Automated Object Detection and Recognition in Satellite Imagery Using CNNs

As increasingly high-resolution satellite photography becomes available, it has led to a growing need for efficient and accurate methods to analyse large databases and extract valuable information. This research presents a comprehensive approach for automated item identification and detection in satellite data using Convolutional Neural Networks (CNNs). This research abstract outlines the critical steps involved in a suggested technique. These steps encompass data cleaning and organisation, model architecture development, training and evaluation, post- processing, deployment, and subsequent management. This study investigates the ability of CNNs to learn intricate spatial characteristics. The research focuses on carefully selecting and annotating training and validation datasets incorporating items of interest. This study uses data augmentation methods to enhance the generalisation capabilities of the model. This research abstract discusses the importance of the preprocessing stage in ensuring the consistent performance of a model by normalising the input data. This research abstract discusses the components of the CNN architecture, namely convolutional, pooling, and fully connected layers. The CNN architecture is specifically designed to capture hierarchical characteristics. This study investigates the utilisation of a cutting-edge optimiser in the training process to optimise a meticulously constructed loss function. This study examines the integration of the model into real-world applications, considering the trade-off between processing speed and resource constraints as crucial considerations for successful deployment. The findings suggest that this methodology holds promise as a valuable tool in various fields.