A Deep Learning-Based Method for Fine-Grained Identification of Airport Operation Vehicles

By combining the YOLOv5 network with data augmentation and multi-scale contrastive learning, the adaptability of airport vehicle detection in diverse environments was solved, achieving efficient and accurate vehicle recognition.

CN116168353BActive Publication Date: 2025-10-28NANJING UNIV OF AERONAUTICS & ASTRONAUTICS
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202310027227.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-01-09
Publication Date
2025-10-28
Estimated Expiration
2043-01-09

AI Technical Summary

Technical Problem

Existing airport vehicle detection algorithms lack adaptability and applicability in handling large image sizes, images containing small objects, and images with varying angles and diverse weather conditions. In particular, their detection accuracy and efficiency are not high in airport environments.

Method used

The YOLOv5 network was used for training, and data augmentation with fog and rain processing was combined. Multi-scale contrastive learning and attention-guided feature fusion were used to improve the robustness and accuracy of vehicle detection. BCE loss and EIoU loss were used as loss functions to calculate the cross-entropy and contrastive loss of multi-scale feature maps.

Benefits of technology

It improves the adaptability and accuracy of airport vehicle detection, enabling efficient and accurate vehicle detection under different weather conditions and perspective changes, and is suitable for the classification and identification of various vehicles.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116168353B_ABST
    Figure CN116168353B_ABST
Patent Text Reader

Abstract

This invention provides a deep learning-based method for fine-grained identification of airport operational vehicles. First, airport vehicle images are acquired and labeled to obtain an airport vehicle dataset. Second, spatial transformations are applied to the images to change the perspective, and fog and rain are added to expand the dataset. Next, the labeled vehicle dataset is fed into a deep learning network for training until the network converges. Finally, the trained deep learning network and weight file are used to detect vehicle targets in test images, and the classification results are output. This invention offers high accuracy and robustness, overcoming the challenges of complex weather and angles that traditional image processing algorithms struggle to handle in vehicle classification.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Vehicle detection method in aerial images of unmanned aerial vehicle based on deep learning

    CN110298227A

  • Typical target detection method and system under complex background based on deep learning

    CN113128564A