Advancing Remote Sensing and Deep Learning with Transformer Models and Semi-Supervised Learning

The recent developments in remote sensing and deep learning have significantly advanced the capabilities of various applications, particularly in the areas of change detection, excretion detection in livestock farming, wildfire prediction, crop segmentation, post-flood building damage assessment, and structural health monitoring anomaly detection. The field is witnessing a shift towards more sophisticated deep learning models, such as transformer-based architectures, which are showing superior performance in tasks traditionally dominated by convolutional neural networks. These models are not only enhancing accuracy but also demonstrating robustness to varying conditions and computational efficiency. Additionally, there is a growing emphasis on semi-supervised and weakly supervised learning paradigms to address the challenges posed by data scarcity and label imbalance. These approaches leverage large amounts of unlabeled data to improve model performance with minimal labeled data, which is particularly relevant in fields like structural health monitoring and crop mapping. The integration of attention mechanisms and consistency regularization in model design is also emerging as a key strategy for improving the sensitivity to subtle changes and enhancing the generalization capabilities of models. Overall, the field is progressing towards more efficient, accurate, and robust solutions that can handle complex and dynamic real-world scenarios.

Sources

GTPC-SSCD: Gate-guided Two-level Perturbation Consistency-based Semi-Supervised Change Detection

Excretion Detection in Pigsties Using Convolutional and Transformerbased Deep Neural Networks

Fire-Image-DenseNet (FIDN) for predicting wildfire burnt area using remote sensing data

Enhancing Crop Segmentation in Satellite Image Time Series with Transformer Networks

Benchmarking Attention Mechanisms and Consistency Regularization Semi-Supervised Learning for Post-Flood Building Damage Assessment in Satellite Images

Transferring self-supervised pre-trained models for SHM data anomaly detection with scarce labeled data

Exact: Exploring Space-Time Perceptive Clues for Weakly Supervised Satellite Image Time Series Semantic Segmentation

Built with on top of