The accessibility of satellite images has significantly improved due to the advancements in remote sensing technologies. To effectively apply remote sensing in various real-world scenarios, it is crucial to explore efficient and scalable strategies for extending its applications to new domains. Deep Learning (DL) techniques play a crucial role in Remote Sensing Imaging, aiding in achieving rapid analysis and accurate classification goals. This research focuses on classifying weather conditions using relevant satellite images, which often pose challenges such as poor quality and uneven data distribution. Despite the widespread use of Convolutional Neural Networks (CNNs) in image classification, they suffer from poor performance, mainly due to their limited global representational capabilities and reliance on local receptive fields. In addressing this issue, we introduce CNN-T, a network architecture that combines the Transformer encoder and CNN features. The CNN-T network, with its robust global modeling capabilities and high inductive biases, emerges as a powerful solution. The CNN extracts local and low-level features from the images, while the transformer encoder captures abstract and high-level semantic information by globally modeling these features. Subsequently, the multilayer perceptron (MLP) classifier receives the data to make the final decision. To assess the recognition performance of satellite images, this study compares the proposed CNN-T model against various DL models, including ResNet-18, AlexNet, and GoogLeNet. Each model's effectiveness is evaluated using a diverse set of measures. Experimental results demonstrate that the CNN-T model outperforms other models on the Large-Scale Cloud Image Database for Meteorological Research (LSCIDMR) dataset, achieving an impressive classification accuracy of 99%. This indicates the effectiveness of the proposed hybrid model in handling challenges associated with satellite image classification.

Deep Learning, Weather Classification, Transformer, Convolutional Neural Network, Satellite Images,