基于多源数据与神经网络的病害识别方法、系统、设备及介质
By using dynamic time warping and multi-attention weight interpolation algorithms to perform timestamp alignment and spatial overlap verification on multi-source data, and combining 3D convolutional networks and deep neural networks for feature fusion, the problem of timestamp deviation and sparse feature completion in crop disease identification of multi-source data is solved, improving the accuracy and positioning precision of disease identification, and is suitable for disease control in facility agriculture.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- WUXI UNIV
- Filing Date
- 2026-03-09
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies for crop disease identification suffer from large timestamp biases in multi-source data, low spatial overlap, large errors in sparse feature completion, and poor feature correlation, resulting in low accuracy in early lesion identification and difficulty in distinguishing similar diseases. Furthermore, they are poorly adaptable to facility agriculture environments.
The dynamic time warping algorithm is used to align timestamps and verify spatial overlap of multi-source data. A multi-attention weight interpolation model is constructed to complete sparse features. Features are then fused using a three-dimensional convolutional network and a deep neural network to generate a multi-channel three-dimensional feature tensor for disease classification and lesion localization.
It achieves high consistency of multi-source data in time and space, accurately completes sparse features, improves the sensitivity of early lesion identification and the ability to distinguish different disease types, and provides accurate three-dimensional positioning coordinates, providing reliable guidance for precision pesticide application in facility agriculture.
Smart Images
Figure CN121810683B_ABST