Answers are grounded in this paper and cite their sources.
Wheat yellow rust is one of the most destructive diseases threatening global wheat production. The pathogen can persist as active sporulating pustules during the early growth stages, later forming parallel rows of yellow pustules on the leaves. Continuous canopy-level monitoring of host-pathogen interaction is therefore essential to curtail its spread. Hyperspectral-image (HSI) time series, which combine high spatial resolution with dense, narrow-band spectra, offer a promising means of capturing host–pathogen interactions at scale. However, the disease-specific signals are heterogeneous in space, spectrum, and time, and they are displayed as subtle, low-amplitude variations within the HSI time-series. How to extract the subtle host-pathogen interaction signals from the redundancy of HSI time-series is challenging. In this study, we introduce a spectral–temporal–spatial Capsule Network (STS-CapsNet), a biology-informed deep-learning framework to disentangle the disease canopy spectral, temporal, and spatial features from HSI time series. The network comprises 1) a dual-branch encoder that extracts spectral–temporal and spatial–temporal features, and 2) a class-capsule decoder that bundles these scalar responses into biologically meaningful vector capsules, thereby modelling part-to-whole relationships among canopy physiological traits. Experiments demonstrate that STS-CapsNet can reveal host–pathogen interactions before visible lesions emerge. At 7 th day after infestation (DAI), the model achieves an overall accuracy (OA) of 86.6%, rising to a maximum OA of 97.4%. To our knowledge, this is the first study to interpret canopy-level yellow-rust dynamics from AI-enhanced remote-sensing observations, and the method shows strong potential for extension to other plant diseases. • Introduce a biology-informed capsule network for hyperspectral time-series analysis. • Dual-branch encoder captures biochemical and structural canopy dynamics linked to yellow-rust infection. • Capsule decoder builds part-to-whole hierarchies. • Achieves early lesion detection at 7 DAI and a peak overall accuracy of 97.4%.
No takes yet. Share an insight, caveat, or question.
Bi et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: