Terrestrial Ecosystem Detection Based on Deep Learning Framework and Satellite Image Geographical Information

Detection of terrestrial ecosystems involves the identification and classification of land cover types and vegetation patterns using remote sensing and geospatial analysis. It provides assistance in environmental monitoring, the assessment of biodiversity, and sustainable land management, among others. The study given proposes a hybrid Deep Learning (DL) system that simultaneously incorporates ResNet‐50 and LSTM for accurate classification of Land Use and Land Cover (LULC) in satellite images. The EuroSAT dataset, consisting of 27,000 labelled samples across 10 land‐use classes, was used to train and evaluate the proposed model. This large, diverse dataset enables robust model development for accurate land‐cover classification in satellite imagery. The sequence of activities involves advanced image pre‐processing to enhance edge preservation and reduce noise, such as bilateral, guided, and median filtering. The next instance is feature extraction performed with HOG, which retains important edge and shape properties. The extracted Histogram of Oriented Gradients (HOG) features are given into the ResNet‐50 backbone, where spatial representations at a higher level are derived. Then, those ResNet‐50 based high‐level features are fed into the LSTM layer to model temporal and sequential dependencies and then provide additional contextual information for spatial transitions. The last classification is done using a fully connected layer with a Softmax activation function that provides the class probabilities. The model gained higher performance with a precision of 0.9903, an accuracy of 0.9904, and recall of 0.9903, which showed its ability to differentiate efficiently between land cover types that are more similar. By integrating spatial and sequential modelling, the architecture proposed gives a very high degree of reliability and robustness in classification with regard to a variety of geographic and environmental conditions, rendering them well out for environmental monitoring, urban planning, and geospatial intelligence.