Apple Target Detection Method in Different Scenarios Based on YOLOv3
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
维普
万方数据
苹果采摘机器人依赖于精准的苹果目标检测来达到智能采摘.然而苹果数量、光照强度、生长阶段等条件使得苹果检测变得复杂,为此以苹果为试材,采用基于YOLOv3(You Only Look Once version 3)的目标检测方法,研究了不同场景下的苹果目标检测效果.针对苹果的外形特征,调整了 YOLOv3的锚点尺寸以优化苹果在不同尺度上的变化.为了验证模型有效性,试验以苹果数量、拍摄不同光照的角度、苹果不同生长阶段为变量,对所提出的算法与 YOLOv2(You Only Look Once version 2)、SSD(Single Shot MultiBox Detector)2种现有算法在目标识别效果进行了比较分析.结果表明:在不同条件下,所提出的方法在苹果目标检测的F1值上均达到了 88％以上,可为苹果采摘机器人的开发提供技术支持.
Apple-picking robots relied on precise apple target detection to achieve intelligent picking.However,conditions such as the number of apples,light intensity,and growth stages made apple detection complex.To address this,apples were used as test materials,and the YOLOv3(You Only Look Once version 3)object detection method was employed to study the effectiveness of apple target detection in different scenarios.The anchor sizes of YOLOv3 were adjusted to optimize the variations in apple size according to the specific characteristics of apple shapes.To verify the effectiveness of the model,experiments were conducted using variables such as the number of apples,different lighting angles,and different growth stages of apples.The proposed algorithm was compared and analyzed with two existing algorithms,YOLOv2(You Only Look Once version 2)and SSD(Single Shot MultiBox Detector),in terms of target recognition effectiveness.The results indicated that under different conditions,the proposed method achieved F1 scores of over 88％in apple target detection,providing technical support for the development of apple-picking robots.
维普
万方数据
