Accurate field-scale crop disease detection is crucial for precise decisions and for highly efficient multi-scale collaboration. UAV-based multispectral imaging technology offers advantages in terms of high efficiency and low cost. Deep learning shows potential for deep representation and fusion of spectral and spatial features. However, traditional manual disease surveys are limited by efficiency and cost, making it difficult to meet the large sample sizes required by deep learning. Therefore, we proposed a method for rice bacterial leaf blight detection using UAV-based multispectral imagery. This method integrates a cross-scale sample-label transfer, and a spectral–spatial dual-branch feature fusion architecture (DualRiceNet). We first used RTK positioning to transfer disease labels from near-ground RGB images to high-altitude multispectral images, effectively expanding the dataset and alleviating the scarcity of labeled samples. DualRiceNet employed a cross-attention mechanism to couple its spectral and spatial branches, thereby isolating disease-specific spatial–spectral patterns from complex interference from the farmland background. DualRiceNet achieved an overall accuracy (OA) of 92.3% on the same-distribution test set. In an independent scenario test set spanning multiple differences in geography, time, phenology, and variety, the model maintained the highest OA of 80.0%. Our method demonstrated an excellent generalization ability to real-world environmental variations in rice fields.

Precise spatiotemporal assessment of habitat suitability is essential for crop pest and disease risk warning and food security. However, most existing approaches focus on disease occurrence, overlook spatial heterogeneity and time series information, and therefore, struggle to capture the habitat dynamics from occurrence to epidemic. We propose a dynamic framework for rice sheath blight (RSB) habitat suitability assessment that integrates rice phenology and disease time series to reveal fine-grained intra-annual spatiotemporal variability via a spatiotemporally adaptive strategy beyond the reach of traditional static models. Remote sensing and meteorological time series data, together with RSB survey data and crowdsourced records from southern China, are integrated in this study. First, the study area is partitioned into sub-regions and sensitive time windows (STWs) based on climate and rice-cropping systems. MaxEnt, combined with natural breaks, is then used to assess RSB occurrence suitability and delineate multi-level suitable areas. Geographical and temporal weighted regression (GTWR) and the coefficient of variation (CV) are finally applied within moderate-to-high occurrence-suitability areas to reconstruct time series of intra-STW epidemic potential dynamics and quantify their temporal variation. Results indicate the optimal phenology-based scheme yields one STW for single-cropping sub-regions and three STWs for double- and mixed-cropping sub-regions. MaxEnt AUC ranges from 0.610 to 0.768, with natural-break thresholds at 0.312 and 0.473. GTWR produces generally robust intra-STW fits (most local R² > 0.4), and the CV highlights localized, time-varying high-fluctuation zones within occurrence-suitable areas that static maps do not reveal. Overall, our method extends crop disease habitat suitability assessment from a static paradigm to a spatiotemporally adaptive, dynamic one, providing useful habitat background constraints for monitoring, early warning, and forecasting of crop pests and diseases under complex cropping systems.