摘要
针对果园环境中柑橘果实相互重叠和被枝叶遮挡,导致机器视觉识别柑橘果实与定位目标柑橘空间位置难度较大的问题,提出了一种基于YOLOv8-SAM的改进算法。通过增加BAM(Bottlenet Attention Module)注意力机制提高模型对被遮挡柑橘的识别准确率,运用SAM(Segment Anything Model)算法对被遮挡柑橘轮廓形状进行识别,并运用边缘检测法结合双目立体相机三维稠密深度点云得到被遮挡柑橘有效轮廓边,使用最小二乘法拟合出被遮挡柑橘的完整轮廓以确定目标柑橘果实更精确的空间坐标位置。试验结果表明:该算法可以准确识别并分离目标柑橘果实,同时更精确地定位柑橘果实空间坐标。改进的YOLOv8-SAM算法在果园环境中对被遮挡柑橘果实的识别平均精度达到91.1%,对被遮挡柑橘形心空间坐标的平均定位误差相比传统定位方法减少了16.22 mm,平均果径误差降低了7.99%,可为柑橘采摘机器人对重叠与被遮挡果实的准确识别提供参考。
In response to the challenges for machine vision to identify citrus fruit and locate the spatial position of target citrus in orchards due to overlapping fruit and occlusion by branches and leaves,a modified algorithm based on YOLOv8-SAM was proposed.The model s accuracy in identifying occluded citrus fruit was improved by adding BAM(Bottlenet Attention Module)attention mechanism.The contour shape of occluded citrus fruit was identified using SAM(Segment Anything Model)algorithm,and effective contour edges were obtained by combining edge detection with a binocular camera s 3D dense point cloud.The complete contour of the occluded citrus fruit was fitted using least squares to determine the more precise spatial coordinate position of the target citrus fruit.The experimental results show that the algorithm can accurately identify and separate the target citrus fruit,and more precisely locate the spatial coordinate of the citrus fruit.The average identification accuracy of the modified YOLOv8-SAM algorithm for occluded citrus fruit in the orchard environment is 91.1%,and the average spatial coordinate positioning error of the citrus fruit s center compared to traditional positioning methods is reduced by 16.22 mm,and the average fruit diameter error is reduced by 7.99%.This algorithm can provide reference for accurate identification of overlapping and occluded citrus fruit by citrus harvesting robots.
作者
王元昊
娄欢欢
罗红品
付兴兰
李光林
WANG Yuanhao;LOU Huanhuan;LUO Hongpin;FU Xinglan;LI Guanglin(College of Engineering and Technology,Southwest University,Chongqing 400715,China)
出处
《西南大学学报（自然科学版）》
CAS
北大核心
2025年第2期171-183,共13页
Journal of Southwest University(Natural Science Edition)
基金
国家自然科学基金项目(31971782)
重庆市科委产业化重点专项(cstc2018jszx-cyzdX0051)。
关键词
柑橘采摘
机器视觉
空间定位
轮廓重建
遮挡果实
图像处理
citrus picking
machine vision
spatial positioning
contour reconstruction
occluded fruit
image processing
