1. A Novel Pyramidal CNN Deep Structure for Multiple Objects Detection in Remote Sensing Images.
- Author
-
Elgamily, Khaled Mohammed, Mohamed, M. A., Abou-Taleb, Ahmed Mohamed, and Ata, Mohamed Maher
- Abstract
This article suggests a novel convolutional neural network (CNN) layering structure based on the pyramidal-shaped CNN model in the state of the art of remote sensing images. The suggested system outperforms the traditional CNN pre-trained models. Consequently, a detailed analysis of several CNN models has indeed been utilized. Furthermore, a comprehensive comparison has been acquired between the proposed Pyramidal Net model and nine different well-known pre-trained models to assess the efficacy of the developed framework. Ten distinct classes have been trained, tested, and validated from two different standardized datasets; NWPU-RESISC45 (Northwestern Polytechnical University Remote Sensing Image Scene Classification) and UC (University of California) Merced Land Use datasets. The utilized system performance has been evaluated based on several metrics: accuracy, recall, precision, IOU, and F1-score. Experimental findings demonstrate that the proposed Pyramidal Net CNN model has achieved an accuracy of 97.1%, recall: 0.96, precision: 0.96, IOU: 0.928, and F1-score: 0.96. The proposed model in comparison with other pre-trained CNN architectures has improved the classification accuracy by a percentage up to 30% taking into consideration a superior training time of 840 s for 5950 images with 10 different classes. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF