AI intelligent logistics warehousing system

Abstract

The invention relates to the technical field of warehouse management, and particularly discloses an AI intelligent logistics warehouse system comprising an acquisition module which acquires video data when a shooting point acquires a truck in a detection interval; the target tracking module is used for tracking and numbering the loaded truck and determining an unloading area code where the loaded truck finally enters; the identification module is used for identifying the logistics box and the bill in the image frame and marking the logistics box as a target; the feature extraction unit extracts features of each level in the image frame and fuses the features of all levels; the detection head divides the fused features into a plurality of network units and predicts a group of bounding boxes for each network unit; each bounding box is associated with a preset anchor point; the detection head predicts the position information of the bounding box corresponding to the network unit according to the anchor point, and predicts the category confidence of the goods; and the result output module outputs the scanning result of the target.

Description

Technical Field

[0001] The present invention relates to the field of warehouse management technology, and in particular to an AI intelligent logistics warehousing system. Background Art

[0002] The demand for logistics is increasing day by day. During the cargo information entry stage, since a large amount of cargo will arrive at the warehouse continuously in multiple batches and at a high frequency, a lot of labor costs are required to extract information for each single box of cargo and enter it into the enterprise ERP system.

[0003] At the same time, due to a large amount of manual intervention in the process, the probability of human errors in the input cargo information has greatly increased, resulting in low efficiency and difficulty in warehouse management. This not only increases the operating costs of enterprises, but also affects customer satisfaction and market competitiveness. Therefore, how to improve the level of informatization and automation in the logistics industry, reduce labor costs, and reduce human errors has become an urgent problem to be solved. Summary of the invention

[0004] The purpose of this invention is to provide an AI intelligent logistics warehousing system to solve the above technical problems.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] AI intelligent logistics warehousing system, including:

[0007] Acquisition module: including a detection area and a shooting point; obtaining the entry time of a truck entering the detection area, and obtaining the exit time of a truck leaving the detection area, and recording the time period between the entry time and the exit time as the detection interval; the shooting point is used to collect video data when the truck is in the detection interval;

[0008] Target tracking module: divides the video data into several image frames, and determines whether the truck is loaded with goods according to the image frames; if the truck is loaded with goods, the truck is recorded as a loading truck; the loading truck is tracked and numbered, and the unloading area number that the loading truck finally enters is obtained;

[0009] Recognition module: including a feature extraction unit and a detection head; the feature extraction unit is a backbone network composed of a number of convolutional layers and residual modules, used to extract the features of the image frame; the detection head is composed of a number of convolutional layers, used to predict the location and category of the goods in the image frame;

[0010] Identify the logistics box and the bill in the image frame, and record the logistics box as the target; the feature extraction unit extracts the features of each level in the image frame, and fuses the features of all levels; the detection head divides all the fused features into a number of network units, and presets a set of bounding boxes for the network units, each of which is associated with a preset anchor point; the detection head predicts the position information of the bounding box corresponding to the network unit according to the anchor point, and predicts the category confidence of the goods, the position information includes the center coordinates, width and height of the bounding box;

[0011] Result output module: obtains the scanned text result according to the bill, and associates the position information, category confidence, unloading area number and the scanned text result to obtain the scanned result of the target.

[0012] As a further solution of the present invention: the process of determining whether a truck is loaded with goods includes:

[0013] The image frame is input into a target tracking network, and the target tracking network is used to identify whether the truck contains cargo; if the truck contains cargo, the truck is recorded as a loaded truck; if the truck does not contain cargo, the truck is recorded as an empty truck.

[0014] As a further solution of the present invention: methods for fusing features of all levels include weighted averaging and splicing.

[0015] As a further solution of the present invention: the process of setting the anchor point includes:

[0016] A number of length values ​​and length-to-width ratio values ​​are set, and the length values ​​and length-to-width ratio values ​​are arranged and combined to obtain a number of groups, and a number of anchor points are obtained according to the length values ​​and length-to-width ratio values ​​of each group; each anchor point is associated with a category confidence and position information, and the category confidence is the item category of the goods.

[0017] As a further solution of the present invention: the process of determining the bounding box corresponding to the network unit includes:

[0018] The confidence value of each bounding box is obtained by non-maximum suppression technology, and the bounding boxes are sorted from high to low according to the confidence value to obtain the bounding box with the highest confidence, which is recorded as the highest confidence box; the intersection and union ratio of each bounding box and the remaining bounding boxes are obtained in turn; if there is a bounding box whose intersection and union ratio with the highest confidence box is greater than a preset threshold, the bounding box is recorded as overlapping with the highest confidence box and the bounding box is deleted; the steps are repeated until only the last bounding box remains.

[0019] As a further solution of the present invention: the preset threshold is usually set to 0.5.

[0020] As a further solution of the present invention: when the logistics box and the bill are not identified in the image frame, continue to identify the next image frame.

[0021] As a further solution of the present invention: the process of obtaining the scanned text result according to the bill includes:

[0022] A segmented image of the bill is obtained in the image frame, wherein the segmented image is an image block where the bill is located in the image frame; the segmented image is input into a text detection model to obtain text area information and text results, wherein the text area information includes the position and size of the text; an image having the same size as the segmented image is generated, and the text result is drawn in the image according to the text area information to obtain a scanned text result.

[0023] Beneficial effects of the present invention:

[0024] The present invention can monitor the entry and exit of trucks in real time by automatically collecting and processing data, reducing the need for manual intervention; the automated target tracking and recognition module improves the speed and accuracy of cargo processing, thereby improving overall work efficiency; it can accurately identify loaded trucks, track and number these trucks, and ensure that the unloading area code of each truck is accurate; through feature extraction and detection heads, it can accurately identify logistics boxes and bills in image frames, thereby improving the accuracy of cargo management; a backbone network composed of a convolutional layer and a residual module is used for feature extraction, which can extract useful information from complex image data; the application of the detection head further refines the judgment of the location and category of the cargo, thereby improving the accuracy of recognition; the result output module associates the location information, category confidence, unloading area code and scanned text results, and provides comprehensive cargo information; it helps to reduce errors and delays, thereby improving logistics efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The present invention will be further described below in conjunction with the accompanying drawings.

[0026] Figure 1 It is a schematic diagram of the process of the AI ​​intelligent logistics warehousing system of the present invention;

[0027] Figure 2 It is a flowchart of the AI ​​intelligent logistics warehousing system of the present invention;

[0028] Figure 3 It is a schematic diagram of the process of the back-end module business in the AI ​​intelligent logistics warehousing system of the present invention;

[0029] Figure 4 It is a flow chart of the front-end display module business in the AI ​​intelligent logistics warehousing system of the present invention. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] See also Figure 1 As shown, the present invention is an AI intelligent logistics warehousing system, comprising:

[0032] Acquisition module: including a detection area and a shooting point; obtaining the entry time of a truck entering the detection area, and obtaining the exit time of a truck leaving the detection area, and recording the time period between the entry time and the exit time as the detection interval; the shooting point is used to collect video data when the truck is in the detection interval;

[0033] It can be understood that the present invention starts from the stage of goods entering the warehouse. In the process of goods entering the warehouse, the pile of goods is used as the identification unit, and video data is obtained from four to five directions (front, back, top and left and right sides of the identification unit) to cover all information surfaces of the identification unit;

[0034] Target tracking module: divides the video data into several image frames, and determines whether the truck is loaded with goods according to the image frames; if the truck is loaded with goods, the truck is recorded as a loading truck; the loading truck is tracked and numbered, and the unloading area number that the loading truck finally enters is obtained;

[0035] As a preferred embodiment of the present invention, the process of determining whether a truck is loaded with goods includes:

[0036] The image frame is input into a target tracking network, and the target tracking network is used to identify whether the truck contains cargo; if the truck contains cargo, the truck is recorded as a loaded truck; if the truck does not contain cargo, the truck is recorded as an empty truck;

[0037] It should be noted that the process of the target tracking network determining whether the truck is loaded with goods includes:

[0038] Acquire appearance features of the truck according to the image frame, the appearance features including changes in vehicle height, contour, and shape; use a convolutional neural network to extract image features of loaded trucks and empty trucks, and classify them to train a deep learning model; determine whether the truck is loaded with goods according to the deep learning model; or combine sensor data on the truck, the sensor data including a weight sensor, monitor dynamic changes of the truck according to the sensor data, such as monitoring the acceleration, changes in the center of gravity, etc. of the truck, and then determine whether the truck is loaded with goods;

[0039] Recognition module: including a feature extraction unit and a detection head; the feature extraction unit is a backbone network composed of a number of convolutional layers and residual modules, used to extract the features of the image frame; the detection head is composed of a number of convolutional layers, used to predict the location and category of the goods in the image frame;

[0040] Identify the logistics box and the bill in the image frame, and record the logistics box as the target; the feature extraction unit extracts the features of each level in the image frame, and fuses the features of all levels; the detection head divides all the fused features into a number of network units, and presets a set of bounding boxes for the network units, each of which is associated with a preset anchor point; the detection head predicts the position information of the bounding box corresponding to the network unit according to the anchor point, and predicts the category confidence of the goods, the position information includes the center coordinates, width and height of the bounding box;

[0041] It should be noted that a rectangular coordinate system is established on the image frame, and the center coordinates are the coordinates corresponding to the center point of the goods on the rectangular coordinate system;

[0042] As a preferred embodiment of the present invention, the method of fusing features of all levels includes weighted averaging and splicing;

[0043] As a preferred embodiment of the present invention, the anchor point setting process includes:

[0044] A number of length values ​​and length-to-width ratio values ​​are set, and the length values ​​and length-to-width ratio values ​​are arranged and combined to obtain a number of groups, and a number of anchor points are obtained according to the length values ​​and length-to-width ratio values ​​of each group; each anchor point is associated with a category confidence and position information, and the category confidence is the item category of the goods;

[0045] It should be noted that this anchor setting method is often used in target detection tasks, especially when using convolutional neural networks (CNNs) for object recognition; by pre-defining a series of anchors of different sizes and shapes, the network can learn how to adjust these anchors during training to best fit the objects in the image; this method helps improve the accuracy and efficiency of detection, especially when dealing with objects with multiple sizes and shapes.

[0046] As a preferred embodiment of the present invention, the process of determining the bounding box corresponding to the network unit includes:

[0047] By using the non-maximum suppression technology, the confidence value of each bounding box is obtained, and the bounding boxes are sorted from high to low according to the confidence value to obtain the bounding box with the highest confidence, which is recorded as the highest confidence box; the intersection and union ratio of each bounding box and the remaining bounding boxes are obtained in turn; if there is a bounding box whose intersection and union ratio with the highest confidence box is greater than a preset threshold, the bounding box is recorded as overlapping with the highest confidence box, and the bounding box is deleted; the steps are repeated until only the last bounding box remains;

[0048] The preset threshold is usually set to 0.5;

[0049] It should be noted that the category confidence in target detection refers to the degree of certainty of the model about the specific category contained in each predicted bounding box, which reflects the probability or confidence of the model that the object in a certain bounding box belongs to a specific category; in deep learning target detection models, category confidence is usually calculated through an activation function called "softmax" or "sigmoid", which can convert the raw scores output by the model into probability values, thereby indicating the model's confidence in each category; in target detection tasks, the model assigns a category confidence to each predicted bounding box; The confidence value can help determine the most likely category of the object in the bounding box; for example, in the YOLOv7 network, each bounding box predicts a category confidence, which indicates the possibility that the object in the bounding box belongs to a certain category; in order to filter out reliable detection results from the model's scanning results, a threshold is usually set; only when the model's confidence in a certain category exceeds the threshold, the prediction is considered reliable; during the training process, the category confidence is also part of the loss function; by minimizing the loss function, the model can learn how to predict the category confidence more accurately, thereby improving the overall detection performance;

[0050] Result output module: obtains the scanned text result according to the bill, and associates the location information, category confidence, unloading area code and the scanned text result to obtain the scanned result of the target;

[0051] As a preferred embodiment of the present invention, the process of obtaining the scanned text result according to the bill includes:

[0052] Acquire a segmented image of the bill in the image frame, wherein the segmented image is an image block where the bill is located in the image frame; input the segmented image into a text detection model to obtain text area information and text results, wherein the text area information includes the position and size of the text; generate an image of the same size as the segmented image, and draw the text result in the image according to the text area information to obtain a scanned text result;

[0053] It can be understood that after obtaining all the information surfaces of the identification unit, the present invention divides the box, quantity and bill of each information surface to be tested, reads the text information of the bill, and then processes and analyzes the read results to meet the enterprise's requirements for data accuracy and intelligent management after warehousing, and finally uploads the processed accurate information to the enterprise ERP to complete the closed loop of the warehousing process;

[0054] The present invention also includes a back-end module, which includes a back-end interface, and the back-end interface receives the scan result and records the scan result as AI data; obtains the company name as a unique ID for account verification to ensure that the received data is legal and valid; the back-end module calls the login interface to verify the user's authority and identity; extracts corresponding bill data from the database according to the user's request, and performs matching operations to ensure the accuracy and completeness of the data, wherein the bill data is a scan text result; performs corresponding operations on the database according to the scan text result; calls corresponding interfaces according to different requirements, such as query interface, modification interface, and deletion interface; classifies and encapsulates data for subsequent management and retrieval; for example, classifies bills according to dimensions such as time and type, and stores this information in a specific format; after all data processing is completed, returns the processing results and status information to other modules;

[0055] The present invention also includes a front-end display module for displaying the bill data of the logistics box in the database.

[0056] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. AI intelligent logistics warehousing system, characterized by: include: Acquisition module: including detection area and shooting points; Obtaining the time when the truck enters the detection zone, and obtaining the time when the truck exits the detection zone, and recording the time period between the time when the truck enters and exits as the detection interval; the shooting point is used to collect video data when the truck is in the detection interval; Target tracking module: divides the video data into several image frames, and determines whether the truck is loaded with goods according to the image frames; if the truck is loaded with goods, the truck is recorded as a loading truck; the loading truck is tracked and numbered, and the unloading area number that the loading truck finally enters is obtained; Identification module: It includes a feature extraction unit and a detection head; the feature extraction unit is a backbone network composed of a plurality of convolutional layers and residual modules, and is used to extract the features of the image frame; the detection head is composed of a plurality of convolutional layers, and is used to predict the location and category of the goods in the image frame; Identify the logistics box and the bill in the image frame, and record the logistics box as the target; the feature extraction unit extracts the features of each level in the image frame, and fuses the features of all levels; the detection head divides all the fused features into a number of network units, and presets a set of bounding boxes for the network units, each of which is associated with a preset anchor point; the detection head predicts the position information of the bounding box corresponding to the network unit according to the anchor point, and predicts the category confidence of the goods, the position information includes the center coordinates, width and height of the bounding box; Result output module: obtains the scanned text result according to the bill, and associates the position information, category confidence, unloading area number and the scanned text result to obtain the scanned result of the target.

2. The AI ​​intelligent logistics warehousing system according to claim 1 is characterized in that: The process of determining whether a truck is loaded with cargo includes: The image frame is input into a target tracking network, and the target tracking network is used to identify whether the truck contains cargo; if the truck contains cargo, the truck is recorded as a loaded truck; if the truck does not contain cargo, the truck is recorded as an empty truck.

3. The AI ​​intelligent logistics warehousing system according to claim 1 is characterized in that: Methods for fusing features at all levels include weighted averaging and splicing.

4. The AI ​​intelligent logistics warehousing system according to claim 1 is characterized in that: The anchor point setting process includes: A number of length values ​​and length-to-width ratio values ​​are set, and the length values ​​and length-to-width ratio values ​​are arranged and combined to obtain a number of groups, and a number of anchor points are obtained according to the length values ​​and length-to-width ratio values ​​of each group; each anchor point is associated with a category confidence and position information, and the category confidence is the item category of the goods.

5. The AI ​​intelligent logistics warehousing system according to claim 1 is characterized in that: The process of determining the bounding box corresponding to the network unit includes: The confidence value of each bounding box is obtained by non-maximum suppression technology, and the bounding boxes are sorted from high to low according to the confidence value to obtain the bounding box with the highest confidence, which is recorded as the highest confidence box; the intersection and union ratio of each bounding box and the remaining bounding boxes are obtained in turn; if there is a bounding box whose intersection and union ratio with the highest confidence box is greater than a preset threshold, the bounding box is recorded as overlapping with the highest confidence box and the bounding box is deleted; the steps are repeated until only the last bounding box remains.

6. The AI ​​intelligent logistics warehousing system according to claim 5 is characterized in that: The preset threshold is usually set to 0.

5.

7. The AI ​​intelligent logistics warehousing system according to claim 1 is characterized in that: When the logistics box and the bill are not recognized in the image frame, continue to recognize the next image frame.

8. The AI ​​intelligent logistics warehousing system according to claim 1 is characterized in that: The process of obtaining the scanned text result according to the bill includes: A segmented image of the bill is obtained in the image frame, wherein the segmented image is an image block where the bill is located in the image frame; the segmented image is input into a text detection model to obtain text area information and text results, wherein the text area information includes the position and size of the text; an image having the same size as the segmented image is generated, and the text result is drawn in the image according to the text area information to obtain a scanned text result.

Images