基于YOLOv5对花生种子质量的识别研究

Identification of Peanut Seed Quality Based on YOLOv5

摘要: 针对传统人工种子筛选方式存在的不精准、不完全、效率低等问题。本文提出一种基于改进YOLOv5的花生种子质量的识别研究算法。首先，我们收集了几种不同质量类型的花生种子图片，利用labelImg工具制作数据集。其次，我们使用轻量级网络ShuffleNetV2取代了原始的骨干网络，在不降低精度的状况下提高网络运行速度。然后本文在YOLOv5颈部层引入高效通道注意力ECA来抑制不重要的特征信息，增强网络的特征提取能力。最后，将改进后的YOLOv5模型在自制的花生种子数据集进行实验。结果表明，所提模型与YOLOv4和原YOLOv5相比，mAP分别提升了2.7%和4.1%。另外该模型具有更好的检测速度，FPS为63帧/s。

Abstract: In response to the issues of imprecision, incompleteness, and low efficiency associated with tradi-tional manual seed screening methods. Aiming at the inaccuracy and incompleteness of traditional artificial seed screening methods, this paper proposes an algorithm for detecting the quality of peanut seeds based on the improved YOLOv5 algorithm. Firstly, we collected several images of peanut seeds of different quality types to create the dataset using the labelImg tool. Secondly, we replace the original backbone network with a lightweight network, ShuffleNetV2, to increase the network speed without decreasing the accuracy. The paper then introduces efficient channel atten-tion (ECA) into the neck layer of YOLOv5 to suppress unimportant feature information and enhance the feature extraction capability of the network. Finally, the improved YOLOv5 model is experi-mented on a homemade peanut seed dataset. The results show that the improved YOLOv5 network improves mAP by 2.7% and 4.1% compared to YOLOv4 and the original YOLOv5. The model has a better detection speed and reaches 65 frames per second (FPS).