Hydroponic lettuce defective leaves identification based on improved YOLOv5s

Area of Science:

• Agricultural Technology
• Computer Vision
• Machine Learning

Background:

• Ensuring the quality and value of hydroponic lettuce post-harvest is crucial.
• Accurate and efficient detection of defective leaves is a significant challenge.

Purpose of the Study:

• To improve the detection accuracy and efficiency of defective leaves in hydroponic lettuce.
• To develop an optimized deep learning model for this task.

Main Methods:

• An image acquisition system was designed for collecting images of defective lettuce leaves.
• The EBG_YOLOv5 model was proposed, optimizing YOLOv5 with ECA attention, bidirectional feature pyramid, and GSConv modules.
• Ablation and comparison experiments were conducted to validate the model's performance.

Main Results:

• EBG_YOLOv5 achieved higher Precision (0.1%), Recall (2.0%), and mAP0.5 (2.6%) compared to YOLOv5s.
• The model size, GFLOPs, and parameters were reduced by 15.3%, 18.9%, and 16.3%, respectively.
• EBG_YOLOv5 demonstrated superior accuracy and a smaller model size than other detection algorithms.

Conclusions:

• The EBG_YOLOv5 model significantly improves the detection of defective leaves in hydroponic lettuce.
• This technology provides a strong foundation for intelligent, non-destructive classification equipment for lettuce.