Improving remote sensing scene classification using dung Beetle optimization with enhanced deep learning approach.

Area of Science:

• Earth and Space Sciences
• Computer Science
• Artificial Intelligence

Background:

• Remote sensing (RS) scene classification is crucial for applications like surveillance, urban planning, and environmental observation.
• Existing convolutional neural network (CNN) methods struggle with RSIs due to complex textures, cluttered scenes, and scale variations.
• Accurate classification of land scenes from remote sensing images (RSIs) remains a challenge.

Purpose of the Study:

• To develop an advanced Remote Sensing Scene Classification using Dung Beetle Optimization with Enhanced Deep Learning (RSSC-DBOEDL) approach.
• To effectively categorize diverse scenes within remote sensing images (RSIs).
• To enhance the accuracy and contextual understanding of RSIs classification.

Main Methods:

• Utilized an enhanced MobileNet model as the primary feature extractor for RSIs.
• Employed Dung Beetle Optimization (DBO) for hyperparameter tuning of the enhanced MobileNet.
• Implemented a multi-head attention-based long short-term memory (MHA-LSTM) network for scene classification.

Main Results:

• The RSSC-DBOEDL approach achieved high accuracy rates of 98.75% on the UC Merced dataset.
• The method demonstrated strong performance with 95.07% accuracy on the EuroSAT dataset.
• Outperformed existing methods in classifying scenes within benchmark RSI datasets.

Conclusions:

• The RSSC-DBOEDL approach offers a superior method for remote sensing scene classification.
• The integration of DBO and MHA-LSTM with enhanced MobileNet significantly improves classification accuracy.
• This technique provides a robust solution for analyzing complex remote sensing images.