RFID luggage tag high-speed reading compensation method and system based on luggage re-identification

Abstract

The invention discloses an RFID luggage tag high-speed reading compensation method and system based on luggage re-identification, and relates to the technical field of luggage re-identification, and the method comprises the steps: carrying out the image feature extraction through employing a ResNet network as a feature extraction backbone network in a luggage detection process; according to the method, global feature enhancement and local feature enhancement are introduced, after extracted image features are processed, multi-granularity information aggregation is executed, model parameters are updated through lost back propagation so as to construct an identification model, and the identification model is applied to an airport luggage sorting system so as to realize automatic sorting and matching of luggage. According to the method, the luggage features are remarkably enhanced, so that the efficiency and accuracy of intelligent luggage searching are improved.

Description

Technical Field

[0001] This invention relates to the field of baggage weight recognition technology, and more specifically, to a high-speed reading compensation method and system for RFID baggage tags based on baggage weight recognition. Background Technology

[0002] In hub airport baggage handling systems, the ability to quickly locate lost baggage tags is a crucial factor affecting passenger baggage punctuality. Currently, baggage information recording and tracking primarily rely on RFID chips within baggage tags. However, effectively tracking and locating baggage when tags are lost is a pressing issue. In the field of computer vision, recovering lost targets is mainly achieved through re-identification (Re-ID) technology. Re-ID relies on target features and uses cross-camera, cross-scene image or video retrieval to determine whether targets from different perspectives share the same identity. Traditional systems are limited by single-camera perspectives, making cross-area target tracking difficult. Re-identification technology overcomes this limitation by extracting the target's depth features.

[0003] As required, the CCTV surveillance system must provide full coverage of the entire baggage handling process. Therefore, baggage image acquisition becomes exceptionally convenient, facilitating the tracking, location, and retrieval of baggage throughout the process. With the development of deep learning, re-identification has shifted from matching methods based on manual features to end-to-end learning based on deep neural networks, significantly improving recognition performance in complex environments. Re-identification tasks are typically divided into different areas such as pedestrian re-identification (Person Re-ID), vehicle re-identification (Vehicle Re-ID), and object re-identification (Object Re-ID). The core of re-identification lies in feature extraction and metric learning. Through deep learning models such as Convolutional Neural Networks (CNN) and Generative Adversarial Networks (GAN), the system can extract global features (such as overall shape and color distribution) and local features (such as facial details and clothing texture) from images, and utilize metric learning algorithms such as Triplet Loss and Contrastive Loss to optimize the feature space, making the feature distances of the same target closer and the feature distances of different targets farther apart.

[0004] Because baggage re-identification tasks differ from other re-identification tasks in many ways—for example, baggage re-identification faces a series of challenges such as transport speeds of up to 10 meters per second, baggage sticking together and stacking, and complex working environments—existing re-identification methods struggle to address each of these challenges. Summary of the Invention

[0005] To address the aforementioned issues, the present invention aims to provide a high-speed reading compensation technology for RFID baggage tags based on baggage re-identification, which aims to enhance the efficiency and accuracy of intelligent baggage retrieval by strengthening baggage features.

[0006] To achieve the above technical objectives, this application provides a high-speed reading compensation method for RFID baggage tags based on baggage re-identification, comprising the following steps:

[0007] During baggage inspection, ResNet network is used as the backbone network for feature extraction to perform image feature extraction;

[0008] Global and local feature enhancements are introduced to process the extracted image features. Multi-granularity information aggregation is then performed, and the model parameters are updated through backpropagation of the loss to build a recognition model. This model is then applied to an airport baggage sorting system to achieve automatic baggage sorting and matching.

[0009] Preferably, when introducing global feature enhancement, a global attention feature enhancement module is introduced after the backbone network and before multi-granularity feature aggregation. The feature map output by the backbone network is used as the input of the global feature enhancement module. After the image is input into the module, the number of channels of the feature map is halved while keeping the size unchanged.

[0010] Preferably, when introducing global feature enhancement, the feature map is mapped to different spaces through three different linear mapping functions, resulting in three different feature vectors. Then, attention is calculated, and the enhanced features are rearranged and convolutionally processed to restore the enhanced feature vectors to resemble the input features. Figure 1 The same size.

[0011] Preferably, when introducing local feature enhancement, the acquired luggage images are used to extract features through a backbone network. The extracted features are divided into fine-grained local features through different granularities. Then, the local attention feature enhancement module is used to enhance the features of salient regions in order to suppress background information.

[0012] Preferably, when performing multi-granularity information aggregation, information is aggregated by adding local features and global features element by element.

[0013] Preferably, when updating model parameters, cross-entropy loss and three sets of losses are used as loss functions to optimize the model, wherein cross-entropy is used to calculate identity loss.

[0014] Based on the same inventive concept, this invention also discloses a high-speed reading compensation system for RFID baggage tags based on baggage weight identification, comprising:

[0015] The feature extraction module is used to perform image feature extraction during the baggage detection process, using a ResNet network as the backbone network for feature extraction.

[0016] The baggage re-identification module introduces global and local feature enhancement. After processing the extracted image features, it performs multi-granularity information aggregation and updates the model parameters through backpropagation of the loss to build a recognition model. This model is then applied to airport baggage sorting systems to achieve automatic baggage sorting and matching.

[0017] Preferably, in the baggage re-identification module, a global attention feature enhancement module is introduced after the backbone network and before multi-granularity feature aggregation. The feature map output from the backbone network is used as the input to the global feature enhancement module. After the image is input to this module, the number of channels in the feature map is halved while maintaining its size. The feature map is then mapped to different spaces using three different linear mapping functions, resulting in three different feature vectors. Attention is then calculated, and the enhanced features are rearranged and convolved to restore the enhanced feature vectors to match the input features. Figure 1 The size of the luggage is as follows; at the same time, the collected luggage images are used to extract features through a backbone network. The extracted features are divided into fine-grained local features through different granularities. Then, the local attention feature enhancement module is used to enhance the features of salient regions to suppress background information.

[0018] Preferably, the baggage re-identification module aggregates information by adding local and global features element by element.

[0019] Preferably, the baggage re-identification module uses cross-entropy loss and three sets of losses as loss functions to optimize the model, wherein cross-entropy is used to calculate the identity loss.

[0020] The present invention discloses the following technical effects:

[0021] This invention employs a multi-granularity information aggregation method to aggregate global and local features of luggage, further enhancing the salient features of luggage. Actual user experience has demonstrated that the intelligent luggage search system can quickly locate luggage with missing tags, improving the efficiency and accuracy of luggage searches and effectively ensuring the on-time delivery rate of luggage. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a model framework diagram as described in this invention;

[0024] Figure 2 It is the global feature enhancement module described in this invention;

[0025] Figure 3 This is the local feature enhancement module described in this invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0027] like Figures 1-3 As shown, this invention provides a high-speed reading compensation technology for RFID baggage tags based on baggage weight identification, specifically including the following:

[0028] This invention designs the overall structure of a model for baggage re-identification as follows: Figure 1 As shown, this model uses ResNet as the backbone network for feature extraction. The overall process can be divided into image feature extraction, local feature enhancement, global feature enhancement, multi-granularity information aggregation, and backpropagation of loss. The local attention enhancement module is mainly used to enhance local features and suppress background information. The global attention enhancement module is mainly used to enhance global features and suppress background information. The loss function adopted in this invention is triplet loss and cross-entropy loss.

[0029] In baggage detection, the highly complex operating environment of the baggage system means that the captured images contain not only baggage but also a great deal of background information. This background information is not only useless for detection and classification but also affects the robustness of the model. Therefore, to enable the neural network to notice baggage features and suppress background information, a global attention feature enhancement module and a local attention feature enhancement module are added to the model. The captured baggage images are processed through a backbone network for feature extraction. The extracted features are divided into global features and local features, which are then enhanced by the attention feature enhancement module to improve the features of salient regions and suppress background information. Furthermore, to better utilize global and local features, multi-granularity information aggregation is added to improve the discriminative power of the features.

[0030] like Figure 2 As shown, a global attention feature enhancement module is introduced after the backbone network and before multi-granularity feature aggregation to enhance the global feature map output by the backbone network. As input to the feature enhancement module, to reduce the computational complexity of the model, the feature map is processed after the image is input into the module. The number of channels is halved while the size remains unchanged. This invention uses 1×1 convolution for the operation, and its mathematical expression is as follows:

[0031] (1)

[0032] Where conv represents 1×1 convolution, , C represents the channel of the feature map, H represents the height of the feature map, and W represents the width of the feature map.

[0033] Next, to enable the network to notice salient luggage areas while suppressing the influence of some background information, this invention uses attention enhancement to give higher weights to luggage target areas and lower weights to background areas. As shown in the figure, the feature map is first expanded using an unfolding operation. The feature map is expanded to facilitate subsequent attention calculations. The expanded feature map is represented as follows:

[0034] (2)

[0035] in, , .

[0036] For the unfolded This invention first maps the feature vectors to different spaces using three different linear mapping functions, resulting in three distinct feature vectors. Then, attention is calculated. The attention calculation method of this invention is as follows:

[0037] , , (3)

[0038] (4)

[0039] (5)

[0040] in, , , Let V represent the parameter matrices of three different linear mappings, V represent the value vector, K represent the key vector, Q represent the query vector, T represent the transposed rank of the vectors, Atten represent the attention weight graph, and D represent the attention scaling factor.

[0041] Finally, to facilitate subsequent classification and regression by the detection network, this invention performs rearrangement and convolution operations on the enhanced features, restoring the enhanced feature vectors to resemble the input features. Figure 1 The sample size. The rearrangement and convolution operations are as follows:

[0042] , , (6)

[0043] , (7)

[0044] in, Represents a 1×1 convolution, X r This represents the rearranged feature map. This indicates a rearrangement operation. , .

[0045] Due to the unique nature of baggage re-identification tasks, the samples exhibit extremely high similarity, differing only in subtle areas. To further focus on fine-grained information about the baggage and improve feature discriminative power, a local attention feature enhancement module was incorporated into the model. First, the acquired baggage images are processed through a backbone network for feature extraction. The extracted features are then divided into fine-grained local features based on different granularities. Finally, the local attention feature enhancement module enhances the features in salient regions and suppresses background information.

[0046] like Figure 3 As shown, a local attention feature enhancement module is introduced after the backbone feature extraction network and before multi-granularity information aggregation to divide the feature map of the backbone network into multiple fine-grained local feature blocks. As input to the local feature enhancement module, to reduce the computational complexity of the model, fine-grained feature maps are processed after the image is input into this module. The number of channels is halved while the size remains unchanged. This invention uses 1×1 convolution for the operation, and its mathematical expression is as follows:

[0047] (8)

[0048] in, express convolution, , k represents the number of blocks in the partition.

[0049] Next, to enable the network to notice locally salient luggage areas while suppressing the influence of some background information, this invention uses local attention enhancement to give higher weights to luggage target areas and lower weights to background areas. As shown in the figure, the feature map is first expanded using a unfolding operation. The feature map is expanded to facilitate subsequent attention calculations. The expanded feature map is represented as follows:

[0050] (9)

[0051] in, , .

[0052] For the unfolded This invention first maps the feature vectors to different spaces using three different linear mapping functions, resulting in three distinct feature vectors. Then, attention is calculated. The attention calculation method of this invention is as follows:

[0053] , , (10)

[0054] (11)

[0055] (12)

[0056] in, , , This represents the parameter matrix for three different linear mappings. D represents the attention scaling factor.

[0057] Finally, to facilitate subsequent classification and regression by the detection network, this invention performs rearrangement and convolution operations on the enhanced features, restoring the enhanced feature vectors to resemble the input features. Figure 1 The sample size. The rearrangement and convolution operations are as follows:

[0058] , , (13)

[0059] , (14)

[0060] in, express convolution, This indicates a rearrangement operation. , .

[0061] To better integrate the enhanced local features with the global features, the enhanced local features need to be concatenated together for easier subsequent operations.

[0062] (15)

[0063] in, This indicates a splicing operation. k represents the number of blocks. The concatenated local features are restored to the same size as the global features.

[0064] Next, information aggregation is performed by element-wise addition of local and global features. The aggregation operation is as follows:

[0065] (16);

[0066] This invention uses two loss functions to optimize the network: cross-entropy loss and three sets of losses. To calculate the identity loss and triplet loss, we use cross-entropy to calculate the identity loss, as follows:

[0067] , (17)

[0068] Where B represents the batch size; This represents the identity label corresponding to the feature; Represents the aggregated features The classification result after passing through the classifier.

[0069] Triple loss:

[0070] This invention selects one baggage image as the anchor point, one baggage image of the same category as the positive sample, and one baggage image of a different category as the negative sample, and vice versa.

[0071] (18)

[0072] in, Indicates Euclidean distance; , , This represents the anchor point of the luggage image, positive sample, and negative sample. The overall loss function in the network is expressed as follows:

[0073] (19)

[0074] During training, these losses are used to jointly optimize the network of the present invention, and the parameters are updated through backpropagation.

[0075] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0076] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0077] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A high-speed reading compensation method for RFID baggage tags based on baggage weight identification, characterized in that, Includes the following steps: During baggage inspection, ResNet network is used as the backbone network for feature extraction to perform image feature extraction; Global and local feature enhancements are introduced to process the extracted image features. Multi-granularity information aggregation is then performed, and the model parameters are updated through backpropagation of the loss to build a recognition model. This model is then applied to an airport baggage sorting system to achieve automatic baggage sorting and matching.

2. The high-speed reading compensation method for RFID baggage tags based on baggage re-identification according to claim 1, characterized in that: When introducing global feature enhancement, a global attention feature enhancement module is introduced after the backbone network and before multi-granularity feature aggregation. The feature map output by the backbone network is used as the input of the global feature enhancement module. After the image is input into the module, the number of channels of the feature map is halved while keeping the size unchanged.

3. The high-speed reading compensation method for RFID baggage tags based on baggage re-identification according to claim 2, characterized in that: When introducing global feature enhancement, the feature map is mapped to different spaces through three different linear mapping functions, resulting in three different feature vectors. Then, attention is calculated, and the enhanced features are rearranged and convolutionally processed to restore the enhanced feature vectors to the same size as the input feature map.

4. The high-speed reading compensation method for RFID baggage tags based on baggage re-identification according to claim 3, characterized in that: When introducing local feature enhancement, the collected luggage images are processed through a backbone network for feature extraction. The extracted features are then divided into fine-grained local features by different granularities. Finally, a local attention feature enhancement module is used to enhance the features of salient regions in order to suppress background information.

5. The high-speed reading compensation method for RFID baggage tags based on baggage re-identification according to claim 4, characterized in that: When performing multi-granularity information aggregation, information is aggregated by adding local and global features element by element.

6. The high-speed reading compensation method for RFID baggage tags based on baggage re-identification according to claim 5, characterized in that: When updating model parameters, cross-entropy loss and three sets of losses are used as loss functions to optimize the model. Cross-entropy is used to calculate identity loss.

7. A high-speed RFID baggage tag reading compensation system based on baggage weight identification, used to execute the high-speed RFID baggage tag reading compensation method based on baggage weight identification as described in claim 1, characterized in that, include: The feature extraction module is used to perform image feature extraction during the baggage detection process, using a ResNet network as the backbone network for feature extraction. The baggage re-identification module introduces global and local feature enhancement. After processing the extracted image features, it performs multi-granularity information aggregation and updates the model parameters through backpropagation of the loss to build a recognition model. This model is then applied to airport baggage sorting systems to achieve automatic baggage sorting and matching.

8. The RFID baggage tag high-speed reading compensation system based on baggage re-identification according to claim 7, characterized in that: The baggage re-identification module introduces a global attention feature enhancement module after the backbone network and before multi-granularity feature aggregation. The feature map output from the backbone network is used as input to the global feature enhancement module. After the image is input to this module, the number of channels in the feature map is halved while maintaining its size. The feature map is then mapped to different spaces through three different linear mapping functions, resulting in three different feature vectors. Attention is then calculated, and the enhanced features are rearranged and convolutionally processed to restore the enhanced feature vectors to the same size as the input feature map. Simultaneously, the acquired baggage image is processed through the backbone network for feature extraction. The extracted features are then divided into fine-grained local features based on different granularities. These fine-grained local features are then enhanced through the local attention feature enhancement module to suppress background information in salient regions.

9. The high-speed reading compensation system for RFID baggage tags based on baggage re-identification according to claim 8, characterized in that: The baggage re-identification module aggregates information by adding local and global features element by element.

10. The RFID baggage tag high-speed reading compensation system based on baggage weight identification according to claim 9, characterized in that: The baggage re-identification module uses cross-entropy loss and three sets of losses as loss functions to optimize the model, where cross-entropy is used to calculate identity loss.

Images