A low-cost UAV-edge-cloud system for agricultural straw burning monitoring: multi-target detection and event-based alerting

Crop residue burning damages air quality and affects agricultural environments. Small and short-duration events are difficult to detect through satellite observations or manual patrols. This study develops a low-cost UAV-Edge-Cloud system for real-time straw burning monitoring that combines local inference with event-based alert reporting. The system detects flame, smoke, and burned residue. These targets represent active combustion, early or smoldering combustion, and post-event evidence. A CBAM-enhanced YOLOv8s model improves detection of small flames. INT8-quantized CBAM-YOLOv8s is deployed on the MaixCAM-Pro camera, a low-cost edge platform capable of capturing RGB imagery and performing model inference locally. The deployed model achieves 0.864 mAP@0.5 and runs at 24.5 frames per second. In field tests, the system attained 91% image-level classification accuracy, a local response around 100 ms and a cloud reporting around 500 ms. This work demonstrates that practical deep learning-based agricultural fire monitoring is achievable on low-cost edge hardware, offering a basis for cost-effective environmental monitoring.