Industrial collaborative warehouse management method and system

Abstract

The application provides an industry collaborative storage management method and system, which comprises the following steps: obtaining an initial image element descriptor of a target cargo image; loading the initial image element descriptor into a cargo image processing network which is previously calibrated; obtaining a target image element descriptor output by the cargo image processing network; the cargo image processing network is calibrated in sequence through a multi-layer clustering cost function and a shielding prediction cost function; and then based on the target image element descriptor, a target cargo image which is free from self-defects and external disturbances is determined. The cargo image processing network which is previously deployed is calibrated based on two different cost functions, so that the network can efficiently eliminate self-defects and external disturbances in the image element descriptor, thereby relieving the operation cost of the network and enhancing the image processing capacity.

Description

Technical Field

[0001] This invention relates to the field of data processing, and more specifically, to an industry collaborative warehousing management method and system. Background Technology

[0002] Machine vision and artificial intelligence (AI) have been widely applied in collaborative warehouse management, driven by the urgent need to improve efficiency, reduce costs, and optimize supply chain operations. Traditional warehouse management often relies on manual operations, which suffers from numerous problems such as human error, high labor intensity, and low efficiency. The introduction of machine vision and AI technologies has brought revolutionary changes to warehouse management. Machine vision acquires real-time images and data from inside the warehouse through devices such as cameras and sensors, while AI can efficiently analyze and process this data. Through deep learning algorithms and pattern recognition technology, machine vision and AI can achieve a range of functions, including goods identification and classification, inventory management, intelligent navigation and picking, anomaly detection and security monitoring, and predictive maintenance. This not only improves the automation and accuracy of warehouse operations but also accelerates logistics processes, reduces labor costs, helps reduce error rates caused by human operation, and improves overall management level and efficiency.

[0003] When identifying and classifying goods, the quality of the goods image directly affects the final identification accuracy. However, due to uncontrollable factors in the goods sorting and transportation process, the goods themselves may be damaged. In addition, the equipment shooting and image transmission processes may also result in unqualified image quality. It is of great significance to use artificial intelligence to overcome defects in the captured package images. How to ensure the accuracy and efficiency of this process is a technical issue that needs to be considered. Summary of the Invention

[0004] The purpose of this invention is to provide a

[0005] The technical solution of this invention is implemented as follows:

[0006] In a first aspect, embodiments of the present invention provide an industry-coordinated warehousing management method, applied to electronic devices, the method comprising:

[0007] Obtain the initial image feature descriptor for the target cargo image;

[0008] The initial image feature descriptor is loaded into a pre-calibrated cargo image processing network to obtain the target image feature descriptor output by the cargo image processing network. The cargo image processing network is obtained by sequential calibration through a multi-layer clustering cost function and a masking prediction cost function.

[0009] Based on the target image element descriptor, a target cargo image with self-caused defects and external disturbances removed is obtained;

[0010] The cargo image processing network is calibrated based on the following operations:

[0011] Obtain a cargo image learning sample set, which includes self-causing defect image feature descriptors, noiseless cargo image annotation information, self-causing defect image annotation information, and multi-layer clustering labels;

[0012] Obtain a preset image processing network component, which includes an intermediate network module, a multi-layer clustering module, and a masking prediction module;

[0013] Based on the cargo image learning sample set, the preset image processing network components are sequentially adjusted by removing self-cause defects and external disturbances until the preset image processing network components meet the set convergence requirements, and the adjusted target image processing network components are obtained and determined as the cargo image processing network.

[0014] In an optional implementation, the masking prediction module includes an image masking prediction module and a self-causing defect masking prediction module. The step of sequentially performing self-causing defect removal and external disturbance removal adjustments on the preset image processing network components based on the cargo image learning sample set until the preset image processing network components meet the set convergence requirements includes:

[0015] The self-causing defect image feature descriptor is loaded into the intermediate network module, and a depth-tuned descriptor is generated based on the intermediate network module.

[0016] The depth calibration descriptor is loaded into the multi-layer clustering module, and clustering calibration tags are generated based on the multi-layer clustering module.

[0017] The depth calibration descriptor is loaded into the image masking prediction module, and a noiseless cargo image calibration descriptor is generated based on the image masking prediction module.

[0018] The depth calibration descriptor is loaded into the self-causing defect masking prediction module, and a self-causing defect image calibration descriptor is generated based on the self-causing defect masking prediction module.

[0019] Based on the noiseless cargo image annotation information, the self-caused defect image annotation information, the multi-layer clustering marker, the noiseless cargo image calibration descriptor, the self-caused defect image calibration descriptor, and the clustering calibration marker, a target cost function is generated. Based on the target cost function, the preset image processing network component is sequentially calibrated for self-caused defect removal and external disturbance removal until the preset image processing network component meets the set convergence requirements.

[0020] In an optional implementation, the step of generating a target cost function based on the noiseless cargo image annotation information, the self-causing defect image annotation information, the multi-layer clustering marker, the noiseless cargo image calibration descriptor, the self-causing defect image calibration descriptor, and the clustering calibration marker, and then sequentially performing self-causing defect removal calibration and external disturbance removal calibration on the preset image processing network component based on the target cost function until the preset image processing network component meets the set convergence requirements, includes:

[0021] Based on the clustering calibration mark and the multi-layer clustering mark, a first cost function is determined;

[0022] Based on the noiseless cargo image calibration descriptor and the noiseless cargo image annotation information, a second cost function is determined;

[0023] Based on the self-causing defect image calibration descriptor and the self-causing defect image annotation information, a third cost function is determined;

[0024] Based on the first cost function, the second cost function, and the third cost function, a target cost function for the preset image processing network component is generated. Based on the target cost function, the preset image processing network component is sequentially adjusted to remove self-cause defects and external disturbances until the preset image processing network component meets the set convergence requirements.

[0025] In an optional implementation, the noiseless cargo image annotation information includes first noiseless cargo image annotation information, and the step of determining the second cost function based on the noiseless cargo image calibration descriptor and the noiseless cargo image annotation information includes:

[0026] Based on the noiseless cargo image calibration descriptor and the first noiseless cargo image annotation information, the self-causing defect removal cost function is determined.

[0027] The self-causing defect removal cost function is used as the second cost function, and the first noiseless cargo image annotation information is the image annotation information obtained by cargo images that do not contain self-causing defects but contain external disturbances.

[0028] or;

[0029] The noiseless cargo image annotation information includes second noiseless cargo image annotation information. The step of determining the second cost function based on the noiseless cargo image calibration descriptor and the noiseless cargo image annotation information includes:

[0030] Based on the noiseless cargo image calibration descriptor and the second noiseless cargo image annotation information, the external disturbance removal cost function is determined.

[0031] The external disturbance removal cost function is used as the second cost function, and the second noise-free cargo image annotation information is the image annotation information obtained by cargo images that do not contain self-caused defects and do not contain external disturbances.

[0032] In an optional implementation, the step of generating a target cost function for the preset image processing network component based on the first cost function, the second cost function, and the third cost function, and then sequentially performing self-cause defect removal and external disturbance removal adjustments on the preset image processing network component based on the target cost function until the preset image processing network component meets the set convergence requirements, includes:

[0033] Obtain image precision constraints;

[0034] The corresponding sequential calibration method is determined based on the image accuracy constraints.

[0035] Using the sequential calibration method, a target cost function for the preset image processing network component is generated based on the first cost function, the second cost function, and the third cost function. Based on the target cost function, the preset image processing network component is sequentially calibrated to remove self-cause defects and external disturbances until the preset image processing network component meets the set convergence requirements.

[0036] In an optional implementation, the sequential calibration method includes a first sequential calibration method. The step involves generating a target cost function for the preset image processing network component based on the first cost function, the second cost function, and the third cost function, and then sequentially performing self-cause defect removal calibration and external disturbance removal calibration on the preset image processing network component based on the target cost function until the preset image processing network component meets a set convergence requirement. This includes:

[0037] When the sequential calibration method is the first sequential calibration method, a target cost function of the preset image processing network component is determined based on the first cost function, the second cost function, and the third cost function. The preset image processing network component is then repeatedly calibrated to remove self-cause defects based on the target cost function until the preset image processing network component meets the set convergence requirements, thereby obtaining a self-cause defect-removed network component. The second cost function is determined by the self-cause defect-removed cost function.

[0038] Based on the first cost function, the second cost function, and the third cost function, a target cost function for the self-cause defect removal network component is determined. Based on the target cost function, the self-cause defect removal network component is repeatedly adjusted to remove external disturbances until the self-cause defect removal network component meets the set convergence requirements. The second cost function is obtained by determining the external disturbance removal cost function.

[0039] In an optional implementation, the sequential calibration method includes a second sequential calibration method. The step involves generating a target cost function for the preset image processing network component based on the first cost function, the second cost function, and the third cost function, and then sequentially performing self-cause defect removal calibration and external disturbance removal calibration on the preset image processing network component based on the target cost function until the preset image processing network component meets a set convergence requirement. This includes:

[0040] When the sequential calibration method is the second sequential calibration method, a target cost function for the preset image processing network component is determined based on the first cost function, the second cost function, and the third cost function. The preset image processing network component is then repeatedly calibrated to remove external disturbances based on the target cost function until the preset image processing network component meets the set convergence requirements, thus obtaining an external disturbance removal network component. The second cost function is determined by the external disturbance removal cost function.

[0041] Based on the first cost function, the second cost function, and the third cost function, a target cost function for the external disturbance removal network component is determined. Based on the target cost function, the external disturbance removal network component is repeatedly adjusted to remove self-cause defects until the external disturbance removal network component meets the set convergence requirements. The second cost function is determined by the self-cause defect removal cost function.

[0042] In an optional implementation, acquiring the cargo image learning sample set includes:

[0043] Obtain a first cargo image learning example, which is a cargo image with self-caused defects and external disturbances captured by a shooting device;

[0044] Image feature descriptor mining is performed on the first cargo image learning example to obtain the self-causing defect image feature descriptor.

[0045] Acquire a second cargo image learning example, which includes a noiseless cargo image that does not contain self-caused defects but contains external disturbances and a noiseless cargo image that does not contain self-caused defects and does not contain external disturbances.

[0046] Image feature descriptor mining is performed on the second cargo image learning example to obtain the first noiseless cargo image annotation information and the second noiseless cargo image annotation information;

[0047] Based on the first cargo image learning example and the second cargo image learning example, multi-layer clustering labels are determined.

[0048] In an optional implementation, the cargo image processing network includes an intermediate network module, a multi-layer clustering module, an image masking prediction module, and an autogenic defect masking prediction module. The step of loading the initial image feature descriptor into a pre-calibrated cargo image processing network to obtain the target image feature descriptor output by the cargo image processing network includes:

[0049] The initial image feature descriptor is loaded into the intermediate network module, and a depth descriptor is generated based on the intermediate network module;

[0050] The depth descriptor is loaded into the image masking prediction module, a noiseless image feature descriptor is generated based on the image masking prediction module, and the noiseless image feature descriptor is used as the target image feature descriptor.

[0051] The step of determining the target cargo image by removing intrinsic defects and extrinsic disturbances based on the target image feature descriptor includes:

[0052] Image reconstruction is performed on the target image feature descriptor to obtain a target cargo image that has been cleared of self-caused defects and external disturbances.

[0053] Secondly, the present invention provides an industrial collaborative warehousing management system, including an image acquisition device and an electronic device, wherein the image acquisition device and the electronic device are communicatively connected, the image acquisition device is used to acquire images of goods and send them to the electronic device, the electronic device includes a memory and a processor, the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the method described above.

[0054] The present invention has at least the following beneficial effects:

[0055] The industrial collaborative warehousing management method and system provided in this invention obtains an initial image feature descriptor of a target cargo image, loads the initial image feature descriptor into a pre-calibrated cargo image processing network, and obtains the target image feature descriptor output by the cargo image processing network. The cargo image processing network is obtained through sequential calibration using a multi-layer clustering cost function and a masking prediction cost function. Then, based on the target image feature descriptor, a target cargo image cleaned of intrinsic defects and external disturbances is determined. This invention calibrates the pre-deployed cargo image processing network based on two different cost functions, enabling the network to efficiently remove intrinsic defects and external disturbances in the image feature descriptor, thus not only alleviating the network's computational overhead but also enhancing its image processing capabilities.

[0056] Other features will be described in part in the following description. These features will be partially discovered by those skilled in the art upon examination of the following content and figures, or may be learned through production or application. The features of the present application can be implemented and obtained by practice or use of various aspects of the methods, tools, and combinations listed in the detailed examples described below. Attached Figure Description

[0057] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the specification, serve to illustrate the technical solutions of this disclosure.

[0058] Figure 1 This is a schematic diagram illustrating an application scenario of the industry collaborative warehousing management method provided in this embodiment of the invention.

[0059] Figure 2 This is a flowchart of an industry collaborative warehousing management method provided in an embodiment of the present invention.

[0060] Figure 3 This is a schematic diagram of the functional module architecture of the image processing device provided in an embodiment of the present invention.

[0061] Figure 4 This is a schematic diagram of the composition of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] In the following description, the terms "some embodiments," "as an implementation / method," and "in one implementation" refer to a subset of all possible embodiments. However, it is understood that "some embodiments," "as an implementation / method," and "in one implementation" can be the same subset or different subsets of all possible embodiments and can be combined with each other without conflict.

[0064] In the following description, the terms "first," "second," "third," and similar terms are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first," "second," and "third" can be interchanged in a specific order or sequence where permissible, so that the embodiments of the invention described herein can be implemented in an order other than that illustrated or described herein. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0065] The industry collaborative warehousing management method provided in this invention can be executed by electronic devices, which can be various types of terminals such as laptops, tablets, desktop computers, and mobile devices (e.g., mobile phones, portable music players, personal digital assistants, dedicated messaging devices, portable gaming devices), or implemented as servers. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0066] Figure 1 This is a schematic diagram illustrating an application scenario of the collaborative warehousing management method provided in this embodiment of the invention. The collaborative warehousing management system 10 provided in this embodiment includes multiple image acquisition devices 100, a network 200, and electronic devices 300. The multiple image acquisition devices 100 and electronic devices 300 are connected via the network 200. The electronic devices 300 are used to execute the method provided in this embodiment of the invention.

[0067] In related technologies, two different deep neural networks are typically used to remove internal defects and external disturbances, respectively. The removal order may be internal defects first, followed by external disturbances, or vice versa. For example, images acquired by a camera are divided into different image sets. Based on a primary cargo image processing network component, each camera obtains a cleaned cargo image. These cleaned cargo images are then processed together and combined using a secondary cargo image processing network component to obtain the final result. This two-stage cargo image defect and disturbance removal process often relies heavily on computational overhead, placing excessive demands on hardware and cost in practical applications. However, reducing the number of parameters to cut costs can weaken image processing performance. Based on this, the present invention proposes a novel industrial collaborative warehousing management method. It obtains the initial image feature descriptor of the target cargo image and loads it into a pre-calibrated cargo image processing network to obtain the target image feature descriptor output by the cargo image processing network. This cargo image processing network is obtained through sequential calibration using a multi-layer clustering cost function and a masking prediction cost function, integrating the two networks to form a new network, thus reducing the computational overhead of the network calibration step. Then, based on the target image feature descriptor, a target cargo image cleaned of intrinsic defects and external disturbances is determined. In this way, network calibration of the deployed cargo image processing network based on different cost functions prompts the network to clean up intrinsic defects and external disturbances in the initial image feature descriptor, not only alleviating the network's computational overhead but also enhancing its image processing capabilities.

[0068] Specifically, embodiments of the present invention provide an industry-coordinated warehousing management method, which is applied to electronic devices, such as... Figure 2 As shown, the method includes:

[0069] Step S101: Obtain the initial image feature descriptor of the target cargo image.

[0070] In this embodiment of the invention, an initial image feature descriptor of the target cargo image to be removed is obtained. The initial image feature descriptor is an image feature obtained from the target cargo image, essentially a feature vector. Methods for obtaining the initial image feature descriptor include, for example, obtaining a color histogram based on the number of pixels of different colors in the cargo image; or obtaining a Histogram of Oriented Gradients (HOG) describing texture and shape based on the gradient direction and intensity in the cargo image; or obtaining a Scale Invariant Feature Transform (SIFT) based on the scale direction of local feature points in the image; or obtaining a high-level feature representation of the image extracted by a convolutional neural network. Of course, multiple methods can be combined, and no specific limitation is imposed.

[0071] Step S102: Load the initial image feature descriptor into the pre-calibrated cargo image processing network to obtain the target image feature descriptor output by the cargo image processing network.

[0072] In some cases, images of target goods captured by imaging devices may simultaneously possess both intrinsic defects and extrinsic disturbances. Intrinsic defects are inherent flaws in the goods themselves, such as incomplete imaging due to dirt or damage. Extrinsic disturbances, on the other hand, are image content disturbances caused by external factors such as sensor noise, electromagnetic interference, and signal transmission interference. Because the neural networks used for both intrinsic defect removal and extrinsic disturbance removal involve a large number of internal network configuration variables (such as various weights and biases) during calibration, this results in significant computational overhead. Conversely, simplifying parameters can diminish the effectiveness of both processes. Therefore, this invention proposes integrating the two networks into a single network to effectively reduce the number of internal configuration variables, alleviate computational overhead during network calibration, and simultaneously ensure the network's effectiveness in removing image defects and disturbances.

[0073] In this embodiment of the invention, the cargo image processing network can generate a target image feature descriptor corresponding to the target cargo image based on the loaded initial image feature descriptor. This is essentially a noise-free image feature descriptor obtained after removing image defect perturbations by eliminating both intrinsic and extrinsic perturbations. As an example of the composition of a cargo image processing network, it may include an intermediate network module (i.e., the intermediate hidden network layer of a neural network), a multi-layer clustering module (i.e., a deep clustering network layer), an image masking prediction module, and an intrinsic defect masking prediction module. The multi-layer clustering module, image masking prediction module, and intrinsic defect masking prediction module can be dense network layers. Their execution data all originate from the output of the intermediate network module. The intermediate network module can determine a depth descriptor based on the loaded initial image feature descriptor. This depth descriptor is an intermediate result generated during image defect perturbation removal. The image masking prediction module can perform masking prediction using deep descriptors (i.e., perform pixel-level semantic segmentation of objects in the cargo image, generate boundary masks for each object, and complete the semantic segmentation of the objects, also known as mask prediction), to obtain target image feature descriptors cleared of intrinsic defects and external disturbances. The intrinsic defect masking prediction module can also perform masking prediction using deep descriptors to obtain image feature descriptors containing intrinsic defects. The multi-layer clustering module performs multi-layer clustering based on the obtained deep descriptors, that is, learns the hidden representation of the image data and groups it into different classification clusters. Specifically, it learns the high-level feature representation of the image based on a deep neural network and combines it with a clustering algorithm for accurate clustering, so as to complete the grouping under unsupervised learning. This can help discover the hidden patterns and structures of the image and assist the image masking prediction module and the intrinsic defect masking prediction module in clearing intrinsic defects and external disturbances. The intermediate network module can be a convolutional neural network.

[0074] Optionally, during network calibration, the network can be calibrated sequentially based on the multi-level clustering cost function corresponding to the multi-level clustering module and the masking prediction cost functions corresponding to the image masking prediction module and the intrinsic defect masking prediction module, respectively. For example, firstly, the intrinsic defect removal network can be calibrated using the multi-level clustering cost function and the masking prediction cost function. If the intrinsic defect removal network converges, the calibration ends. Here, the masking prediction cost function corresponding to the image masking prediction module uses noiseless cargo image annotation information that does not contain intrinsic defects but contains external disturbances. Then, the extrinsic disturbance removal network is calibrated. The previously calibrated intrinsic defect removal network is determined as the extrinsic disturbance removal network, and the extrinsic disturbance removal network is calibrated using the multi-level clustering cost function and the masking prediction cost function. The calibration ends when the extrinsic disturbance removal network converges. Here, the masking prediction cost function corresponding to the image masking prediction module uses noiseless cargo image annotation information that does not contain intrinsic defects or external disturbances. Thus, the final extrinsic disturbance removal network, in other words, the cargo image processing network has the performance to simultaneously perform intrinsic defect removal and extrinsic disturbance removal. In this embodiment of the invention, each annotation information can be implemented using a label.

[0075] It's important to understand that the multi-layer clustering module of the cargo image processing network is a binary classification cost based on image data clustering. Because the multi-layer clustering cost exhibits intra-class compactness and inter-class separation, it better enables the image masking prediction module and the intrinsic defect masking prediction module to remove intrinsic defects and extrinsic disturbances in the image during calibration, thus effectively enhancing the network's image defect and disturbance removal performance. Sequential calibration ensures that both intrinsic defect removal and extrinsic disturbance removal tasks achieve optimal calibration performance in their respective calibration stages, facilitating the enhancement of the cargo image processing network's performance in these areas. Based on this, the cargo image processing network obtained through the above calibration can obtain depth descriptors based on multi-layer CNNs. Then, the image masking prediction module can use the depth descriptors to perform masking prediction to determine the binary image (mask) of the cargo image, which is the target image feature descriptor.

[0076] Step S103: Based on the target image feature descriptor, determine the target cargo image after removing self-caused defects and external disturbances.

[0077] Optionally, image reconstruction is performed on the acquired target image feature descriptors to determine the target cargo image cleared of intrinsic defects and extrinsic disturbances. For example, the target image feature descriptors are fused or combined with the descriptors of the original cargo image. This process can be performed through simple weighted summation, stitching, stacking, etc. Then, the fused descriptors are used for reconstruction to generate a denoised image. Various methods can be used for reconstruction, such as deconvolution, generative adversarial networks, etc., without specific limitations.

[0078] In this embodiment of the invention, an initial image feature descriptor of the target cargo image is obtained. This initial image feature descriptor is then loaded into a pre-calibrated cargo image processing network to obtain the target image feature descriptor output by the cargo image processing network. This cargo image processing network is obtained through sequential calibration using a multi-layer clustering cost function and a masking prediction cost function. Then, based on the target image feature descriptor, a target cargo image cleaned of intrinsic defects and extrinsic disturbances is determined. Therefore, by calibrating the pre-deployed cargo image processing network using different cost functions, the network is prompted to clean up intrinsic defects and extrinsic disturbances in the initial image feature descriptor, which not only alleviates the network's computational overhead but also enhances its image processing capabilities.

[0079] The following describes the calibration process of the cargo image processing network, specifically including the following steps:

[0080] Step T101: Obtain a set of cargo image learning examples.

[0081] It is important to understand that the calibration of the preset image processing network components can be performed in advance based on the acquired cargo image learning sample set. Subsequently, each time the initial image feature descriptor of the target cargo image is cleared of image defects and disturbances, the calibrated cargo image processing network is used to determine the target image feature descriptor cleared of self-caused defects and external disturbances, without having to calibrate the preset image processing network components again each time image defect disturbances are cleared.

[0082] Optionally, the process of acquiring a training sample set of cargo images may include the following operations:

[0083] T1011, acquire the first cargo image learning example.

[0084] T1012, image feature descriptor mining is performed on the first cargo image learning example to obtain the self-causing defect image feature descriptor.

[0085] T1013, Obtain the second cargo image for learning.

[0086] T1014, perform image feature descriptor mining on the second cargo image learning example to obtain the annotation information of the first noiseless cargo image and the annotation information of the second noiseless cargo image.

[0087] T1015, based on the first cargo image learning example and the second cargo image learning example, determines multi-layer clustering labels.

[0088] The first cargo image learning example is a cargo image with inherent defects and external disturbances captured by the imaging device. The second cargo image learning example is a noiseless cargo image that does not contain inherent defects but contains external disturbances, and a noiseless cargo image that does not contain inherent defects or external disturbances. The multi-layer clustering label is the feature ratio of the first cargo image learning example and the second cargo image learning example at each pixel.

[0089] For example, images of target cargo with inherent defects and external disturbances can be captured directly using the imaging equipment.

[0090] Alternatively, noiseless cargo images (i.e., cargo images without defects and disturbances) can be obtained from the intrinsic defect removal calibration sample library as second cargo image learning examples. To sequentially calibrate the preset image processing network components, noiseless cargo images containing external disturbances but not intrinsic defects can be obtained, as well as noiseless cargo images containing neither intrinsic defects nor external disturbances. Then, image feature descriptor mining is performed on the noiseless cargo images containing external disturbances but not intrinsic defects to obtain first noiseless cargo image annotation information, and image feature descriptor mining is performed on the noiseless cargo images containing neither intrinsic defects nor external disturbances to obtain second noiseless cargo image annotation information. Specifically, the intrinsic defect image annotation information, the first noiseless cargo image annotation information, and the second noiseless cargo image annotation information can be represented as embedded feature vectors.

[0091] Optionally, the multi-layer clustering label can be determined based on the comparison of the pixel values ​​of the first cargo image learning example and the second cargo image learning example at each pixel. For example, the ratio of the pixel values ​​of the cargo image that does not contain self-caused defects but contains external disturbances and the cargo image with self-caused defects can be determined as the multi-layer clustering label, or the ratio of the pixel values ​​of the cargo image that does not contain self-caused defects and does not contain external disturbances and the cargo image with self-caused defects can be determined as the multi-layer clustering label. The multi-layer clustering label is used to determine the multi-layer clustering cost function.

[0092] Step T102: Obtain the preset image processing network component.

[0093] Because the time requirements for goods entering and leaving the warehouse are high in the warehousing process, the efficiency of image defect and disturbance removal needs to be improved. Therefore, the internal configuration variables of the goods image processing network cannot be too numerous, otherwise too much time will be spent on computation. Conversely, reducing the number of internal configuration variables will lead to insufficient image defect and disturbance removal performance. Therefore, this embodiment of the invention integrates two networks, enabling the goods image processing network to simultaneously perform internal defect removal and external disturbance removal, while maintaining the image defect and disturbance removal capability without reducing the number of internal configuration variables.

[0094] Optionally, in this embodiment of the invention, the preset image processing network component may include an intermediate network module, a multi-layer clustering module, and a masking prediction module. The preset image processing network component is a model that shares lower-level parameters for multi-dimensional output. The multi-layer clustering module assists the image masking prediction module and the self-causing defect masking prediction module in performing masking prediction, enabling them to accurately identify self-causing defects and external disturbances in the cargo image during network calibration. The intermediate network module can be a convolutional neural network, such as a residual neural network. The masking prediction module includes a cargo image binary image unit (i.e., a mask unit) and a self-causing defect binary image unit. The image masking prediction module can determine the binary image contour of the cargo image, which can be considered a mask, i.e., noiseless cargo image annotation information. The self-causing defect masking prediction module can determine the binary image contours of self-causing defects and external disturbances, which can also be considered a mask, i.e., self-causing defect image annotation information. It should be understood that processing the cargo image only requires the binary image contour output by the image masking prediction module, which does not introduce additional image defect disturbance removal computational overhead and improves the speed of image defect disturbance removal.

[0095] Step T103: Based on the cargo image learning sample set, the preset image processing network component is sequentially adjusted by removing self-cause defects and removing external disturbances until the preset image processing network component meets the set convergence requirements, and the adjusted target image processing network component is obtained and determined as the cargo image processing network.

[0096] Because the calibrated target image processing network component, namely the cargo image processing network, simultaneously executes two cleaning branches: internal defect removal and external disturbance removal. If these two cleaning branches are calibrated at the same time, the calibration of the preset image processing network component will be difficult to achieve the optimal calibration result. Therefore, the calibration of the two branches can be separated by sequential calibration. Specifically, embodiments of the present invention provide two sequential calibration methods, such as performing internal defect removal calibration first and then external disturbance removal calibration, or performing external disturbance removal calibration first and then internal defect removal calibration. The goal of internal defect removal calibration is to enable the network to have the performance of internal defect removal, and the goal of external disturbance removal calibration is to enable the network to have the performance of external disturbance removal. In this way, each of the two cleaning branches can achieve the optimal calibration result by adjusting the individual stages, thereby enhancing the cargo image processing network's ability to remove image defects and disturbances.

[0097] In an optional implementation, the preset image processing network components are sequentially adjusted for self-caused defects and external disturbances based on the cargo image learning sample set until the preset image processing network components meet the set convergence requirements. Specifically, this may include:

[0098] Step T01: Load the self-caused defect image feature descriptor into the intermediate network module, and generate a depth-tuned descriptor based on the intermediate network module.

[0099] Step T02: Load the depth tuning descriptor into the multi-layer clustering module, and generate clustering tuning tags based on the multi-layer clustering module.

[0100] Step T03: Load the depth calibration descriptor into the image masking prediction module, and generate a noiseless cargo image calibration descriptor based on the image masking prediction module.

[0101] Step T04: Load the depth calibration descriptor into the self-causing defect masking prediction module, and generate a self-causing defect image calibration descriptor based on the self-causing defect masking prediction module.

[0102] Step T05: Generate a target cost function based on the noiseless cargo image annotation information, the self-caused defect image annotation information, the multi-layer clustering label, the noiseless cargo image calibration descriptor, the self-caused defect image calibration descriptor, and the clustering calibration label. Then, based on the target cost function, sequentially perform self-caused defect removal calibration and external disturbance removal calibration on the preset image processing network component until the preset image processing network component meets the set convergence requirements.

[0103] The depth calibration descriptor is an intermediate result generated by the intermediate network module of the preset image processing network component. It is loaded as a universal value into the multi-layer clustering module, image masking prediction module, and self-causing defect masking prediction module to achieve shared lower-layer parameters and reduce the number of internal configuration variables in the network. The image masking prediction module and the self-causing defect masking prediction module can generate noise-free cargo image calibration descriptors and self-causing defect image calibration descriptors respectively using the depth calibration descriptor. The multi-layer clustering module can generate clustering calibration markers using the depth calibration descriptor. Optionally, a target cost function is generated based on the noise-free cargo image annotation information, self-causing defect image annotation information, multi-layer clustering markers, noise-free cargo image calibration descriptors, self-causing defect image calibration descriptors, and clustering calibration markers. Based on the target cost function, the preset image processing network component is sequentially calibrated for self-causing defect removal and external disturbance removal until the preset image processing network component meets the set convergence requirements, specifically including:

[0104] Step T051: Determine the first cost function based on the clustering calibration mark and the multi-layer clustering mark.

[0105] The first cost function is the multi-level clustering cost function, for example, the first cost function C d (m d ,m d '), m d For clustering and calibration markers, m d ' is a multi-level cluster marker.

[0106] Step T052: Determine the second cost function based on the noiseless cargo image calibration descriptor and the noiseless cargo image annotation information.

[0107] For the two sequential calibration methods, two different second cost functions can be determined based on different noiseless cargo image annotation information. For example, the calibration descriptor m can be based on the noiseless cargo image. n And the first noise-free cargo image annotation information m nl Determine the cost function C for eliminating self-cause defects. c (m n ,m nl The self-causing defect elimination cost function is used as the second cost function C. c (m n ,m nl ).

[0108] For example, the descriptor m can be adjusted based on a noiseless cargo image. n Second noiseless cargo image annotation information m n2 Determine the cost function C for self-causing defect removal. c (m n ,m n2 The self-causing defect removal cost function is used as the second cost function Cc(m). n ,m n2 ).

[0109] Step T053: Determine the third cost function based on the self-caused defect image calibration descriptor and self-caused defect image annotation information.

[0110] For example, the third cost function C b (m b ,m b '), where m b For self-causing defect image calibration descriptor, m b 'This provides annotation information for the self-causing defect image. The second cost function C...' c and the third cost function C b (m b ,m b ') is the masked prediction cost function.

[0111] Step T054: Based on the first cost function, the second cost function, and the third cost function, a target cost function for the preset image processing network component is generated. Based on the target cost function, the preset image processing network component is sequentially adjusted to remove self-cause defects and remove external disturbances until the preset image processing network component meets the set convergence requirements.

[0112] For example, based on the first cost function C dThe second cost function C c and the third cost function C b The target cost function C for generating the preset image processing network components can be implemented using the following formula:

[0113] C = w1·C d +w2·C c +w3·C b

[0114] Where w1, w2, and w3 are weights, the objective cost function C sequentially performs self-cause defect removal and external disturbance removal on the preset image processing network components until the preset image processing network components meet the set convergence requirements.

[0115] Because the requirements for image accuracy, i.e., the completeness and clarity of the displayed information, vary in different cargo image usage scenarios. For example, for simple size measurement and contour detection tasks, the accuracy requirement is lower, and the focus can be on eliminating external disturbances. For complex information recognition and surface integrity detection tasks, the accuracy requirement is higher, and the focus can be on eliminating internal defects. In an optional implementation, image accuracy constraints can be obtained based on the application requirements of the cargo image processing network, and a corresponding sequential calibration method can be determined based on the image accuracy constraints. Then, through the sequential calibration method, a target cost function for the preset image processing network component is generated based on the first cost function, the second cost function, and the third cost function. Based on the target cost function, the preset image processing network component is sequentially calibrated to eliminate internal defects and external disturbances until the preset image processing network component meets the set convergence requirements. Here, the image accuracy constraints are used to indicate the task type corresponding to the cargo image processing network. The sequential calibration methods include a first sequential calibration method and a second sequential calibration method. The first sequential calibration method is used to prioritize the elimination of internal defects, first performing calibration to eliminate self-caused defects, and then performing calibration to eliminate external disturbances. The second sequential calibration method is used to prioritize the elimination of external disturbances, first performing calibration to eliminate external disturbances, and then performing calibration to eliminate self-caused defects.

[0116] Optionally, when focusing on eliminating intrinsic defects, intrinsic defect elimination calibration is performed first, followed by extrinsic disturbance elimination calibration. Using a first sequential calibration method, a target cost function for a preset image processing network component is determined based on a first cost function, a second cost function, and a third cost function. The second cost function is obtained by determining the intrinsic defect elimination cost function. Then, the preset image processing network component is repeatedly calibrated for intrinsic defect elimination based on the target cost function until it meets the set convergence requirements, resulting in an intrinsic defect elimination network component. This intrinsic defect elimination network component only provides intrinsic defect elimination. Then, based on the first, second, and third cost functions, a target cost function for the intrinsic defect elimination network component is determined. The second cost function is obtained by determining the extrinsic disturbance elimination cost function. Furthermore, the intrinsic defect elimination network component is repeatedly calibrated for extrinsic disturbance elimination based on the target cost function until it meets the set convergence requirements. In this way, performing independent self-cause defect removal and calibration first can prevent the calibration process from being affected by external disturbances, so that the resulting target image processing network component has better self-cause defect removal capabilities.

[0117] Optionally, when the focus is on eliminating external disturbances, external disturbance elimination calibration is performed first, followed by self-causing defect elimination calibration. Using a second sequential calibration method, a target cost function for the preset image processing network component is determined based on the first, second, and third cost functions. The second cost function is obtained by determining the external disturbance elimination cost function. Then, based on the target cost function, the preset image processing network component is repeatedly calibrated to eliminate external disturbances until it meets the set convergence requirements, resulting in an external disturbance elimination network component. This external disturbance elimination network component only provides external disturbance elimination. Then, based on the first cost function, the second cost function, and the third cost function, the target cost function of the external disturbance removal network component is determined. The second cost function is obtained by determining the self-cause defect removal cost function. Then, based on the target cost function, the external disturbance removal network component is repeatedly adjusted for self-cause defect removal until the external disturbance removal network component meets the set convergence requirements. In this way, performing independent external disturbance removal adjustment first can prevent the adjustment process from being affected by self-cause defects, so that the obtained target image processing network component has better external disturbance removal capabilities.

[0118] The convergence requirements mentioned above can be that the number of adjustments has reached the maximum, the network cost no longer changes, or it is less than the preset cost value; there are no specific limitations.

[0119] The calibration of the preset image processing network components based on multi-task calibration adopts multi-layer clustering cost and masking prediction cost for joint calibration. However, the masking prediction cost is only used in the verification stage of the target image processing network component, namely the cargo image processing network. In the application of the cargo image processing network, the output result of the masking prediction branch is used as the mask after the image defect disturbance is cleared, which is the target image feature descriptor.

[0120] Step T104: Obtain the initial image feature descriptor of the target cargo image.

[0121] Step T105: Load the initial image feature descriptor into the intermediate network module, and generate the depth descriptor based on the intermediate network module.

[0122] Step T106: Load the depth descriptor into the image masking prediction module, generate a noiseless image feature descriptor based on the image masking prediction module, and use the noiseless image feature descriptor as the target image feature descriptor.

[0123] Optionally, after capturing an image of the target cargo, image feature descriptor mining is performed on the target cargo image to obtain an initial image feature descriptor. The initial image feature descriptor is then loaded into the intermediate network module of the cargo image processing network component. A depth descriptor is generated based on the intermediate network module. The depth descriptor can then be loaded into the image masking prediction module. A noiseless image feature descriptor is generated based on the image masking prediction module, and the noiseless image feature descriptor is used as the target image feature descriptor.

[0124] Step T107: Perform image reconstruction on the target image feature descriptor to obtain the target cargo image after removing self-caused defects and external disturbances.

[0125] Optionally, after obtaining the target image feature descriptor, image reconstruction can be performed on the target image feature descriptor to obtain the target cargo image. In this embodiment of the invention, a cargo image learning sample set and a preset image processing network component are obtained. Based on the cargo image learning sample set, the preset image processing network component is sequentially adjusted for self-cause defect removal and external disturbance removal until the preset image processing network component meets the set convergence requirements, resulting in the adjusted target image processing network component, which is then determined as the cargo image processing network. The initial image feature descriptor is then loaded into an intermediate network module, a depth descriptor is generated based on the intermediate network module, the depth descriptor is loaded into an image masking prediction module, and a noiseless image feature descriptor is generated based on the image masking prediction module. The noiseless image feature descriptor is used as the target image feature descriptor, and image reconstruction is performed on the target image feature descriptor to determine the target cargo image cleared of self-cause defects and external disturbances. In summary, only the target image feature descriptor output by the image masking prediction module of the cargo image processing network is used for cargo image clearing processing, eliminating the computational overhead of adding an image defect disturbance removal step, thus increasing the speed of image defect disturbance removal.

[0126] Based on the above embodiments, this invention provides an image processing apparatus. Figure 3 This is an image processing device 340 provided in an embodiment of the present invention, such as... Figure 3 As shown, the device 340 includes:

[0127] Feature mining module 341 is used to obtain initial image feature descriptors of the target cargo image;

[0128] Feature processing module 342 is used to load the initial image feature descriptor into a pre-calibrated cargo image processing network to obtain the target image feature descriptor output by the cargo image processing network. The cargo image processing network is obtained by sequential calibration through a multi-layer clustering cost function and a masking prediction cost function.

[0129] Image processing module 343 is used to determine, based on the target image feature descriptor, a target cargo image that has been cleared of self-caused defects and external disturbances;

[0130] Network calibration module 344 is used to calibrate the cargo image processing network based on the following operations:

[0131] Obtain a cargo image learning sample set, which includes self-causing defect image feature descriptors, noiseless cargo image annotation information, self-causing defect image annotation information, and multi-layer clustering labels;

[0132] Obtain a preset image processing network component, which includes an intermediate network module, a multi-layer clustering module, and a masking prediction module;

[0133] Based on the cargo image learning sample set, the preset image processing network components are sequentially adjusted by removing self-cause defects and external disturbances until the preset image processing network components meet the set convergence requirements, and the adjusted target image processing network components are obtained and determined as the cargo image processing network.

[0134] The description of the above device embodiments is similar to that of the above method embodiments, and has similar beneficial effects. For technical details not disclosed in the device embodiments of the present invention, please refer to the description of the method embodiments of the present invention for understanding.

[0135] If the technical solution of this invention involves personal or private information, the product using this invention has clearly informed the user of the personal information processing rules and obtained the user's voluntary consent before processing the personal information. If the technical solution of this invention involves sensitive personal information, the product using this invention has obtained the user's separate consent before processing the sensitive personal information, and simultaneously meets the requirement of "express consent," while collecting information within the scope of laws and regulations. For example, at personal information collection devices such as cameras, clear and prominent signs are set up to inform users that they have entered the scope of personal information collection and that personal information will be collected. If an individual voluntarily enters the collection scope, it is deemed that they have agreed to the collection of their personal information; or on personal information processing devices, while using clear signs / information to inform users of the personal information processing rules, authorization is obtained through pop-up messages or by asking users to upload their personal information themselves; wherein, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the types of personal information processed.

[0136] It should be noted that, in the embodiments of the present invention, if the above methods are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of the present invention, or the part that contributes to related technologies, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of the present invention are not limited to any specific hardware and software combination.

[0137] This invention provides an electronic device, including a memory and a processor. The memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the above-described method.

[0138] This invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method. The computer-readable storage medium can be transient or non-transient.

[0139] This invention provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program. When the computer program is read and executed by a computer, it implements some or all of the steps in the above-described method. This computer program product can be implemented specifically through hardware, software, or a combination thereof. In one optional embodiment, the computer program product is specifically embodied as a computer storage medium; in another optional embodiment, the computer program product is specifically embodied as a software product, such as a software development kit (SDK), etc.

[0140] It should be noted that, Figure 4 This is a schematic diagram of a hardware entity of XXX provided in an embodiment of the present invention, such as... Figure 4 As shown, the hardware entity of the electronic device 300 includes a processor 310, a communication interface 320, and a memory 330. The processor 310 typically controls the overall operation of the electronic device 300. The communication interface 320 enables the electronic device to communicate with other terminals or servers via a network. The memory 330 is configured to store instructions and applications executable by the processor 310, and can also cache data to be processed or already processed by the processor 310 and various modules in the electronic device 300 (e.g., image data, audio data, voice communication data, and video communication data). It can be implemented using flash memory (FLASH) or random access memory (RAM). Data transmission between the processor 310, the communication interface 320, and the memory 330 can be performed via a bus 340. It should be noted that the description of the above storage medium and device embodiments is similar to the description of the above method embodiments, and has similar beneficial effects. For technical details not disclosed in the storage medium and device embodiments of the present invention, please refer to the description of the method embodiments of the present invention for understanding.

[0141] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of the invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of the invention, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the invention. The sequence numbers of the above-described embodiments of the invention are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0142] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0143] In the several embodiments provided by this invention, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0144] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0145] In addition, in the embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0146] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0147] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0148] The above description is merely an embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. An industry-wide collaborative warehousing management method, characterized in that, Applied to electronic devices, the method includes: Obtain the initial image feature descriptor for the target cargo image; The initial image feature descriptor is loaded into a pre-calibrated cargo image processing network to obtain the target image feature descriptor output by the cargo image processing network. The cargo image processing network is obtained by sequential calibration through a multi-layer clustering cost function and a masking prediction cost function. Based on the target image element descriptor, a target cargo image with self-caused defects and external disturbances removed is obtained; The cargo image processing network is calibrated based on the following operations: The process of acquiring a cargo image learning sample set specifically includes: acquiring a first cargo image learning sample, which is a cargo image captured by a camera device and contains inherent defects and external disturbances, wherein the inherent defects are defects inherent to the cargo itself, and the external disturbances are image content disturbances caused by external factors; performing image element descriptor mining on the first cargo image learning sample to obtain image element descriptors for inherent defects; and acquiring a second cargo image learning sample, which includes noise-free cargo images that do not contain inherent defects but contain external disturbances, as well as images that do not contain inherent defects. The first cargo image learning example set contains a noiseless cargo image with inherent defects but no external disturbances. Image feature descriptor mining is performed on the second cargo image learning example to obtain annotation information for the first and second noiseless cargo images. Based on the first and second cargo image learning examples, multi-level clustering labels are determined. The cargo image learning example set includes inherent defect image feature descriptors, noiseless cargo image annotation information, inherent defect image annotation information, and multi-level clustering labels. The inherent defect image annotation information is a binary image contour of both inherent defects and external disturbances. Obtain a preset image processing network component, which includes an intermediate network module, a multi-layer clustering module, and a masking prediction module; Based on the cargo image learning sample set, the preset image processing network component is sequentially adjusted by removing self-cause defects and removing external disturbances until the preset image processing network component meets the set convergence requirements, and the adjusted target image processing network component is obtained and determined as the cargo image processing network. The shielding prediction module includes an image shielding prediction module and a self-causing defect shielding prediction module. The step of sequentially performing self-causing defect removal and external disturbance removal adjustments on the preset image processing network components based on the cargo image learning sample set, until the preset image processing network components meet the set convergence requirements, includes: The self-causing defect image feature descriptor is loaded into the intermediate network module, and a depth-tuned descriptor is generated based on the intermediate network module; The depth calibration descriptor is loaded into the multi-layer clustering module, and clustering calibration tags are generated based on the multi-layer clustering module; The depth calibration descriptor is loaded into the image masking prediction module, and a noiseless cargo image calibration descriptor is generated based on the image masking prediction module. The depth calibration descriptor is loaded into the self-causing defect masking prediction module, and a self-causing defect image calibration descriptor is generated based on the self-causing defect masking prediction module. Based on the noiseless cargo image annotation information, the self-caused defect image annotation information, the multi-layer clustering marker, the noiseless cargo image calibration descriptor, the self-caused defect image calibration descriptor, and the clustering calibration marker, a target cost function is generated. Based on the target cost function, the preset image processing network component is sequentially calibrated for self-caused defect removal and external disturbance removal until the preset image processing network component meets the set convergence requirements.

2. The method according to claim 1, characterized in that, The process involves generating a target cost function based on the noiseless cargo image annotation information, the self-causing defect image annotation information, the multi-layer clustering marker, the noiseless cargo image calibration descriptor, the self-causing defect image calibration descriptor, and the clustering calibration marker. Then, based on the target cost function, the preset image processing network component is sequentially calibrated for self-causing defect removal and external disturbance removal until the preset image processing network component meets the set convergence requirements. This includes: Based on the clustering calibration mark and the multi-layer clustering mark, a first cost function is determined; Based on the noiseless cargo image calibration descriptor and the noiseless cargo image annotation information, a second cost function is determined; Based on the self-causing defect image calibration descriptor and the self-causing defect image annotation information, a third cost function is determined; Based on the first cost function, the second cost function, and the third cost function, a target cost function for the preset image processing network component is generated. Based on the target cost function, the preset image processing network component is sequentially adjusted to remove self-cause defects and external disturbances until the preset image processing network component meets the set convergence requirements.

3. The method according to claim 2, characterized in that, The noiseless cargo image annotation information is the first noiseless cargo image annotation information. The step of determining the second cost function based on the noiseless cargo image calibration descriptor and the noiseless cargo image annotation information includes: Based on the noiseless cargo image calibration descriptor and the first noiseless cargo image annotation information, the self-causing defect removal cost function is determined. The self-causing defect removal cost function is used as the second cost function, and the first noiseless cargo image annotation information is the image annotation information obtained by cargo images that do not contain self-causing defects but contain external disturbances. or The noiseless cargo image annotation information is the second noiseless cargo image annotation information. The step of determining the second cost function based on the noiseless cargo image calibration descriptor and the noiseless cargo image annotation information includes: Based on the noiseless cargo image calibration descriptor and the second noiseless cargo image annotation information, the external disturbance removal cost function is determined. The external disturbance removal cost function is used as the second cost function, and the second noise-free cargo image annotation information is the image annotation information obtained by cargo images that do not contain self-caused defects and do not contain external disturbances.

4. The method according to claim 3, characterized in that, Based on the first cost function, the second cost function, and the third cost function, a target cost function for the preset image processing network component is generated. Then, based on the target cost function, the preset image processing network component undergoes sequential self-cause defect removal and external disturbance removal adjustments until the preset image processing network component meets the set convergence requirements, including: Obtain image accuracy constraints; The corresponding sequential calibration method is determined based on the image accuracy constraints. Using the sequential calibration method, a target cost function for the preset image processing network component is generated based on the first cost function, the second cost function, and the third cost function. Based on the target cost function, the preset image processing network component is sequentially calibrated to remove self-cause defects and external disturbances until the preset image processing network component meets the set convergence requirements.

5. The method according to claim 4, characterized in that, The sequential calibration method includes a first sequential calibration method, wherein, based on the first cost function, the second cost function, and the third cost function, a target cost function for the preset image processing network component is generated, and the preset image processing network component is sequentially calibrated by removing intrinsic defects and removing extrinsic disturbances based on the target cost function until the preset image processing network component meets the set convergence requirements, including: When the sequential calibration method is the first sequential calibration method, the target cost function of the preset image processing network component is determined based on the first cost function, the second cost function and the third cost function, and the preset image processing network component is repeatedly calibrated to remove self-cause defects based on the target cost function until the preset image processing network component meets the set convergence requirements, thereby obtaining the self-cause defect removal network component, wherein the second cost function is the self-cause defect removal cost function; Based on the first cost function, the second cost function, and the third cost function, a target cost function for the self-cause defect removal network component is determined, and the self-cause defect removal network component is repeatedly adjusted to remove external disturbances based on the target cost function until the self-cause defect removal network component meets the set convergence requirements, wherein the second cost function is the external disturbance removal cost function.

6. The method according to claim 4, characterized in that, The sequential calibration method includes a second sequential calibration method. Through this sequential calibration method, a target cost function for the preset image processing network component is generated based on the first cost function, the second cost function, and the third cost function. Based on the target cost function, the preset image processing network component is sequentially calibrated to remove intrinsic defects and external disturbances until the preset image processing network component meets the set convergence requirements. This includes: When the sequential calibration method is the second sequential calibration method, the target cost function of the preset image processing network component is determined based on the first cost function, the second cost function, and the third cost function. The preset image processing network component is repeatedly calibrated to remove external disturbances based on the target cost function until the preset image processing network component meets the set convergence requirements, thereby obtaining the external disturbance removal network component. The second cost function is the external disturbance removal cost function. Based on the first cost function, the second cost function, and the third cost function, a target cost function for the external disturbance removal network component is determined, and the external disturbance removal network component is repeatedly adjusted to remove self-cause defects based on the target cost function until the external disturbance removal network component meets the set convergence requirements. The second cost function is the self-cause defect removal cost function.

7. The method according to claim 1, characterized in that, The cargo image processing network includes an intermediate network module, a multi-layer clustering module, an image masking prediction module, and an autogenic defect masking prediction module. The step of loading the initial image feature descriptor into the pre-calibrated cargo image processing network to obtain the target image feature descriptor output by the cargo image processing network includes: The initial image feature descriptor is loaded into the intermediate network module, and a depth descriptor is generated based on the intermediate network module; The depth descriptor is loaded into the image masking prediction module, a noiseless image feature descriptor is generated based on the image masking prediction module, and the noiseless image feature descriptor is used as the target image feature descriptor. The step of determining the target cargo image by removing intrinsic defects and extrinsic disturbances based on the target image feature descriptor includes: Image reconstruction is performed on the target image feature descriptor to obtain a target cargo image that has been cleared of self-caused defects and external disturbances.

8. An industry collaborative warehousing management system, characterized in that, The method includes an image acquisition device and an electronic device, the image acquisition device and the electronic device being communicatively connected, the image acquisition device being used to acquire images of goods and send them to the electronic device, the electronic device including a memory and a processor, the memory storing a computer program that can run on the processor, and the processor executing the computer program to implement the method according to any one of claims 1 to 7.

Images