Land Use and Land Cover Classification Meets Deep Learning: A Review.

Area of Science:

• Earth Observation
• Remote Sensing
• Artificial Intelligence

Background:

• Land Use and Land Cover (LULC) classification is crucial for environmental monitoring and resource management.
• High-dimensional remote sensing data and limited labeled samples present significant challenges for traditional LULC classification.
• Deep learning has emerged as a powerful tool for advancing LULC classification accuracy.

Purpose of the Study:

• To provide a systematic review of deep-learning-based approaches for LULC classification.
• To cover key components of deep learning networks, datasets, evaluation metrics, and strategies.
• To discuss challenges and future directions in the field.

Main Methods:

• Review of five typical deep learning network architectures and their benefits.
• Summary of baseline datasets and performance metrics (OA, AA, F1, MIOU) for LULC classification.
• Exploration of deep learning strategies including CNNs, AEs, GANs, and RNNs.

Main Results:

• Deep learning methods have achieved remarkable results in LULC classification.
• Identification of common deep learning strategies applied to LULC classification.
• Analysis of challenges related to limited training samples in remote sensing image classification.

Conclusions:

• Deep learning offers new possibilities for LULC classification research and development.
• Addressing challenges like limited training data is key for future advancements.
• Future development will likely focus on more sophisticated deep learning models and techniques.