> initializing system...

Click or press any key to skip

← back to publications

YOLO-AgriNet: A Deep Learning-Based Model for Real-Time Plant Disease Detection in Precision Agriculture

Co-author · 2025

Journal of Computer and Communications, Vol.13 No.8

Real-time plant disease detection model (YOLOv8 extended with CBAM and ASPP), optimized for precision agriculture under low-resource tropical conditions. 84.5% mAP@0.5, 45 FPS, deployable on smartphones.

Details

  • Problem: visual disease detection by farmers is subjective and imprecise (e.g. only 40% accuracy for Fusarium wilt), especially in low-resource tropical regions like Cameroon.
  • Proposed model: YOLO-AgriNet, a YOLOv8 extension integrating attention modules (CBAM), a multi-scale context module (Atrous Spatial Pyramid Pooling), and an extra Stage Layer 5 to better detect small, early-stage symptoms.
  • Data: the public PlantDoc dataset (2,569 images, 13 species, 8,851 labeled instances, 30 disease categories) supplemented with 200 locally collected images from Cameroon.
  • Overall results: 84.5% mAP@0.5 (vs. 65.2% for YOLOv8), 45 FPS real-time inference, 78.8% recall, 83.1% F1-score.
  • Small objects (early symptoms): +24.6% mAP@0.5 over YOLOv8, and 46.7% fewer false positives.
  • Deployable on low-cost hardware: 32 FPS on an Android smartphone, under 500MB memory footprint, a 60% reduction in operational cost versus cloud-based solutions.