Answers are grounded in this paper and cite their sources.
In today's world, the use of hyperspectral image (HSI) has significantly increased in a variety of industries, from agricultural to environmental monitoring due to its capacity to capture rich spectral information. This detailed data allows us to detect subtle differences in materials. However, the classification performance of HSI suffers from the low availability of labeled sample data. This paper overcomes this challenge using transfer learning, harnessing the extensive knowledge of the ImageNet dataset, a cross-domain resource, to enhance our analysis of hyperspectral images (Indian Pine and Salinas). However, traditional pre-trained models like ResNet50 and VGG16 cannot utilize the HSI's rich spectral information due to their 3-channel constraint from ImageNet training. To overcome this, we propose modified ResNet50 and VGG16 architectures, enabling them to effectively extract features from multi-channel HSI. The overall accuracy achieved is 95.67% for Indian Pine and 99.58% for Salinas, respectively, with the modified VGG16 outperforming the existing pre-trained models.
No takes yet. Share an insight, caveat, or question.
Jannat et al. (2024) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: