A LIGHTWEIGHT DETECTION ALGORITHM FOR MULTIPLE WHEAT DISEASES BASED ON IMPROVED YOLOV11
TL;DR
The improved model achieves drastically reduced parameters and computational overhead, outperforming original lightweight YOLOv11 variants in detection accuracy, delivering a practical high-precision, low-latency solution for intelligent crop disease diagnosis in farmlands.
Abstract
Wheat planting areas face complex field environments, while embedded edge devices are restricted by limited computing power. Conventional detection models fail to balance recognition accuracy and real-time inference speed for multiple wheat diseases. This paper proposes a lightweight multi-disease detection algorithm based on improved YOLOv11. Taking YOLOv11s as the baseline, lightweight optimizations are conducted on the backbone, neck and detection head. A self-built wheat disease image dataset is constructed for model evaluation. The improved model achieves drastically reduced parameters and computational overhead, outperforming original lightweight YOLOv11 variants in detection accuracy. Ablation experiments verify the independent performance gain of each optimized module. After ONNX export and TensorRT FP16 quantization, the model is deployed on Raspberry Pi 5 and attains an ultra-high inference frame rate. The proposed lightweight algorithm addresses the challenge of real-time wheat disease detection on field edge terminals, delivering a practical high-precision, low-latency solution for intelligent crop disease diagnosis in farmlands.