Sparrow Search Algorithm Enhanced Deep Learning for Remote Sensing Scene Detection and Classification

Abstract

Remote sensing (RS) scene classification using deep learning has become a rapidly advancing field, offering substantial improvements in accuracy and automation. Deep learning (DL) methods excel in extracting intricate features from RS imagery, eliminating the need for manual feature engineering required by traditional machine learning (ML) techniques. Unlike ML, where domain-specific features must be handcrafted, DL algorithms learn relevant features automatically during training. Convolutional neural networks (CNNs) are widely used in RS scene classification for their ability to capture spatial dependencies and local patterns within images. This paper introduces a novel approach, the Sparrow Search Algorithm with Deep Learning Assisted Remote Sensing Scene Detection and Classification (SSADL-RSSDC). The SSADL-RSSDC method automates the identification and classification of multiple scene labels in RS images. It begins with preprocessing using a median filtering (MF) approach, followed by feature extraction using a deep residual network (ResNet) to learn hierarchical representations of input data. The Sparrow Search Algorithm (SSA) is employed to optimize the hyperparameters of the ResNet model. For the final detection and classification, the Extreme Learning Machine (ELM) is used. The SSADL-RSSDC technique's performance is evaluated on a benchmark RS dataset, demonstrating superior results compared to other models.