Related Experiment Video
Updated: Jun 23, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Hybridizing Deep Neural Networks and Machine Learning Models for Aerial Satellite Forest Image Segmentation
Clopas Kwenda1, Mandlenkosi Gwetu2, Jean Vincent Fonou-Dombeu1
1School of Mathematics, Statistics and Computer Science, University of KwaZulu Natal, Pietermaritzburg 3209, South Africa.
Journal of Imaging
|June 26, 2024
Summary
This study introduces a hybrid deep learning and machine learning model for accurate forest cover segmentation from aerial images. The novel approach significantly improves segmentation performance, outperforming existing methods.
Area of Science:
- Remote Sensing and Geospatial Analysis
- Machine Learning and Artificial Intelligence
- Environmental Monitoring and Climate Change Mitigation
Background:
- Accurate forest cover monitoring is crucial for climate change mitigation and socio-economic activities.
- Traditional image segmentation methods struggle with spatial and textural feature extraction, leading to suboptimal forest cover classification.
- Deep neural networks offer advanced feature extraction capabilities but often require substantial data and computational resources.
Purpose of the Study:
- To develop and evaluate a novel hybrid approach combining deep neural networks (VGG16, ResNet50) and machine learning classifiers (Random Forest, LSVM, kNN, LDA, GNB) for aerial satellite image segmentation.
- To enhance the accuracy and efficiency of forest and non-forest region segmentation.
- To compare the performance of the hybrid model against traditional methods and existing studies.
Main Methods:
- Feature extraction from aerial satellite forest images using pre-trained VGG16 and ResNet50 deep neural network models.
- Segmentation of forest and non-forest regions using five machine learning classifiers (Random Forest, LSVM, kNN, LDA, GNB) trained on extracted deep features.
- Performance evaluation using Accuracy, Jaccard index, and Root Mean Square Error (RMSE) on a deep globe challenge dataset.
Main Results:
- The hybrid Random Forest model achieved the highest performance with 94% accuracy, 0.913 Jaccard index, and 0.245 RMSE.
- The hybrid approach significantly improved the segmentation performance of all tested machine learning classifiers compared to their standalone use.
- The proposed model demonstrated superior segmentation capabilities, outperforming other models in related studies.
Conclusions:
- The hybrid deep learning and machine learning approach offers a powerful and effective solution for accurate forest cover segmentation from aerial imagery.
- The integration of deep neural networks for feature extraction enhances the performance of traditional machine learning classifiers for this task.
- This methodology provides a robust tool for environmental monitoring and climate change research.
