DI3CL: Contrastive Learning With Dynamic Instances and Contour Consistency for SAR Land-Cover Classification Foundation Model

arXiv — cs.CV•Thursday, November 13, 2025 at 5:00:00 AM

The recent publication of the DI3CL model marks a significant advancement in SAR land-cover classification, addressing the prevalent reliance on supervised learning methods that depend heavily on large labeled datasets. This dependency has limited the scalability and generalization of existing approaches. DI3CL introduces a Dynamic Instance module that enhances contextual awareness and a Contour Consistency module that focuses on the geometric contours of land-cover objects, improving structural discrimination. With a robust pre-training framework, DI3CL is designed to serve as a general-purpose foundation model, facilitating the development of various downstream applications. The model is trained on a large-scale dataset comprising 460,532 SAR images, which enhances its robustness and adaptability across different classification tasks. This innovation not only accelerates the deployment of SAR classification models but also opens avenues for more efficient and effective applications i…

— via World Pulse Now AI Editorial System

Read Original

Was this article worth reading? Share it

Recommended Readings

arXiv — stat.ML16 hours ago

PCA++: How Uniformity Induces Robustness to Background Noise in Contrastive Learning

PositiveArtificial Intelligence

The article presents PCA++, an advanced method in contrastive learning designed to enhance the recovery of shared signal subspaces from high-dimensional data affected by background noise. Building on the limitations of PCA+, which struggles under strong noise, PCA++ employs a hard uniformity constraint to enforce identity covariance on projected features. This approach ensures stability in high-dimensional settings and offers a closed-form solution through a generalized eigenproblem, demonstrating its effectiveness in mitigating background interference.

Read full article

via arXiv — stat.ML

arXiv — cs.CL2 days ago

LANE: Lexical Adversarial Negative Examples for Word Sense Disambiguation

PositiveArtificial Intelligence

The paper titled 'LANE: Lexical Adversarial Negative Examples for Word Sense Disambiguation' introduces a novel adversarial training strategy aimed at improving word sense disambiguation in neural language models (NLMs). The proposed method, LANE, focuses on enhancing the model's ability to distinguish between similar word meanings by generating challenging negative examples. Experimental results indicate that LANE significantly improves the discriminative capabilities of word representations compared to standard contrastive learning approaches.

Read full article

via arXiv — cs.CL

arXiv — cs.CV2 days ago

Detection of Bark Beetle Attacks using Hyperspectral PRISMA Data and Few-Shot Learning

PositiveArtificial Intelligence

Bark beetle infestations pose a significant threat to the health of coniferous forests. A recent study introduces a few-shot learning method that utilizes contrastive learning to detect these infestations through satellite hyperspectral data from PRISMA. The approach involves pre-training a CNN encoder to extract features from hyperspectral data, which are then used to estimate the proportions of healthy, infested, and dead trees. Results from the Dolomites indicate that this method surpasses traditional PRISMA spectral bands and Sentinel-2 data in effectiveness.

Read full article

via arXiv — cs.CV

arXiv — cs.CV2 days ago

OpenUS: A Fully Open-Source Foundation Model for Ultrasound Image Analysis via Self-Adaptive Masked Contrastive Learning

PositiveArtificial Intelligence

OpenUS is a newly proposed open-source foundation model for ultrasound image analysis, addressing the challenges of operator-dependent interpretation and variability in ultrasound imaging. This model utilizes a vision Mamba backbone and introduces a self-adaptive masking framework that enhances pre-training through contrastive learning and masked image modeling. With a dataset comprising 308,000 images from 42 datasets, OpenUS aims to improve the generalizability and efficiency of ultrasound AI models.

Read full article

via arXiv — cs.CV