AS-Faster-RCNN: An Improved Object Detection Algorithm for Airport Scene Based on Faster R-CNN

• Published in: IEEE access Vol. 13; pp. 36050 - 36064
• Main Authors: He, Zhige, He, Yuanqing
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Airport scene; Airports; Algorithms; Atmospheric modeling; CBAM; Classification algorithms; Convolution; DCN; Deep learning; Deformable models; Degradation; faster-RCNN; Feature extraction; Formability; Object detection; Object recognition; objection detection; Proposals; ResNet
• DOI: 10.1109/ACCESS.2025.3539930
• ISSN: 2169-3536

Currently, the rapid development of the aviation industry has made the safety of the airport becomes more and more important. The most important part of this is the capability of discriminate the different type of objects correctly. However, the existing detection models have the problems of degradation, lacking of detection capability for deformed and small objects and single feature extraction, causing low detection accuracy. To overcome these problems, we design an object detection method for airport scene named AS-Faster-RCNN. Firstly the ResNet-101 substitute for VGG-16 as the backbone network to improve the ability of detecting small objects, prevent the degradation and enhance the ability of detecting the small objects. Secondly, The DCN (Deformable Convolution Network) is employed in the backbone to strengthen the ability of extracting features for deformed objects. Finally, the CBAM (Convolutional Block Attention Module) is added to the backbone to extract multidimensional features to enhance performance of the model. We design some experiemnts to prove the feasibility of the method and the results demonstrate the mAP(mean Average Precision) has increased by 5.3% comapred to the basline model, and compared with other object detection models, its mAP also increased to a certain extent.