An AI-Driven Multimodal Sensing Framework Integrating UAV Imagery and Environmental Sensors for Intelligent Farmland Monitoring

. 2026 May 30;26(11):3456. doi: 10.3390/s26113456

Algorithm 1 Workflow of the Multimodal Target Perception Network

Input:
UAV image I, Environmental sensor data S, GPS coordinates P
Output:
Pest detection results $Y_{1}$ , Crop growth assessment $Y_{2}$
1:
Extract multi-scale visual features $F_{v}$ from I using convolutional extractors
2:
Embed S into environmental feature vectors $F_{s}$ via temporal encoding
3:
Map P into spatial representation $F_{p}$ using positional functions
4:
Perform cross-modal alignment: $F_{c m} = Align (F_{v}, F_{s}, F_{p})$ to bridge semantic gaps
5:
Apply multi-scale pyramid processing on $F_{c m}$ to handle scale variance
6:
Compute adaptive weights $w_{v}$ , $w_{s}$ , $w_{p}$ via dynamic gating mechanism
7:
Generate joint representation: $F_{joint} = w_{v} F_{v} + w_{s} F_{s} + w_{p} F_{p}$
8:
Input $F_{joint}$ into task-specific heads
9:
return $Y_{1}$ (Detection), $Y_{2}$ (Assessment)