High-efficiency logistics line

Abstract

The present application relates to the field of machine vision technology, and in particular to a high-efficiency logistics line. The high-efficiency logistics line includes the following parts: an information acquisition module, which is used to obtain the source information of the logistics package pile sorted by the current logistics transfer station, and thereby obtain the weather conditions of the transportation route from the source to the current logistics transfer station in the current transportation process; a transfer learning module, which is used to transfer learning the pre-trained logistics waybill recognition basic model based on the weather conditions from the source to the current logistics transfer station using the waybill data collected from the transportation route, so as to generate a logistics waybill recognition model that is adapted to the corresponding damage type of the transportation route under different weather conditions; an identification module, which selects the logistics waybill recognition model trained with the waybill data collected during the corresponding weather conditions according to the weather conditions from the source to the current logistics transfer station, so as to identify the logistics waybill. The present application can identify express waybills more accurately.

Description

Technical Field

[0001] The present application relates to the field of machine vision technology, and in particular to a high-efficiency logistics line. Background Art

[0002] With the rapid development of e-commerce and the logistics industry, the number of express parcels has exploded. To meet the demand for efficient and accurate logistics, many logistics companies have introduced automated parcel sorting and identification systems. These systems typically rely on scanning and identifying package labels to capture key data such as destination and recipient information, enabling automated parcel sorting and forwarding.

[0003] However, in actual logistics and transportation, express delivery labels are often damaged by various factors. For example, in rainy or snowy weather, packages may be exposed to moisture or rain, causing the information on the label to become blurred or blurred. During transportation, packages may be damaged, worn, or contaminated by physical forces such as friction and collision. These damages can adversely affect the accuracy and efficiency of the package identification system, potentially preventing the package from being correctly identified and delaying delivery.

[0004] Current package identification systems typically use fixed image processing and character recognition algorithms to identify various types of shipping labels. However, these algorithms have varying degrees of success when it comes to identifying shipping labels with varying levels of damage, making it difficult to adapt to complex damage scenarios. Summary of the Invention

[0005] In order to solve the above technical problems or at least partially solve the above technical problems, the present application provides a high-efficiency logistics line that can more accurately identify express delivery bills.

[0006] The present application provides a high-efficiency logistics line, which includes the following parts:

[0007] An information acquisition module is used to obtain the source information of the logistics package pile sorted by the current logistics transfer station, and thereby obtain the weather conditions of the transportation route from the source to the current logistics transfer station during the current transportation process;

[0008] A transfer learning module is used to perform transfer learning on the pre-trained logistics waybill recognition basic model based on the weather conditions from the source to the current logistics transfer station, using the waybill data collected from the transportation route, to generate a logistics waybill recognition model that is adapted to the corresponding damage types under different weather conditions on the transportation route;

[0009] The recognition module selects the logistics waybill recognition model obtained by training the waybill data collected during the corresponding weather conditions according to the weather conditions from the source to the current logistics transfer station to recognize the logistics waybill.

[0010] Optionally, the high-efficiency logistics line also includes the following parts:

[0011] The high-wear identification module identifies logistics waybills with a wear degree above a preset wear threshold as waybills to be processed;

[0012] The pre-worn bill transformation module is used to train a bill transformation model based on the worn bill image data in the transportation route from the current logistics transfer station to the next logistics node, transform the bill to be processed using the bill transformation model, and generate a pre-worn bill image of the current bill to be processed;

[0013] A data transmission module, used for sending the pre-worn bill image and corresponding bill information to the next logistics node;

[0014] The classification module further performs the following steps:

[0015] Used at the next logistics node, when the waybill cannot be recognized normally, the current package waybill image is used to compare with all pre-worn waybill images to identify the waybill information of the closest pre-worn waybill image as the waybill information of the currently unrecognizable logistics waybill.

[0016] Optionally, a waybill transformation model is trained based on the worn-out image data of the waybill in the transportation route from the current logistics transfer station to the next logistics node, including the following steps:

[0017] Collect images of waybills before and after actual transportation from the current logistics transfer station to the next logistics node under different weather conditions to construct the first training dataset;

[0018] A generative adversarial network is used as the model framework of the face order transformation model. The weather information during actual transportation and the face order image before actual transportation in the first training data set are used as inputs of the model framework of the face order transformation model. The face order image after actual transportation is used as the label of the model framework of the face order transformation model. The model framework of the face order transformation model is trained to obtain the face order transformation model.

[0019] Optionally, the logistics bill recognition basic model is obtained by the following steps:

[0020] Collect logistics waybill data under different weather conditions. The logistics waybill data includes images of normal conditions, images of images of images of images of images of images of images of images of images of images of images of images of images of images of images of images of images of images of images of images of images of images of images of images of images of images of images of images of images in order to construct a second training dataset.

[0021] The convolutional neural network is used as the network framework of the logistics waybill recognition basic model. The logistics waybill image data in the second training data machine is used as the input of the waybill recognition basic model, and the corresponding waybill information is used as the label of the waybill recognition basic model. The logistics waybill recognition basic model is trained in this way.

[0022] Optionally, using the bill data collected from the transportation process, transfer learning is performed on the pre-trained logistics bill recognition basic model to generate a logistics bill recognition model that is adapted to the corresponding damage types of the transportation route under different weather conditions, including the following steps:

[0023] Collecting the waybill data generated during the transportation process under different weather conditions of the transportation route;

[0024] Add a preset number of fully connected layers before the output layer of the final logistics waybill recognition basic model, and initialize the weights of the added fully connected layers to obtain the logistics waybill recognition training model to be transferred;

[0025] The logistics waybill recognition training model to be transferred is trained using the waybill data generated during the transportation process under different weather conditions, so as to obtain a logistics waybill recognition model that is adapted to the corresponding damage types of the transportation route under different weather conditions.

[0026] Optionally, the different weather conditions include rainy days and sunny days.

[0027] The technical solution provided by this application has the following advantages compared with the existing technology:

[0028] One of its benefits is that damage patterns vary under different weather conditions. Furthermore, since the transporters and handling environments are often fixed, the damage patterns caused by these weather conditions tend to form a fixed pattern. Directly using a general recognition model on the delivery document data fails to exploit this fixed pattern, making it difficult to accurately identify the logistics information recorded on the damaged delivery document.

[0029] In this application, the system first uses the transfer learning module to perform targeted transfer learning on the pre-trained recognition model for the corresponding damage caused under different weather conditions on the transportation route, thereby obtaining a logistics waybill recognition model that adapts to the corresponding damage types of the transportation route under different weather conditions.

[0030] In this way, the present application realizes the ability to dynamically adjust the recognition direction of the face bill recognition model according to actual environmental conditions and damage characteristics, thereby improving the recognition accuracy and efficiency of damaged face bills.

[0031] The second beneficial effect is that it should be noted that if we only select different recognition models to identify and read the waybill information based on different conditions (such as weather conditions, environmental factors), this method is usually more suitable for dealing with the problem of identification after the first damage occurs during sorting at the transfer station.

[0032] However, during the initial transport phase, the package label may become damp due to rain, causing the paper to soften and the ink to smudge. Later, during handling, friction and collisions can cause the label to tear and fray. This combination of damage exacerbates the severity of the label damage, presenting a complex and unpredictable pattern. This complex damage often exceeds the processing capabilities of the recognition model, rendering the aforementioned methods ineffective. Therefore, simply adapting the recognition model to each specific condition cannot effectively address the difficulties in label recognition caused by the accumulation of wear and tear during multiple transports.

[0033] This application uses the system to identify waybills whose degree of wear exceeds a preset threshold at the current logistics transfer station, and estimates that these waybills may be further damaged in the next transportation stage due to the large degree of wear they have already suffered, resulting in inability to be properly identified at the next logistics node.

[0034] In the current logistics transfer station, the system of this application uses a high wear recognition module to identify logistics waybills with a degree of wear exceeding a preset threshold, and marks them as waybills to be processed. For these waybills to be processed, the pre-worn waybill conversion module uses the collected waybill wear image data to train a waybill conversion model in advance based on the transportation route and weather conditions from the current logistics transfer station to the next logistics node. This model can simulate the further damage that may occur to the waybill in the next transportation stage. The waybill to be processed is transformed using this model, and further pre-worn waybill images are generated based on the wear that has affected the information, and these pre-worn waybill images and corresponding waybill information are sent to the next logistics node through the data transmission module. When the package arrives at the next logistics node, if the waybill cannot be normally identified due to severe wear, the classification module will use the damaged waybill image of the current package to compare with the pre-worn waybill image transmitted in advance, and obtain the waybill information of the closest pre-worn waybill image through matching, so as to achieve effective identification of the unrecognizable waybill.

[0035] The principle of this method does not rely on the restoration of damaged information. Even for highly damaged and unrecognizable waybill information, this method can be used to screen out the closest waybill information. It only needs to be screened out and then manually confirmed, which is still helpful in determining the waybill information of the damaged waybill. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A schematic flow chart of the method performed by the high-efficiency logistics line provided in an embodiment of the present application. DETAILED DESCRIPTION

[0037] The technical solution in this application will be described below with reference to the accompanying drawings.

[0038] The following description sets forth many specific details to facilitate a thorough understanding of the present application. However, the present application may also be implemented in other ways than those described herein. It is apparent that the embodiments described in the specification are only some of the embodiments of the present application, not all of them. It should be noted that the embodiments of the present application and the features therein may be combined with each other unless there is a conflict.

[0039] The present application provides a high-efficiency logistics line, which includes the following parts:

[0040] S101: An information acquisition module is used to obtain the source information of the logistics package pile sorted by the current logistics transfer station, and thereby obtain the weather conditions of the transportation route from the source to the current logistics transfer station during the current transportation process.

[0041] Specifically, the high-efficiency logistics line of the present application is used in the logistics sorting line of a logistics transfer station. It is necessary to sort the logistics parcels sent to the transfer station before sending them to the next transfer station. In an embodiment of the present application, when the logistics parcels begin to be transferred on the parcel sorting conveyor belt, the packaging and shipping location of the current logistics parcels is identified by manually entering or recognizing an identification code, which serves as the source information of the logistics parcels.

[0042] The weather conditions of the transport route include at least the weather at the packaging location at the time of the packaging batch, which is at least classified as sunny or rainy. The weather at the packaging location at the time of the packaging batch can be obtained through the API interface of a third-party weather data provider.

[0043] S102: A transfer learning module is used to perform transfer learning on the pre-trained logistics waybill recognition basic model based on the weather conditions from the source to the current logistics transfer station, using the waybill data collected from the transportation route, to generate a logistics waybill recognition model that is adapted to the corresponding damage types of the transportation route under different weather conditions.

[0044] Specifically, the logistics bill recognition basic model is obtained through the following steps:

[0045] Collect logistics waybill data under different weather conditions, including images of the waybill in normal state, wet in rainy weather, damaged by friction and wear, and damaged by collision, as well as the waybill information recorded in the waybill images, to construct a second training dataset;

[0046] The convolutional neural network is used as the network framework of the logistics waybill recognition basic model. The logistics waybill image data in the second training data machine is used as the input of the waybill recognition basic model, and the corresponding waybill information is used as the label of the waybill recognition basic model. The logistics waybill recognition basic model is trained in this way.

[0047] Specifically, using the waybill data collected from the transportation process, transfer learning is performed on the pre-trained logistics waybill recognition basic model to generate a logistics waybill recognition model that is adapted to the corresponding damage types of the transportation route under different weather conditions, including the following steps:

[0048] Collect the logistics bill data generated during the transportation process under different weather conditions of the transportation route.

[0049] Specifically, in the embodiment of the present application, the logistics waybill image and the corresponding waybill information after the transportation process are collected at least under sunny and rainy conditions.

[0050] Add a preset number of fully connected layers before the output layer of the final logistics waybill recognition basic model, and initialize the weights of the added fully connected layers to obtain the logistics waybill recognition training model to be transferred;

[0051] Specifically, the preset number here is generally set to 3-5 layers.

[0052] The weight initialization operation can generally set the weight of the newly added fully connected layer to 0, or other independently set values.

[0053] The logistics waybill recognition training model to be transferred is trained using the waybill data generated during the transportation process under different weather conditions, so as to obtain a logistics waybill recognition model that is adapted to the corresponding damage types of the transportation route under different weather conditions.

[0054] Specifically, after the above transfer learning operation, at least the logistics waybill recognition model for the transportation route on sunny days and the logistics waybill recognition model for rainy days will be learned.

[0055] S103: The recognition module selects a logistics waybill recognition model trained with the waybill data collected during the corresponding weather conditions according to the weather conditions from the source to the current logistics transfer station to recognize the logistics waybill.

[0056] S104: A high wear identification module identifies logistics waybills with a wear degree above a preset wear threshold as waybills to be processed.

[0057] Specifically, the preset wear threshold is a manually set value.

[0058] The degree of wear of different logistics labels is evaluated using the following formula:

[0059]

[0060] Among them, the confidence score is the confidence level of the logistics waybill recognition model in its output results, generally speaking, it is the probability of the output result in the softmax layer.

[0061] The average image recognition time is the average processing time for the system to identify the label, and can also be a manually set benchmark value.

[0062] The current image recognition time is the processing time for the logistics waybill.

[0063] α and β are weight coefficients, both of which are artificially set values.

[0064] S105: A pre-worn face order transformation module is used to train a face order transformation model based on the face order wear image data in the transportation route from the current logistics transfer station to the next logistics node, transform the face order to be processed through the face order transformation model, and generate a pre-worn face order image of the current face order to be processed.

[0065] Specifically, the waybill transformation model is trained based on the worn image data of the waybill in the transportation route from the current logistics transfer station to the next logistics node, including the following steps:

[0066] Collect images of waybills before and after actual transportation from the current logistics transfer station to the next logistics node under different weather conditions to construct the first training dataset;

[0067] A generative adversarial network is used as the model framework of the face order transformation model. The weather information during actual transportation and the face order image before actual transportation in the first training data set are used as inputs of the model framework of the face order transformation model. The face order image after actual transportation is used as the label of the model framework of the face order transformation model. The model framework of the face order transformation model is trained to obtain the face order transformation model.

[0068] Specifically, a pre-worn waybill image is generated based on images with a wear degree above a preset wear threshold and a waybill transformation model corresponding to the transportation route.

[0069] S106: A data transmission module is used to send the pre-worn bill image and corresponding bill information to the next logistics node.

[0070] S107: At the next logistics node, the classification module of the next logistics node further performs the following steps:

[0071] Used to compare the current package's waybill image with all pre-worn waybill images at the next logistics node when the waybill cannot be recognized normally, so as to identify and obtain the waybill information of the pre-worn waybill image that is closest to the unrecognizable waybill image, as the waybill information of the currently unrecognizable logistics waybill.

[0072] Then, the unrecognizable waybills and the matched waybill information are manually reviewed.

[0073] In summary, the beneficial effects of the high-efficiency logistics line provided by the embodiments of the present application are discussed as follows:

[0074] One of its benefits is that damage patterns vary under different weather conditions. Furthermore, since the transporters and handling environments are often fixed, the damage patterns caused by these weather conditions tend to form a fixed pattern. Directly using a general recognition model on the delivery document data fails to exploit this fixed pattern, making it difficult to accurately identify the logistics information recorded on the damaged delivery document.

[0075] In this application, the system first uses the transfer learning module to perform targeted transfer learning on the pre-trained recognition model for the corresponding damage caused under different weather conditions on the transportation route, thereby obtaining a logistics waybill recognition model that adapts to the corresponding damage types of the transportation route under different weather conditions.

[0076] In this way, the present application realizes the ability to dynamically adjust the recognition direction of the face bill recognition model according to actual environmental conditions and damage characteristics, thereby improving the recognition accuracy and efficiency of damaged face bills.

[0077] The second beneficial effect is that it should be noted that if we only select different recognition models to identify and read the waybill information according to different conditions (such as weather conditions, environmental factors), this method is usually more suitable for dealing with the problem of identification after the first damage occurs during sorting at the transfer station.

[0078] However, during the initial transport phase, the package label may become damp due to rain, causing the paper to soften and the ink to smudge. Later, during handling, friction and collisions can cause the label to tear and fray. This combination of damage exacerbates the severity of the label damage, presenting a complex and unpredictable pattern. This complex damage often exceeds the processing capabilities of the recognition model, rendering the aforementioned methods ineffective. Therefore, simply adapting the recognition model to each specific condition cannot effectively address the difficulties in label recognition caused by the accumulation of wear and tear during multiple transports.

[0079] This application uses the system to identify waybills whose degree of wear exceeds a preset threshold at the current logistics transfer station, and estimates that these waybills may be further damaged in the next transportation stage due to the large degree of wear they have already suffered, resulting in inability to be properly identified at the next logistics node.

[0080] In the current logistics transfer station, the system of this application uses a high wear recognition module to identify logistics waybills with a degree of wear exceeding a preset threshold, and marks them as waybills to be processed. For these waybills to be processed, the pre-worn waybill conversion module uses the collected waybill wear image data to train a waybill conversion model in advance based on the transportation route and weather conditions from the current logistics transfer station to the next logistics node. This model can simulate the further damage that may occur to the waybill in the next transportation stage. The waybill to be processed is transformed using this model, and further pre-worn waybill images are generated based on the wear that has affected the information, and these pre-worn waybill images and corresponding waybill information are sent to the next logistics node through the data transmission module. When the package arrives at the next logistics node, if the waybill cannot be normally identified due to severe wear, the classification module will use the damaged waybill image of the current package to compare with the pre-worn waybill image transmitted in advance, and obtain the waybill information of the closest pre-worn waybill image through matching, so as to achieve effective identification of the unrecognizable waybill.

[0081] The principle of this method is that it does not require the restoration of damaged information. Even for highly damaged and unrecognizable waybill information, this method can filter out the closest waybill information. It only needs to be filtered out and then manually confirmed, which is still helpful in determining the waybill information of the damaged waybill.

[0082] In summary, the logistics waybill identification process generally occurs in three stages: collection, transit and distribution.

[0083] During the process from collection to the transfer station: Since the waybill recognition problem encountered at the initial transfer station is a problem of recognition after the first damage occurs, the solution proposed in this application can dynamically adjust the recognition direction of the recognition model, and perform more efficient and accurate recognition of the logistics waybill based on the weather and common damage conditions at the source.

[0084] During the process of distribution to the delivery personnel at the transfer station: When the waybill has been worn out after multiple transportations and cannot be recognized normally in the logistics process, a pre-worn waybill image can be generated in advance while it can still be accurately recognized during transfer. During distribution and delivery, the pre-worn waybill image can be used to match the unrecognizable waybill, thereby obtaining the damaged and difficult-to-recognize waybill information more accurately and quickly, thereby achieving faster logistics delivery.

[0085] It should also be noted that the embodiments of the present application utilize the advantages of both methods and organically combine them.

[0086] During the initial recognition process, it is not appropriate to generate a pre-worn face sheet image in advance, because the location, shape, and size of the contamination that caused the unrecognizable face sheet image cannot be determined at the outset, and subsequent morphological changes of the contamination cannot be inferred, and naturally, effective face sheet contamination inference cannot be made. Therefore, this application takes advantage of the fact that the contamination type is single from the beginning during the collection to the transfer recognition process, and adopts a method of dynamically adjusting the recognition direction of the recognition model to achieve accurate and efficient recognition.

[0087] Subsequently, we also took into account the problem of severely damaged logistics waybills, which could not be identified due to the accumulation of contamination after multiple transfers. We adopted the method of generating pre-worn waybill images in advance. During distribution and delivery, we could use the pre-worn waybill images to match the unrecognizable waybills, so as to obtain the information of damaged and difficult-to-identify waybills more accurately and quickly.

[0088] Therefore, the high-efficiency logistics line provided in the embodiment of the present application can more accurately identify logistics waybills.

[0089] It should be noted that, in this document, relational terms such as "first" and "second" are used solely to distinguish one entity or operation from another, and do not necessarily require or imply any actual relationship or order between these entities or operations. Furthermore, the terms "include," "comprises," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, article, or device. Without further limitation, an element defined by the phrase "comprising a..." does not preclude the presence of additional identical elements in the process, method, article, or device comprising the element. Furthermore, in the description of the embodiments of this application, unless otherwise specified, " / " represents or. For example, A / B can represent either A or B. "And / or" herein is merely a description of an associative relationship between associated objects, indicating that three relationships can exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, in the description of the embodiments of the present application, “plurality” refers to two or more than two.

[0090] The foregoing description is intended only to provide specific embodiments of the present application, which will enable those skilled in the art to understand and implement the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments described herein, but is intended to be construed in the broadest manner consistent with the principles and novel features disclosed herein.

Claims

1. High efficiency logistics line, characterized by: The high-efficiency logistics line includes the following parts: An information acquisition module is used to obtain the source information of the logistics package pile sorted by the current logistics transfer station, and thereby obtain the weather conditions of the transportation route from the source to the current logistics transfer station during the current transportation process; A transfer learning module is used to perform transfer learning on the pre-trained logistics waybill recognition basic model based on the weather conditions from the source to the current logistics transfer station, using the waybill data collected from the transportation route, to generate a logistics waybill recognition model that is adapted to the corresponding damage types under different weather conditions on the transportation route; The recognition module selects the logistics waybill recognition model obtained by training the waybill data collected during the corresponding weather conditions according to the weather conditions from the source to the current logistics transfer station to recognize the logistics waybill.

2. The high-efficiency logistics line according to claim 1, characterized in that: The high-efficiency logistics line also includes the following parts: The high-wear identification module identifies logistics waybills with a wear degree above a preset wear threshold as waybills to be processed; The pre-worn bill transformation module is used to train a bill transformation model based on the worn bill image data in the transportation route from the current logistics transfer station to the next logistics node, transform the bill to be processed using the bill transformation model, and generate a pre-worn bill image of the current bill to be processed; A data transmission module, used for sending the pre-worn bill image and corresponding bill information to the next logistics node; The classification module performs the following steps: Used at the next logistics node, when the waybill cannot be recognized normally, the current package waybill image is used to compare with all pre-worn waybill images to identify the waybill information of the closest pre-worn waybill image as the waybill information of the currently unrecognizable logistics waybill.

3. The high-efficiency logistics line according to claim 2, characterized in that: The transport bill transformation model is trained based on the worn image data of the transport bill from the current logistics transfer station to the next logistics node, including the following steps: Collect images of waybills before and after actual transportation from the current logistics transfer station to the next logistics node under different weather conditions to construct the first training dataset; A generative adversarial network is used as the model framework of the face order transformation model. The weather information during actual transportation and the face order image before actual transportation in the first training data set are used as inputs of the model framework of the face order transformation model. The face order image after actual transportation is used as the label of the model framework of the face order transformation model. The model framework of the face order transformation model is trained to obtain the face order transformation model.

4. The high-efficiency logistics line according to claim 1, characterized in that: The logistics bill recognition basic model is obtained through the following steps: Collecting logistics waybill data under different weather conditions, including images of waybills in normal state, images of waybills soaked in rain, images of waybills damaged by friction and wear, and images of waybills damaged by collision, and the waybill information recorded in the images of the waybills, to construct a second training dataset; The convolutional neural network is used as the network framework of the logistics waybill recognition basic model. The logistics waybill image data in the second training data machine is used as the input of the waybill recognition basic model, and the corresponding waybill information is used as the label of the waybill recognition basic model. The logistics waybill recognition basic model is trained in this way.

5. The high-efficiency logistics line according to claim 1, characterized in that: Using the waybill data collected from the transportation process, transfer learning is performed on the pre-trained logistics waybill recognition basic model to generate a logistics waybill recognition model that is adapted to the corresponding damage types of the transportation route under different weather conditions, including the following steps: Collecting the waybill data generated during the transportation process under different weather conditions of the transportation route; Add a preset number of fully connected layers before the output layer of the final logistics waybill recognition basic model, and initialize the weights of the added fully connected layers to obtain the logistics waybill recognition training model to be transferred; The logistics waybill recognition training model to be transferred is trained using the waybill data generated during the transportation process under different weather conditions, so as to obtain a logistics waybill recognition model that is adapted to the corresponding damage types of the transportation route under different weather conditions.

6. The high-efficiency logistics line according to claim 1, characterized in that: The different weather conditions include rainy days and sunny days.

Images