苹果采摘机械手图像数据集制作

李青斌，邱绪云^*，徐文鑫

山东交通学院汽车工程学院，山东 济南  250357

摘要：为获得用于苹果采摘机械手目标识别模型的苹果图片数据集，提出快速制作苹果采摘机械手图像数据集的方法，通过标注软件获取少量苹果图片数据集，采用图片数据集获得的目标检测模型作为识别模型标记苹果图片。经计算可知：获取的目标检测模型在验证集上的损失为6.803，满足识别苹果图片要求；该模型在标记苹果图片时的正确率为95.8%。由获取的苹果图片数据集训练的目标检测模型的平均识别精度为99.7%，满足采摘机械手的目标识别要求。

关键词：目标识别；目标检测模型；图片数据集；标注

Construction of an image dataset for apple-picking manipulator

LI Qingbin，QIU Xuyun^*, XU Wenxin

School of Automotive Engineering, Shandong Jiaotong University, Jinan 250357,China

Abstract:A method is proposed to quickly create an apple-picking manipulator image dataset for the goal of developing a target recognition model for the apple-picking manipulator. A small set of apple images is obtained through annotation software. The target detection model obtained from this image dataset is used to label more apple images to acquire a larger apple image dataset. The computed loss of the obtained target detection model on the validation set is 6.803, meeting the requirements for apple image recognition. The accuracy of this model in labeling apple images is 95.8%. The average recognition accuracy of the target detection model trained on the acquired apple image dataset is 99.7%, satisfying the requirements for the apple-picking manipulator′s target recognition.

Keywords:target recognition; target detection model; image dataset; label