Satellite Image Scene Classification Using Roulette Fitness-Distance Balance-Based Osprey Optimization Algorithm and Bidirectional-Long Short-Term Memory Approach

Remote Sensing Scene Categorization (RSSC) is essential for identifying various applications, such as farms, highways, rivers, and urban areas. Accurate classification and high spatial resolution performs important role in analysis of satellite images. However, effective feature representation is essential for optimal performance, as processing sensor data can be complex. The irrelevant features causing a dimensionality challenges. This research proposes the Roulette Fitness-Distance Based Osprey Optimization Algorithm (RFDB-OOA) for feature selection to identify the most relevant features for efficient scene classification. OOA is combined with the Roulette Fitness-Distance Balance-based Method (RFDB), improve exploration ability of the algorithm and enhance a balance between exploration and exploitation. Generative Adversarial Network (GAN) utilized for augmentation process to generate augmented images, which are then processed by AlexNet and VGG19 models to extract discriminative features. Bidirectional Long Short-Term Memory (Bi-LSTM) network is utilized for accurate classification of images. Simulation results illustrates that the RFDB-OOA-BiLSTM approach achieves 98.67% and 99.70% on both the AID and UCM datasets as compared to the existing methods like CNN-VGG 16 and Multi-Level Semantic Feature Clustering Network (MLFC-Net).