Abstract
This study presents a new efficient technique for land-use classification. The presented model uses an optimized version of MobileNetV3 architecture. The network optimization has been done based on a Greedy variant of the Osprey Optimizer (GOO). The availability of high-resolution satellite imagery has made accurate and efficient land-use classification crucial for various applications like environmental monitoring, urban planning, natural resource management, and disaster management. However, traditional approaches often suffer from low accuracy or high computational costs, making them unsuitable for large-scale datasets. To overcome these challenges, a GOO-optimized MobileNetV3 has been introduced as a lightweight yet effective solution for land-use classification. By fine-tuning the hyperparameters and weights of MobileNetV3, GOO enhances both convergence speed and model performance. The proposed model has been evaluated on three publicly available benchmark datasets and compared it against several state-of-the-art techniques, including AlexNet, HGVGG19, Joint Deep Learning, DE-UNet, Shapley Additive Explanations (SHAP). The experimental results demonstrate that the model surpasses existing models across multiple metrics. Furthermore, we conduct comprehensive ablation studies to validate the effectiveness of each component in the framework. The findings highlight the potential of combining deep learning architectures with bioinspired optimization algorithms to enhance land-use classification tasks while reducing computational complexity.