Classification of land use/land cover using artificial intelligence (ANN-RF)

Area of Science:

• Remote Sensing
• Machine Learning
• Geospatial Analysis

Background:

• Deep learning methods for land use/land cover classification present challenges in complexity, cost, and processing time.
• There is a need for simpler, more efficient machine learning approaches in remote sensing applications.

Purpose of the Study:

• To enhance the accuracy of machine learning for land use/land cover classification.
• To develop and evaluate a novel hybrid model combining artificial neural networks and random forest for improved classification performance.

Main Methods:

• A novel hybrid model, Artificial Neural Network with Random Forest (ANN_RF), was developed.
• Multispectral satellite imagery from Sentinel-2A and Landsat-8, along with a normalized digital elevation model, were used as input data.
• The model was applied to classify land use/land cover for Sana'a city in 2016.

Main Results:

• The proposed ANN_RF model demonstrated superior accuracy compared to individual ANN classifiers using Sentinel-2A and Landsat-8 data.
• The ANN_RF model achieved high classification accuracy without employing deep learning techniques.
• The model effectively utilized traditional artificial neural networks with a significant number of simulated neurons.

Conclusions:

• The ANN_RF model offers a more accurate and potentially less complex alternative to deep learning for land use/land cover classification.
• This research contributes to the advancement of machine learning in remote sensing, providing a valuable tool for researchers and specialists.
• The findings suggest that hybrid machine learning approaches can effectively leverage satellite data for detailed land cover mapping.