Multi-Deep-Model Ensemble Technique for Aircraft Detection in Remote Sensing Images
Laveena Herman, Meenakshy Balakrishnan, Anakha Anie Jose, Kavya Chandran, A. Shyna · 2023
Aircraft detection in remote sensing images is a challenging problem due to the small and densely distributed nature of objects. In this work, an ensemble-based multi-deep learning model is proposed which exploits the properties of both attention-based ResNet34 and YOLO V5 models for accurate aircraft object detection. Convolutional Block Attention Module (CBAM) is incorporated into ResNet34 to enhance the informative regions in the feature maps and suppress irrelevant information. To accurately locate the position of the object, a Multilevel fusion module is incorporated to fuse the output of shallow and deep class activation maps extracted from different layers of ResNet-CBAM. The pre-trained YOLO V5 model, utilizing fully labeled data, is used as the second model for ensembling. To integrate the results from both models, a non-maximum suppression ensemble technique is used. The performance of the proposed work is evaluated on the WSADD dataset in terms of different quantitative metrics such as Precision, Recall, and F1 score, and it has been found that the proposed methods yield better performance compared to the state-of-the-art methods.