<![CDATA[With its high accuracy and efficiency, deep learning has greatly improved the classification of remote sensing (RS) photos. In order to categorize RS photos, this research analyzes the effectiveness of three cutting-edge deep learning models: ResNet-50, EfficientNetB2, and MobileNetV2. The models' accuracy on training and validation data were noted after they were trained and assessed on a dataset containing a variety of situations. Our findings illustrate each model's advantages and disadvantages and shed light on how well suited each is for various RS image categorization applications. The ResNet-50 model performed well in our study, achieving 74.41% training accuracy and 75.00% validation accuracy. With a training accuracy of 74.66% and a higher validation accuracy of 80.33%, the EfficientNetB2 model performed marginally better, demonstrating its strong generalization capabilities. On the other hand, the MobileNetV2 model had severe overfitting, as evidenced by its validation accuracy of 22.79%, which was much lower than its extraordinary high training accuracy of 99.21%. In order to achieve balanced performance between training and validation datasets in remote sensing image classification tasks, these results emphasize the significance of model architecture and regularization strategies. The proposed model can be utilized for sustainable remote sensing based applications in global water, environment and air health.]]>
Copyrights © 2024
