying speed of goods in each segment based on the load rate of each segment.

[0153] In this embodiment, the load rate of each segment is compared with a preset load rate threshold. If the load rate is less than or equal to the preset load rate threshold, transmission is performed based on the initial transmission speed. If the load rate is greater than the preset load rate threshold but less than 1, transmission is performed at a 20% reduction based on the initial transmission speed.

[0154] The working principle and beneficial effects of the above technical solution are as follows: the final target path is segmented, and then the initial transmission speed for different segment areas is determined by consulting the product type-transmission speed table; the path is further subdivided, and the initial transmission speed is determined for different segments based on product type, making the transmission speed planning more refined. Different segment areas may have different environmental and equipment conditions. Determining an appropriate initial speed based on product type can better adapt to the characteristics of each segment, improving transmission efficiency while ensuring the quality of product transmission.

[0155] To achieve the above objectives, such as Figure 3 As shown, a second aspect of the present invention proposes a transmission method based on AI product recognition, including steps S1-S4:

[0156] S1: Real-time acquisition of image data and sensor data of the goods to be transported in the conveying equipment;

[0157] S2: Based on the AI ​​model, feature extraction is performed on the image data and sensor data of the goods to be delivered, and the extracted features are combined to obtain the first feature;

[0158] S3: Input the first feature into the pre-trained product recognition model for recognition and determine the recognition result;

[0159] S4: Adjust the delivery information of the goods to be delivered based on the recognition results.

[0160] The working principle and beneficial effects of the above technical solution are as follows: by combining image data and sensor data, the accuracy of image recognition is improved; based on the accurate image recognition results, the transmission information of goods in the transmission device is adjusted, which improves the coordination between image recognition and goods transmission, effectively improves the speed and accuracy of goods transmission, and improves the efficiency of goods transmission.

[0161] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A delivery system based on AI product recognition, characterized in that, include: The acquisition module is used to acquire image data and sensor data of the goods to be transported in the conveying equipment in real time; The AI ​​feature combination module is used to extract features from the image data and sensor data of the goods to be delivered based on the AI ​​model, and combine the extracted features to obtain the first feature; The product recognition module is used to input the first feature into a pre-trained product recognition model for recognition and to determine the recognition result; The adjustment module is used to adjust the delivery information of the goods to be delivered based on the recognition results; Adjustment modules include: A submodule is built to obtain historical transmission data from the transmission device and to build a path database based on the historical transmission data. The second combination submodule is used to extract features from other related dimensions of the classified goods to be transmitted based on the goods classification information in the recognition results, to obtain supplementary features of related dimensions, and to combine the supplementary features of related dimensions with the first feature to obtain the second feature; The determination submodule is used to calculate the feature-weighted cosine similarity between the second feature and the path database, compare the feature-weighted cosine similarity with a preset similarity threshold, and take the path with the feature-weighted cosine similarity greater than or equal to the preset similarity threshold as the initial target path; The filtering submodule is used to filter the initial target path and determine the final target path; The adjustment submodule is used to dynamically adjust the conveying speed of the goods to be delivered in the final target path; Adjust the submodules, including: The determination unit is used to segment the final target path into different segment areas and query the commodity type-transmission speed table to determine the initial transmission speed of different segment areas. The second acquisition unit is used to acquire the load rate of each segment of the final target path in real time. The dynamic adjustment unit is used to dynamically adjust the initial conveying speed of goods in each segment based on the load rate of each segment.

2. The delivery system based on AI product recognition as described in claim 1, characterized in that, The data acquisition module includes: The first acquisition submodule is used to acquire images of the goods to be transported in the conveying equipment based on the industrial camera, and obtain image data of the goods to be transported. The second acquisition submodule is used to acquire the weight data of the goods to be transported in the conveying device based on the weight sensor, and to acquire the size data of the goods to be transported based on the laser volume measuring instrument, and to acquire the size data of the goods to be transported; and to use the weight data and size data of the goods to be transported as sensor data of the goods to be transported.

3. The delivery system based on AI product recognition as described in claim 1, characterized in that, The AI ​​feature combination module includes: The AI ​​feature extraction submodule is used to extract features from the image data and sensor data of the goods to be delivered based on the AI ​​model, so as to obtain the target features. The normalization submodule is used to normalize the target features to obtain the normalized target features. The first combination submodule is used to combine the normalized target features to generate the first feature.

4. The delivery system based on AI product recognition as described in claim 1, characterized in that, Training methods for product recognition models include: Acquire image data, weight data, and dimension data of goods historically transported by the conveying equipment; The image data, weight data, and size data of historically delivered goods are preprocessed and labeled to obtain a labeled dataset of historically delivered goods. The labeled historical transmitted product dataset is divided into a product recognition training dataset and a product recognition test dataset based on a preset ratio. The training image data in the commodity recognition training dataset is input into the convolutional layer of the neural network model to extract features from the training image data and obtain the local features corresponding to the training image data. The generated local features are then input into the pooling layer, which performs dimensionality reduction on the local features output by the convolutional layer to obtain the visual feature vector. The preprocessed weight and size data of the products are concatenated to generate a structured feature vector; Visual feature vectors and structured feature vectors are concatenated to form a fused feature vector; the fused feature vector is input into a fully connected layer, and deep correlations between features are learned through nonlinear transformation; finally, the output layer outputs N probability values ​​for product recognition. The training samples in the product recognition training dataset are sequentially input into the neural network model for training. When the training results meet the requirements, the initial product recognition model is obtained. The initial product recognition model was tested using a product recognition test dataset. When the test results met the requirements, a well-trained product recognition model was obtained.

5. The delivery system based on AI product recognition as described in claim 1, characterized in that, Build submodules, including: The first acquisition unit is used to acquire historical transmission data of the transmission device; The first extraction unit is used to extract the transmission nodes in the historical transmission path included in the historical transmission data to obtain the target node; The second extraction unit is used to extract the spatiotemporal features of the target node in the historical transmission path and determine the spatiotemporal feature information corresponding to the target node. The first construction unit is used to construct a four-dimensional grouped feature based on the target node and spatiotemporal feature information, consisting of a starting point, an end point, spatiotemporal features, and a commodity type; and to construct a path database based on the four-dimensional grouped feature.

6. The delivery system based on AI product recognition as described in claim 1, characterized in that, The second combined submodule includes: The classification unit is used to classify the goods to be delivered based on the commodity classification information included in the identification results; The third extraction unit is used to extract other related dimensions of the classified goods to be delivered, including goods type features, environmental related features, operation adaptation features, and time constraint features. The second construction unit is used to construct supplementary features for the association dimension based on the extracted product type features, environmental association features, operation adaptation features, and time constraint features. The combination unit is used to combine the supplementary features of the related dimension with the first feature to obtain the second feature.

7. The delivery system based on AI product recognition as described in claim 1, characterized in that, The filtering submodule includes: The graph construction unit is used to extract the dependencies between nodes in the historical transmission path included in the historical transmission data, and to construct a node state association graph. The first filtering unit is used to acquire the current task data and pre-allocated task volume of each node in the initial target path in real time, and calculate the conflict probability of each node in each initial target path. If the maximum conflict probability is greater than the preset conflict probability threshold, the downstream associated nodes of the node with the maximum conflict probability in the initial target path are queried based on the node state association graph to determine several downstream associated nodes. The propagation impact value of the node with the maximum conflict probability in the initial target path and several downstream associated nodes are calculated respectively. The sum of the maximum conflict probability and the maximum propagation impact value is calculated, and the initial target path with a sum greater than or equal to the preset conflict sum value threshold is deleted. The second filtering unit is used to obtain the average conflict probability, spatiotemporal load entropy and total path distance of the remaining initial target paths; based on the average conflict probability, spatiotemporal load entropy and total path distance of the remaining initial target paths, each remaining initial target path is evaluated, and the remaining initial target path corresponding to the maximum evaluation value is taken as the final target path.

8. A delivery method based on AI product recognition, characterized in that, include: Real-time acquisition of image data and sensor data of goods to be transported in the conveying equipment; Based on the AI ​​model, feature extraction is performed on the image data and sensor data of the goods to be delivered, and the extracted features are combined to obtain the first feature; The first feature is input into a pre-trained product recognition model for recognition, and the recognition result is determined. Adjust the delivery information of the goods to be delivered based on the recognition results; Based on the recognition results, the delivery information of the goods to be delivered is adjusted, including: Obtain historical transmission data from the transmission equipment and build a path database based on the historical transmission data; Based on the product classification information in the recognition results, features are extracted from other related dimensions of the classified products to be transmitted to obtain supplementary features of related dimensions. The supplementary features of related dimensions are then combined with the first feature to obtain the second feature. Calculate the feature-weighted cosine similarity between the second feature and the path database, compare the feature-weighted cosine similarity with a preset similarity threshold, and take the path with the feature-weighted cosine similarity greater than or equal to the preset similarity threshold as the initial target path; Filter the initial target paths to determine the final target path; The delivery speed of the goods to be delivered along the final destination path is dynamically adjusted; The delivery speed of the goods to be delivered along the final destination path is dynamically adjusted, including: The final target path is segmented into different segment areas, and the product type-transmission speed table is queried to determine the initial transmission speed of each segment area. Real-time acquisition of load rates for each segment of the final target path; The initial conveying speed of goods in each segment is dynamically adjusted based on the load rate of each segment.

Images