Timely crop disease detection, pathogen identification, and infestation severity assessments can aid disease prevention and control efforts to mitigate crop-yield decline. However, improved disease monitoring methods are needed that can extract high-resolution, accurate, and rich color and spatial features from leaf disease spots in the field to achieve precise fine-grained disease-severity classification and sensitive disease-recognition accuracy. Here, we propose a neural-network-based method incorporating an improved Rouse spatial pyramid pooling strategy to achieve crop disease detection against a complex background. For neural network construction, first, a dual-attention module was introduced into the cross-stage partial network backbone to enable extraction of multi-dimensional disease information from the channel and space perspectives. Next, a dilated convolution-based spatial pyramid pooling module was integrated within the network to broaden the scope of the collection of crop-disease-related information from images of crops in the field. The neural network was tested using a set of sample data constructed from images collected at a rate of 40 frames per second that occupied only 17.12 MB of storage space. Field data analysis conducted using the miniaturized model revealed an average precision rate approaching 90.15% that exceeded the corresponding rates obtained using comparable conventional methods. Collectively, these results indicate that the proposed neural network model simplified disease-recognition tasks and suppressed noise transmission to achieve a greater accuracy rate than is obtainable using similar conventional methods, thus demonstrating that the proposed method should be suitable for use in practical applications related to crop disease recognition.

Quantitative assessment of historical urban waterlogging on the Qinghai–Tibet Plateau (QTP) is severely hindered by the lack of early instrumental records. To bridge this data gap during the initial rapid urbanization period (1985–2003), this study proposes an integrated retrospective framework combining Large Language Models (LLMs)-based semantic mining, spatial reconstruction, and Extreme Gradient Boosting (XGBoost)- SHapley Additive exPlanations (SHAP) modeling under a Spatial Block Cross-Validation (SBCV) strategy. Historical disaster archives were transformed into spatially explicit training samples, enabling the reconstruction of a high-resolution urban waterlogging susceptibility atlas across the QTP. The results indicate that high-susceptibility areas are predominantly concentrated within urbanized river valleys and account for approximately 45% of the total urban built-up area across the QTP. The proposed framework achieved an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.9793 under the SBCV strategy, indicating good spatial transferability within the study area. SHAP analysis revealed that geomorphic variables contributed more strongly than most climatic variables, highlighting the important role of a geomorphic confinement effect in shaping susceptibility patterns. Comparative analyses further suggest a spatial transition from basin-dominated accumulation patterns to increasingly valley-confined susceptibility distributions under stronger topographic constraints. In addition, surface albedo and land surface temperature were identified as influential predictors, likely reflecting integrated thermal-hydrological conditions associated with antecedent soil moisture and local urban thermal dynamics. This study establishes a historical risk baseline for the QTP and provides a reproducible and cost-effective framework for historical hazard assessment in other data-scarce mountainous and high-altitude regions.