An Iterative Pseudo Label Generation framework for semi-supervised hyperspectral image classification using the Segment Anything Model DOI Creative Commons
Zheng Zhao, Guangyao Zhou,

Qixiong Wang

и другие.

Frontiers in Plant Science, Год журнала: 2024, Номер 15

Опубликована: Дек. 23, 2024

Hyperspectral image classification in remote sensing often encounters challenges due to limited annotated data. Semi-supervised learning methods present a promising solution. However, their performance is heavily influenced by the quality of pseudo labels. This limitation particularly pronounced during early stages training, when model lacks adequate prior knowledge. In this paper, we propose an Iterative Pseudo Label Generation (IPG) framework based on Segment Anything Model (SAM) harness structural information for semi-supervised hyperspectral classification. We begin using small number labels as SAM point prompts generate initial segmentation masks. Next, introduce spectral voting strategy that aggregates masks from multiple bands into unified mask. To ensure reliability labels, design spatial-information-consistency-driven loss function optimizes IPG adaptively select most dependable These selected serve iterative SAM. Following suitable iterations, resultant can be employed enrich training data model. Experiments conducted Indian Pines and Pavia University datasets demonstrate even simple 2D CNN trained with our generated significantly outperforms eight state-of-the-art methods.

Язык: Английский

A robust two-stage framework for maize above-ground biomass prediction integrating spectral remote sensing and allometric growth model DOI

Mohan Yang,

Qiang Wu, Jianbo Qi

и другие.

Computers and Electronics in Agriculture, Год журнала: 2025, Номер 235, С. 110398 - 110398

Опубликована: Апрель 19, 2025

Язык: Английский

Процитировано

0
The Inversion of Rice Leaf Pigment Content: Using the Absorption Spectrum to Optimize the Vegetation Index DOI Creative Commons
Longfei Ma, Yuanjin Li,

Ningge Yuan

и другие.

Agriculture, Год журнала: 2024, Номер 14(12), С. 2265 - 2265

Опубликована: Дек. 11, 2024

The pigment content of rice leaves plays an important role in the growth and development rice. accurate rapid assessment is great significance for monitoring status This study used Analytical Spectra Device (ASD) FieldSpec 4 spectrometer to measure leaf reflectance spectra varieties during entire period under nitrogen application rates simultaneously measured content. leaf’s absorption were calculated based on physical process spectral transmission. An examination was conducted variations composition among distinct cultivars, alongside a thorough dissection interrelations distinctions between spectra. Based vegetation index proposed by previous researchers order invert content, spectrum replace original data optimize index. results showed that chlorophyll carotenoid contents different regular changes whole period, more obvious differences than After replacing absorptivity-sensitive bands (400 nm, 550 680 red-edge bands) with absorptivities would index, correlation which combines absorptivity reflectivity, significantly improved. model’s validation indicate inversion model, improved using spectra, outperforms traditional index-based model. this demonstrate potential spectroscopy quantitative crop phenotypes.

Язык: Английский

Процитировано

0
An Iterative Pseudo Label Generation framework for semi-supervised hyperspectral image classification using the Segment Anything Model DOI Creative Commons
Zheng Zhao, Guangyao Zhou,

Qixiong Wang

и другие.

Frontiers in Plant Science, Год журнала: 2024, Номер 15

Опубликована: Дек. 23, 2024

Hyperspectral image classification in remote sensing often encounters challenges due to limited annotated data. Semi-supervised learning methods present a promising solution. However, their performance is heavily influenced by the quality of pseudo labels. This limitation particularly pronounced during early stages training, when model lacks adequate prior knowledge. In this paper, we propose an Iterative Pseudo Label Generation (IPG) framework based on Segment Anything Model (SAM) harness structural information for semi-supervised hyperspectral classification. We begin using small number labels as SAM point prompts generate initial segmentation masks. Next, introduce spectral voting strategy that aggregates masks from multiple bands into unified mask. To ensure reliability labels, design spatial-information-consistency-driven loss function optimizes IPG adaptively select most dependable These selected serve iterative SAM. Following suitable iterations, resultant can be employed enrich training data model. Experiments conducted Indian Pines and Pavia University datasets demonstrate even simple 2D CNN trained with our generated significantly outperforms eight state-of-the-art methods.

Язык: Английский

Процитировано

0