Published Online:October 2024
Product Name:The IUP Journal of Computer Sciences
Product Type:Article
Product Code:
Author Name:Okonkwo Ngozi Ukamaka , Ogwueleka Francisca Nonyelum , Prasad Rajesh and Muhammad Sanusi
Availability:YES
Subject/Domain:Engineering
Download Format:PDF
Pages:13
The paper investigates the effectiveness of Convolutional Neural Networks (CNN), ResNet, and EfficientNet in classifying high and low resolution satellite images from diverse sources and applications. The models are evaluated based on precision, recall, accuracy and F1 score. The CNN model demonstrates a well-balanced performance with precision at 85%, recall at 95% and an overall accuracy of 91%, resulting in a robust F1 score of 90%. ResNet exhibits moderate accuracy (68%) and a tradeoff between precision (65%) and recall (67%), reflected in F1 score of 64%. EfficientNet stands out with exceptional precision (98%) and commendable accuracy (95%), maintaining a strong F1 score of 92%. These findings highlight the diverse capabilities of each model, providing valuable insights for selecting an optimal approach based on specific requirements for classifying high and low resolution satellite images across various sources and applications.
Satellites are advanced technological apparatus installed in an orbit for data gathering and communication network operations (Yang and Yu, 2023). They are critical in a variety of applications such as urban planning, environmental monitoring and military intelligence. They function as self-contained communication systems, outfitted with transponders that can receive and retransmit Earth’s communications, a dual-purpose radio signal receiver and transmitter (Lu et al., 2023).