Exploring New Convolutional Neural Network Architectures for Satellite Image Categorization: A Comparative Study

With deep learning, satellite images can accurately identify land cover elements such as annual crops, forests, herbaceous vegetation, pastures, permanent crops, and rivers. Deep learning has become a valuable tool in this regard. In this field, convolutional neural networks (CNNs) have shown great promise, with pre-trained models such as DenseNet121, EfficientNetB0, VGG16, and InceptionV3 showing encouraging outcomes. To outperform earlier CNN models, this research explores the possibilities of new CNN designs for satellite image categorization. We suggest a comparative study where these more recent architectures are refined using a dataset of satellite images. The results are shown together with a description of how well the above-mentioned CNN models work for this task. The performance is assessed based on accuracy and a discussion of the model’s suitability for this task is included with the result.