A Novel Technique for Robust Training of Deep Networks With Multisource Weak Labeled Remote Sensing Data

2025-10-07

AI Summary

In remote sensing scene classification, deep networks rely heavily on large-scale, high-quality annotated data—costly to acquire—while abundant multi-source weak labels (e.g., outdated digital maps) are accessible but highly noisy. To address this, we propose a robust training framework that jointly leverages a small set of high-reliability labels and multiple weak label sources. Our method models the class-level confusion patterns of each source via learnable transition matrices and dynamically weights gradient contributions from different label sources during optimization, enabling noise-aware weak supervision. It integrates deep neural networks, differentiable transition matrix estimation, and a principled multi-source label fusion mechanism. Extensive experiments across multiple remote sensing benchmarks demonstrate that our approach significantly improves generalization and training stability under low-quality supervision, outperforming state-of-the-art weakly supervised and noise-robust learning methods.

Technology Category

• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Computer Vision: Remote Sensing / Geospatial AI
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Deep learning has gained broad interest in remote sensing image scene classification thanks to the effectiveness of deep neural networks in extracting the semantics from complex data. However, deep networks require large amounts of training samples to obtain good generalization capabilities and are sensitive to errors in the training labels. This is a problem in remote sensing since highly reliable labels can be obtained at high costs and in limited amount. However, many sources of less reliable labeled data are available, e.g., obsolete digital maps. In order to train deep networks with larger datasets, we propose both the combination of single or multiple weak sources of labeled data with a small but reliable dataset to generate multisource labeled datasets and a novel training strategy where the reliability of each source is taken in consideration. This is done by exploiting the transition matrices describing the statistics of the errors of each source. The transition matrices are embedded into the labels and used during the training process to weigh each label according to the related source. The proposed method acts as a weighting scheme at gradient level, where each instance contributes with different weights to the optimization of different classes. The effectiveness of the proposed method is validated by experiments on different datasets. The results proved the robustness and capability of leveraging on unreliable source of labels of the proposed method.

Problem

• Training deep networks with limited reliable remote sensing labels
• Leveraging multiple weak label sources while handling their errors
• Developing robust training strategies for noisy multisource datasets

Innovation

• Combining multisource weak labeled data with reliable datasets
• Embedding transition matrices to weigh label reliability
• Implementing gradient-level weighting for robust optimization