Satellite image (SI) supported earth observation and environmental monitoring is one of the prime tasks. Recently, the artificial intelligence (AI)-based SI examination is widely discussed in the literature. This research aims to develop a DL-technique to detect desert and mountain from the chosen SI database. Various stages in this DL-approach includes; image collection and resizing, feature extraction using a chosen DL-model, feature reduction and serial features concatenation, classification, and implementing threefold cross validation to confirm the performance. In this work, the DenseNet-variants-based approach is considered to extract the image features and then a 50% feature reduction is executed to reduce the deep features. The reduced deep features from two chose models are integrated serially to get a new feature vector, and this feature vector is then considered to detect the desert/mountain from the chosen SI-data. The outcome of this comparison confirms that the proposed approach provides 100% accuracy when Random Forest (RF)-based classification is executed.