基于改进的YOLOv5的蜜柚果树识别方法

Identification method of pomelo fruit tree based on improved YOLOv5

关键词:  YOLOv5  蜜柚果树识别  注意力机制

Key words:  YOLOv5  Pomelo fruit tree identification  attention mechanism

摘要:

摘要：【目的】为推动井冈蜜柚果园实现智能化喷药，提升资源利用效率，提出一种基于改进的YOLOv5的蜜柚果树检测方法。【方法】研究采用无人机M30T搭载CMOS影像传感器进行蜜柚果园的影像采集，共采集504张高分辨率图像，并使用LabelImg工具进行图像标注。研究基于NVIDIA 4090图形处理器配置深度学习环境，使用Pytorch1.7.1和CUDA 12.2框架，对YOLOv5目标检测算法进行改进，引入注意力机模块，提高蜜柚果树的检测精度和速度。【结果】在测试集上，改进的YOLOv5算法在蜜柚果树目标检测的准确性和速度方面表现出色，相较于SSD、YOLOv4以及YOLOv5等流行目标检测模型，改进的YOLOv5算法在平均检测精度（mAP）达到90.09%与检测速度（it/s）达到5.28it/s，相比于SSD模型、YOLOv4 模型和YOLOv5模型，精度分别提高了8.33%、12.74%、1.72%。【结论】提出了一种基于改进YOLOv5的深度学习模型，用于蜜柚果树的实时识别，提高了检测准确性和速度。基于深度学习的蜜柚果树识别模型，能够实现对蜜柚果树的精准检测，对井冈蜜柚果园的智能喷药的发展有重要意义。

Abstract:

Abstract:［Purpose］To promote the intelligent pesticide spraying in Jinggang pomelo orchards and enhance resource utilization efficiency, this study proposes a pomelo tree detection method based on an improved YOLOv5 algorithm.［Method］A DJI M30T drone equipped with a CMOS imaging sensor was employed to collect 504 high-resolution images of pomelo orchards. Image annotation was performed using the LabelImg tool. A deep learning environment was configured with an NVIDIA 4090 GPU, utilizing PyTorch 1.7.1 and CUDA 12.2 frameworks. The YOLOv5 algorithm was enhanced by integrating an attention mechanism to improve detection accuracy and speed.［Result］On the test dataset, the improved YOLOv5 model demonstrated superior performance in both accuracy and speed for pomelo tree detection. Compared to other mainstream models (SSD, YOLOv4, and original YOLOv5), the improved YOLOv5 achieved a mean average precision (mAP) of 90.09% and a detection speed of 5.28 iterations per second (it/s), surpassing the precision of SSD, YOLOv4, and original YOLOv5 by 8.33%, 12.74%, and 1.72%, respectively.［Conclusion］The proposed improved YOLOv5-based deep learning model enables real-time and precise identification of pomelo trees, significantly enhancing detection accuracy and speed. This approach holds substantial importance for advancing intelligent pesticide spraying technologies in Jinggang pomelo orchards, providing technical support for precision agricultural practices.