Related Experiment Video
Updated: Aug 10, 2026
07:05
Applying Hyperspectral Reflectance Imaging to Investigate the Palettes and the Techniques of Painters
Published on: June 18, 2021
Local receptive field constrained stacked sparse autoencoder for classification of hyperspectral images
Summary
This study introduces a locally connected stacked sparse autoencoder (SSA) with local receptive fields (LRFs) for hyperspectral image classification. The new method improves accuracy and reduces running time compared to existing techniques.
Area of Science:
- Machine Learning
- Remote Sensing
- Computer Vision
Background:
- Stacked Sparse Autoencoders (SSA) are popular for hyperspectral image classification (HSI).
- Traditional SSA architectures use fully connected layers, which can be inefficient for capturing local correlations in HSI data.
Purpose of the Study:
- To propose a novel, locally connected SSA framework incorporating biologically inspired Local Receptive Fields (LRFs).
- To enhance feature extraction and classification accuracy for hyperspectral images while reducing computational time.
Main Methods:
- A hierarchical Local Receptive Field (LRF) constrained Stacked Sparse Autoencoder (SSA) architecture was developed.
- The receptive field constraint was dynamically updated based on spatial distances between nodes.
- An efficient Random Forest classifier was integrated with the SSA model (SSARF).
Main Results:
- The proposed SSARF method demonstrated significant improvements in overall accuracy on the Indian Pines (0.72%-10.87%) and Kennedy Space Center (0.74%-7.90%) datasets.
- The SSARF model achieved lower running times compared to similar methodologies.
- The LRF constraint effectively captured local spectral feature correlations.
Conclusions:
- The hierarchical LRF-constrained SSA offers a more efficient and accurate approach to hyperspectral image classification.
- This biologically inspired method provides a promising direction for advancing feature representation in HSI analysis.
