Sweet potato virus disease monitoring method based on unmanned aerial vehicle multispectral semantic segmentation
By constructing a multi-source prior-guided sweet potato image dataset and asynchronous hierarchical feature extraction, combined with a difference-aware gating fusion module and a modal reliability calibration cross-attention module, the problem of low disease identification accuracy in traditional monitoring methods was solved, and accurate monitoring and quantitative assessment of sweet potato viral diseases were achieved.
CN122391934APending Publication Date: 2026-07-14XUZHOU NORMAL UNIVERSITY
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XUZHOU NORMAL UNIVERSITY
- Filing Date
- 2026-05-18
- Publication Date
- 2026-07-14
Smart Images
Figure CN122391934A_ABST
Abstract
The application discloses a sweet potato virus disease monitoring method based on unmanned aerial vehicle multi-spectral semantic segmentation, and first constructs a spectral prior prompt module integrated with a multi-spectral vegetation index, so that explicit injection of multi-source physiological prior information is realized; then a double-branch hierarchical architecture named INXMRCANet is built, spatial features are extracted by using an InceptionNeXt convolution improved network, and physiological features are synchronously extracted by using a spectral spatial encoder; the core lies in that a modal reliability calibration cross-attention module is designed, feature alignment is performed by predicting pixel-level offset, weights of each modal and a difference compensation branch are dynamically calculated by using a reliability estimator, and a window cross-attention mechanism is guided to perform feature refining.
Need to check novelty before this filing date? Find Prior Art