Land Resource Use Classification Using Deep Learning in Ecological Remote Sensing Images.

Area of Science:

• Ecological remote sensing
• Geospatial analysis
• Machine learning for environmental monitoring

Background:

• Traditional remote sensing image classification methods struggle with integrating diverse deep learning features, leading to suboptimal performance.
• Effective land resource use classification is crucial for ecological monitoring and sustainable management.

Purpose of the Study:

• To propose an improved land resource use classification method for ecological remote sensing images.
• To enhance the integration of deep learning features for better classification performance.
• To validate the proposed method using real-world remote sensing data.

Main Methods:

• A seven-layer convolutional neural network (CNN) was constructed for feature extraction.
• Features from fully connected layers were fused with principal component analysis (PCA)-reduced pooled layer features.
• A support vector machine (SVM) classifier was employed for the final land resource classification.

Main Results:

• The proposed method achieved high classification accuracy (0.9472).
• The misclassification rate was low (0.0528), and the Kappa coefficient was high (0.9435).
• The method demonstrated clear image edge recognition and excellent overall performance.

Conclusions:

• The developed CNN-based method effectively integrates deep learning features for land resource classification in ecological remote sensing.
• The fusion strategy significantly improves classification accuracy and reliability.
• The approach shows strong potential for practical applications in land resource management and ecological studies.