Luggage wheel target identification method applied to airport luggage detection system

Abstract

The invention provides a luggage wheel target identification method applied to an airport luggage detection system. The method comprises the following steps: applying a network structure with a better luggage wheel identification effect to a target identification network and successfully loading the network structure to luggage detection system software; according to the invention, a dual-path self-learning convolution sub-module is adopted, so that a network DS-CCNet with a higher luggage wheel recognition rate is designed and invented; a detection head Pect in a middle recognition area is emphasized, so that the recognition speed is increased, and meanwhile, the luggage wheel recognition effect is not reduced; the final test data proves that the luggage wheel identification effect of the improved network model DSYOLO network is much higher than that of the original identification method.

Description

Technical Field

[0001] This invention relates to the field of computer vision, specifically to a method for recognizing luggage wheels in an airport baggage inspection system. Background Technology

[0002] With the development of the air transport industry and the continuous advancement of smart airport construction, automated baggage sorting and real-time monitoring systems have become crucial for improving operational efficiency and ensuring transportation safety. Target detection technology, as the core of this system, places higher demands on the accuracy and real-time performance of the algorithm. However, baggage targets vary greatly in scale: distant baggage appears as small targets in images with weak feature information, while nearby baggage may occupy most of the frame. Secondly, there are many difficult samples: key components such as baggage wheels and handles are extremely small and prone to motion blur during conveyor belt movement, making feature extraction difficult. Furthermore, airport operations are typically deployed on embedded devices with limited computing resources, placing extremely high demands on the algorithm's lightweight design and inference speed. During baggage transportation, factors such as the baggage's weight, size, and surface material may cause baggage to stagnate on the conveyor belt or fall around the conveyor belt due to sorting machine errors, resulting in abnormal baggage status.

[0003] Each category corresponds to different tasks. Luggage, specifically hard-shell suitcases with handles, wheels, and other distinctive features, is often cubic in shape and relatively heavy when checked in. The wheels are usually movable and rotatable, and are relatively fragile compared to the suitcases and transport equipment. Hard-shell bags experience significant inertia and friction when transported on conveyor belts. Therefore, if the conveyor belt moves up and down during sorting or transport, and the luggage is misaligned, the wheels may fall off first or get stuck on the transport barriers, causing damage or even detachment of the wheels. In severe cases, this can even render the airport's physical baggage sorting system inoperable. Damage to suitcases during check-in is common. Detecting wheel position allows us to determine the luggage's movement by observing wheel position and direction, reminding staff or passengers to correct their luggage's position.

[0004] Existing high-performance detection models often have a large number of parameters and high computational costs, making them difficult to deploy directly; while lightweight models often sacrifice accuracy, especially in small target detection where performance degrades significantly. Therefore, developing a novel target detection method that combines high accuracy with strong small target detection capabilities to meet the comprehensive requirements of airport baggage inspection systems for reliability, real-time performance, and ease of deployment has become an important technical need to be addressed in this field. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention proposes a method for identifying baggage wheels as targets in airport baggage inspection systems. This method mainly includes the following technical solutions.

[0006] Step S1: A dual-path self-learning convolutional module was designed, and the CCnet attention mechanism was improved to construct a network DS-CCNet with higher attention to the wheels. This mainly replaced the post-processing module for the luggage wheel feature maps.

[0007] Step S2: Based on the Pconv approach, the Detect head was reconstructed, and a detection head PDetect that focuses on the central recognition region was designed, with the deep convolution module being removed.

[0008] The beneficial effects of this invention include: An integrated and improved module was used to form a DS-YOLO network. The improved network was trained and tested on an airport baggage dataset to achieve high-precision baggage wheel recognition. In experimental tests, the baggage wheel recognition accuracy was improved to 88.2%. Analysis of the model data showed that the number of model parameters and computational cost were significantly reduced, making it more suitable for deployment on embedded devices. Compared with previous object detection models, its accuracy in recognizing baggage wheels was significantly improved, and the false negative rate was significantly reduced. Attached Figure Description

[0009] Figure 1 This is an overall flowchart of a baggage wheel target recognition method applied to an airport baggage inspection system according to the present invention; Figure 2 This is an example diagram of DS-CCNet in a baggage wheel target recognition method applied to an airport baggage inspection system according to the present invention; Figure 3 This is a diagram of the DS-YOLO network in a baggage wheel target recognition method applied to an airport baggage inspection system according to the present invention. Figure 4 This is an example diagram of a baggage wheel target recognition method applied to an airport baggage inspection system according to the present invention. Detailed Implementation Method 1

[0010] The baggage inspection process will now be described with reference to the accompanying drawings in the embodiments of the present invention.

[0011] like Figure 1 As shown, a method for detecting baggage and baggage wheels in an airport baggage inspection system includes the following steps: 1. After the original image is input into the network, it passes through the P2 detection layer to extract shallow details, improving the detection capability of luggage wheels and adding color and structural information of luggage wheels to the feature map; 2. Feature diagrams of the color and structural information of luggage wheels were processed. Figure 1 In step S1: The feature map is generated by using the DS-CCNet network, which focuses more on luggage wheels, to create a feature map that is more focused on the wheels. This feature map is then output to step S2. 3. Feature images with more prominent luggage wheels have been processed. Figure 1 In step S2: Multiple feature maps are input to the designed PDetect inspection head to reduce computation while outputting recognition results; 4. The output results of the original image after being processed by the DS-YOLO network formed by the integrated improvement module are shown in the appendix. Figure 3 This is the DS-YOLO overall network. Data testing revealed improved network recognition accuracy, indicating a rise in luggage wheel recognition precision. Detailed Implementation Method 2

[0012] The baggage inspection process will now be described with reference to the accompanying drawings in the embodiments of the present invention.

[0013] like Figure 1 As shown, a method for identifying baggage wheels in an airport baggage inspection system includes the following steps: 1. After the original image is input into the network, it passes through the P2 detection layer to extract shallow details, improving the detection capability of luggage wheels and adding color and structural information of luggage wheels to the feature map; 2. The feature image after adding color and structural information of the luggage wheels in P2, after... Figure 1 In step S1, the feature map is processed by the DS-CCNet network, which has a higher focus on the wheel, to form a more focused feature map, and then output to step S3. For example, if the feature map is attached... Figure 4 (a) Attention is drawn to the luggage wheels more than other areas; 3. Multiple feature images that draw more attention to luggage wheels were processed. Figure 1 In step S2, multiple feature maps are input to the designed PDetect inspection head to reduce computation while outputting recognition results; 4. The output results of the original image after being processed by the DS-YOLO network formed by the integrated improvement module are shown in the appendix. Figure 3 It is the DS-YOLO overall network. Data testing was conducted to obtain the network's recognition performance, showing improved wheel detection accuracy and a better recognition result. Figure 4 (b).

[0014] To verify the effectiveness of the baggage wheel target recognition method of the present invention applied to airport baggage inspection systems, the present invention also provides the following experiments.

[0015] (1) Experimental data and experimental environment.

[0016] The baggage data used in this invention comes from surveillance cameras in both real and simulated airport environments. The surveillance video data was processed as follows: depending on the acquisition environment, frames of 3 or 5 were extracted from videos with different baggage densities. Only valid images containing baggage items were selected through manual screening. Image enhancement was performed on the baggage, including adjusting image saturation and adding salt-and-pepper noise. The images were finely labeled, and the item types were set into six categories: baggage (hardcase), baggage wheels, etc. A total of 11,025 airport baggage images were collected.

[0017] The experiment used Windows 11 operating system, NVIDIA GeForce RTX 4060 graphics card, Python 3.11, and PyTorch deep learning framework.

[0018] (2) Experimental results and analysis

[0019] The random airport baggage dataset was trained and tested using the ensemble-improved DS-YOLO algorithm. Some results are shown in Table 1: Table 1 Performance Comparison of Detection Results Model AP mAP50 AP (hard case) AP (wheel) Parameters GFLOPs YOLOV8s 0.913 0.922 0.901 0.841 11127906 28.4 DS-YOLOv8 0.933 0.931 0.931(+0.03) 0.882(+0.041) 10264248 28.2

[0020] Experimental results show that the DS-YOLO model achieved a highest average accuracy of 93.3%, with a 3% improvement in baggage detection accuracy and a 4.1% improvement in baggage wheel recognition accuracy, significantly outperforming other comparative models. This directly proves the effectiveness of the proposed improved module in enhancing baggage wheel detection accuracy. Furthermore, DS-YOLOv8 demonstrates a significant advantage in computational efficiency. Compared to YOLOv8s, the model reduces GFLOPs by 0.2 and the number of parameters by approximately 863,658. This result indicates that DS-YOLOv8 successfully reduces the computational complexity of the model while maintaining high accuracy, making it an effective method for baggage and baggage wheel target detection in airport baggage inspection systems.

[0021] It should be noted that the above embodiments are only used to explain the technical solutions of the present invention and do not constitute a limitation. Those skilled in the art, based on their understanding of the present invention, can still modify the technical solutions in the embodiments or make equivalent substitutions for some technical features, and such modifications or substitutions should all be considered within the scope covered by the technical solutions of the present invention.

Claims

1. A luggage wheel target recognition method applied in an airport luggage detection system, characterized in that, The method comprises the following steps: S1: an invention two-way self-learning convolutional submodule is included, and a network structure DS-CCNet with higher wheel recognition rate is designed, S2: a PDetect inspection head focusing on the recognition area of the wheels in the middle of the image is designed, and the wheels can be accurately recognized.

2. The luggage wheel target recognition method applied in the airport luggage detection system according to claim 1, characterized in that: The double-path self-learning convolution sub-module in step S1 can simultaneously extract local spatial context information and detailed texture information. A two-path self-learning convolution structure is invented, the first path is a 1×1 point convolution w (1) which retains the detailed texture wheel color information and obtains a dynamic weight w1, the second path is a 3×3 deep convolution w (2) which extracts local spatial wheel structure context information and a 1×1 convolution w (3) which adjusts the channel and obtains a dynamic weight w2, and a feature map O with higher wheel attention is obtained, such as .

3. The luggage wheel object recognition method for use in an airport luggage detection system according to claim 1, characterized in that: In step S3, the PDetect detection head uses a partial convolution PConv instead of a deep convolution, focuses on the middle part of the image recognition area, inputs multiple feature maps, outputs the category probability of each wheel anchor point, and outputs the wheel boundary box coordinates. And can significantly reduce the amount of calculation, the calculation formula is .

4. The luggage wheel object recognition method according to claim 1, applied to an airport luggage detection system, characterized in that: The recognition accuracy of the wheels by the DS-YOLO network structure designed through the S1 and S2 operations can reach 93.1% and 89.8% respectively, and the Parameters is reduced by 863658 compared with YOLOv8s.

Images