ResM-FusionNet for efficient landslide detection algorithm with a hybrid architecture.

Area of Science:

• Geological hazard assessment
• Remote sensing and geospatial analysis
• Deep learning for environmental monitoring

Background:

• Landslides represent a significant geological hazard, impacting environments and human populations.
• Accurate identification of landslide-prone areas is vital for effective disaster response, risk assessment, and urban planning.
• Existing landslide detection methods face challenges in complex terrains and precise boundary delineation.

Purpose of the Study:

• To propose a novel deep learning-based landslide detection method, ResM-FusionNet.
• To enhance segmentation accuracy and boundary detail detection in landslide-prone regions.
• To evaluate the performance of ResM-FusionNet against existing state-of-the-art models.

Main Methods:

• ResM-FusionNet utilizes ResNet-50 for feature extraction and a multi-layer perceptron decoder for improved segmentation.
• A novel loss function, RLoss, is introduced, incorporating masking, semantic weighting, and Top-K pixel averaging.
• The model was trained and validated on remote sensing datasets.

Main Results:

• ResM-FusionNet achieved 94.33% accuracy, 85.73% F1-score, and a Kappa coefficient of 70.12%.
• The model outperformed SegFormer, DeepLabv3, and UNet in accuracy by significant margins.
• Excellent performance in boundary detection was observed, with an IoU of 0.7545, 85.61% precision, and 83.92% recall.

Conclusions:

• ResM-FusionNet offers a robust and accurate solution for landslide detection.
• The proposed method demonstrates superior performance, particularly in complex terrains and for detailed boundary segmentation.
• This advancement is critical for improving geological hazard management and spatial planning.