基于改进YOLOv5的辣椒目标检测算法研究

Research on Chili Target Detection Algorithm Based on Improved YOLOv5

摘要: 在自然环境下，辣椒果实生长比较密集，目标检测算法很难区分被枝叶遮挡或被其他辣椒遮挡的情况，这对自动检测辣椒果实增加了困难，在本文中，我们采用现有的辣椒数据集，基于YOLOv5算法模型进行了辣椒图像识别。为了更好的对辣椒进行检测，减小因为枝叶遮挡和被其他辣椒遮挡而造成的检测难度，我们在原来的YOLOv5模型中加入了CBAM注意力机制，构建了CM-YOLO模型，我们的模型mAP可以达到93.1%，比原来的YOLOv5高0.4个百分点。我们所构建的CM-YOLO模型提高了对辣椒的检测能力，同样该模型也可以应用于农产品识别和分类等其他领域。

Abstract: In natural environments, the growth of chili peppers is relatively dense, making it difficult for ob-ject detection algorithms to distinguish between situations obstructed by branches and leaves or other chili peppers. This increases the difficulty of automatically detecting chili peppers. In this ar-ticle, we used existing chili pepper datasets and conducted chili image recognition based on the YOLOv5 algorithm model. In order to better detect chili peppers and reduce the detection difficulty caused by branch and leaf occlusion and being obstructed by other chili peppers, we added the CBAM attention mechanism to the original YOLOv5 model and constructed a CM-YOLO model. Our model’s mAP can reach 93.1%, which is 0.4 percentage points higher than the original YOLOv5 model. The CM-YOLO model we have constructed has improved the detection ability of chili peppers, and can also be applied to other fields such as agricultural product recognition and classification.