Investigations on Deep Learning Pre-trained Model VGG-19 Using Transfer Learning for Remote Sensing Image Classification on Benchmark Datasets

Retrieving remote sensing images involves efficiently and accurately selecting relevant images from large archives. The selection process is based on comparing the features of a query image with the features of images in the archives. Extracting the most relevant features from both the query image and the archive images is crucial. Image classification serves as a fundamental step in this process. Features are extracted to enable further processing using either traditional machine learning or modern deep learning approaches. Recent studies have shown that deep learning methods outperform traditional machine learning methods in image classification. In this study, the features of remote sensing images are extracted using the VGG-19 deep learning model on four popular benchmark datasets UC MERCED, AID, NWPU-RESISC45 and PatterNet. The experimental results demonstrate that the best performance in feature extraction using the VGG-19 deep learning model is achieved with the PatterNet dataset. The evaluation metrics indicate a precision of 92%, recall of 91%, F1 score of 90%, accuracy of 91% and a minimum validation loss of 0.3. These results highlight the effectiveness of the PatterNet dataset for remote sensing image classification and retrieval.