In order to detect rice false smut quickly and accurately, a lightweight false smut detection model, YOLOv8n-MBS, was proposed in this study. The model introduces the C2f_MSEC module to replace C2f in the backbone network for better extraction of key features of false smut, enhances the feature fusion capability of the neck network for different sizes of false smut by using a weighted bidirectional feature pyramid network, and designs a group-normalized shared convolution lightweight detection head to reduce the number of parameters in the model head to achieve model lightweight. The experimental results show that YOLOv8n-MBS has an average accuracy of 93.9%, a parameter count of 1.4 M, and a model size of 3.3 MB. Compared with the SSD model, the average accuracy of the model in this study increased by 4%, the number of parameters decreased by 89.8%, and the model size decreased by 86.9%; compared with the YOLO series of YOLOv7-tiny, YOLOv5n, YOLOv5s, and YOLOv8n models, the YOLOv8n-MBS model showed outstanding performance in terms of model accuracy and model performance detection; compared to the latest YOLOv9t and YOLOv10n models, the average model accuracy increased by 2.8% and 2.2%, the number of model parameters decreased by 30% and 39.1%, and the model size decreased by 29.8% and 43.1%, respectively. This method enables more accurate and lighter-weight detection of false smut, which provides the basis for intelligent management of rice blast disease in the field and thus promotes food security.

Rapid and accurate detection of rice foliar diseases is essential for yield prediction and food security. This study proposes a multi-size rice leaf disease detection model, YOLOv7-tiny, for fast and accurate detection of rice leaf diseases. The MobileNetV3 lightweight network is introduced to replace the backbone network of YOLOv7-tiny, which reduces the size of the model parameters and improves the extraction capability of features of different sizes; the RCS-OSA is used to replace the original ELAN-1 module, which improves the extraction capability of interlayer features; the TSCODE detector head is designed to enhance the extraction capability of the model for small targets; and the MPDIoU loss function is used to improve the model’s convergence speed and effect. The experimental results show that the average accuracy of ofYOLOv7-TMRTM is 97.9%, and compared with the baseline YOLOv7-tiny model, the accuracy of leaf spot detection is improved for different sizes and types of small target detection results, the YOLOv7-TMRTM model improves mAP0.5 by 4.4%, recall by 4.7% and precision by 8.8% compared to YOLOv7-tiny. The comparison with Faster RCNN, SSD, YOLOv4, YOLOv5s, YOLOv8s, and other mainstream target detection models shows that this method greatly solves the field environment. The problem of small spots and fuzzy edges of photographed rice diseases provides a basis for intelligent management of diseases in the field, which in turn promotes food security in China.

Rice leaf blast can develop rapidly under field conditions, creating a need for fast, non-destructive, and spatially explicit monitoring. UAV multispectral imagery and synchronized ground disease surveys were collected from artificially induced rice leaf blast experiments conducted in 2024 and 2025. After image registration, U-Net canopy segmentation, and ROI quality screening, 1801 curated 3 × 3-pixel canopy ROIs were retained. Each ROI was represented by four reflectance bands and 10 vegetation indices selected by Pearson correlation analysis, and the resulting dataset supported model development and spatial mapping. The improved 1DCNN-Transformer branch combined multi-scale Inception convolution, SE recalibration, and Focal Loss with RF and GBDT probability fusion. In the hold-out evaluation, the model achieved an overall accuracy of 98.90% and a weighted F1-score of 98.89%. Five repetitions of five-fold grouped cross-validation were performed, with identical 14-feature vectors constrained to the same fold. RF, GBDT, and equal RF-GBDT probability fusion achieved mean accuracies of 99.63%, 99.29%, and 99.33%, respectively, supporting strong class separability after duplicate-group isolation. The resulting severity and prescription maps provide an end-to-end digital workflow from canopy extraction to spatial decision support.