Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing Hyperspectral Image Prediction with Contrastive Learning in Low-Label Regime

Salma Haidar, José Oramas

TL;DR

The qualitative analysis confirms that the contrastive-based encoder can produce well-separated representations for different classes and identify location-based features, even though it was not explicitly trained on spatial cues, suggesting the method’s potential to uncover implicit spatial information.

Abstract

Self-supervised contrastive learning is an effective approach for addressing the challenge of limited labelled data. This study builds upon the previously established two-stage patch-level, multi-label classification method for hyperspectral remote sensing imagery. We evaluate the method's performance for both the single-label and multi-label classification tasks, particularly under scenarios of limited training data. The methodology unfolds in two stages. Initially, we focus on training an encoder and a projection network using a contrastive learning approach. This step is crucial for enhancing the ability of the encoder to discern patterns within the unlabelled data. Next, we employ the pre-trained encoder to guide the training of two distinct predictors: one for multi-label and another for single-label classification. Empirical results on four public datasets show that the predictors trained with our method perform better than those trained under fully supervised techniques. Notably, the performance is maintained even when the amount of training data is reduced by $50\%$. This advantage is consistent across both tasks. The method's effectiveness comes from its streamlined architecture. This design allows for retraining the encoder along with the predictor. As a result, the encoder becomes more adaptable to the features identified by the classifier, improving the overall classification performance. Qualitative analysis reveals the contrastive-learning-based encoder's capability to provide representations that allow separation among classes and identify location-based features despite not being explicitly trained for that. This observation indicates the method's potential in uncovering implicit spatial information within the data.

Enhancing Hyperspectral Image Prediction with Contrastive Learning in Low-Label Regime

TL;DR

The qualitative analysis confirms that the contrastive-based encoder can produce well-separated representations for different classes and identify location-based features, even though it was not explicitly trained on spatial cues, suggesting the method’s potential to uncover implicit spatial information.

Abstract

Self-supervised contrastive learning is an effective approach for addressing the challenge of limited labelled data. This study builds upon the previously established two-stage patch-level, multi-label classification method for hyperspectral remote sensing imagery. We evaluate the method's performance for both the single-label and multi-label classification tasks, particularly under scenarios of limited training data. The methodology unfolds in two stages. Initially, we focus on training an encoder and a projection network using a contrastive learning approach. This step is crucial for enhancing the ability of the encoder to discern patterns within the unlabelled data. Next, we employ the pre-trained encoder to guide the training of two distinct predictors: one for multi-label and another for single-label classification. Empirical results on four public datasets show that the predictors trained with our method perform better than those trained under fully supervised techniques. Notably, the performance is maintained even when the amount of training data is reduced by . This advantage is consistent across both tasks. The method's effectiveness comes from its streamlined architecture. This design allows for retraining the encoder along with the predictor. As a result, the encoder becomes more adaptable to the features identified by the classifier, improving the overall classification performance. Qualitative analysis reveals the contrastive-learning-based encoder's capability to provide representations that allow separation among classes and identify location-based features despite not being explicitly trained for that. This observation indicates the method's potential in uncovering implicit spatial information within the data.

Paper Structure

This paper contains 23 sections, 5 equations, 10 figures, 13 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Multi-label classification in hyperspectral image analysis
  4. Self-supervised contrastive learning (SSCL)
  5. Self-supervised learning for Hyperspectral image analysis.
  6. Methodology
  7. Model Architecture
  8. Experiments
  9. Implementation Details
  10. Multi-Label Classification
  11. Performance Across Different Training Schemes
  12. Impact of the Amount of Labelled Data
  13. Single-Label Classification
  14. Impact of the Amount of Labelled Data
  15. Effect of the Temperature parameter from the Contrastive Learning stage
  16. ...and 8 more sections

Figures (10)

  • Figure 1: Two-stage method: In the first stage a contrastive-based encoder is trained using unlabelled data to maximise the agreement between hidden representations of two views of the same instance. In the second stage a classifier is fine-tuned with the pre-trained encoder on labelled examples for multi-label prediction.
  • Figure 2: Pavia University and Salinas ground truth visualisations.
  • Figure 3: Houston 2013 and Houston 2018 ground truth visualisations.
  • Figure 4: CL-based Multi-label Classifier: Model Accuracy Performance (in$\%$)- Impact of Reduced Labelled Data
  • Figure 5: CL-based Single-label Classifier: Model Accuracy Performance (in$\%$)- Impact of Reduced Labelled Data
  • ...and 5 more figures