arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2005.03990cs.CVcs.LGeess.IV
收藏

基于深度学习的开心果检测与计数

Detecting and Counting Pistachios based on Deep Learning

Iran University of Science and Technology(伊朗科技大学);Sharif University of Technology(谢里夫理工大学)

Mohammad Rahimzadeh, Abolfazl Attar

摘要:

开心果是一种营养丰富的坚果,根据外壳的形状分为两类:开口和闭口。开口的开心果在价格、价值和需求方面都比闭口的开心果高。由于这些差异,生产公司精确计算每种数量是相当可观的。本文旨在提出一种新的系统,利用计算机视觉对不同类型的开心果进行计数。我们介绍并分享了一个新的开心果数据集,包括六个视频,总长度为167秒,3927个标记的开心果。与许多其他作品不同,我们的模型在视频中计算开心果,而不是图像。对视频中的对象进行计数需要在视频帧之间分配每个对象,使得每个对象被计数一次。我们工作中的两个主要挑战是开心果的咬合和开心果在不同框架中的变形的存在,因为在运输线上移动和滚动的开口开心果可能在某些框架中表现为闭合的口,而在其他框架中表现为开口。我们的新模型首先在RetinaNet对象检测器网络上进行训练,使用我们的数据集检测视频帧中不同类型的开心果。收集检测后,我们将其应用到一个新的计数器算法的基础上,一个新的跟踪器分配开心果在连续的帧具有高精度。我们的模型能够分配开心果,转动并改变它们的外观(例如,张着嘴的开心果看起来闭着嘴),这样就不会数错了。我们的算法执行非常快,并取得了良好的计数结果。我们的算法在六个视频(9486帧)的计算准确率为94.75%。

英文摘要:

Pistachios are nutritious nuts that are sorted based on the shape of their shell into two categories: Open-mouth and Closed-mouth. The open-mouth pistachios are higher in price, value, and demand than the closed-mouth pistachios. Because of these differences, it is considerable for production companies to precisely count the number of each kind. This paper aims to propose a new system for counting the different types of pistachios with computer vision. We have introduced and shared a new dataset of pistachios, including six videos with a total length of 167 seconds and 3927 labeled pistachios. Unlike many other works, our model counts pistachios in videos, not images. Counting objects in videos need assigning each object between the video frames so that each object be counted once. The main two challenges in our work are the existence of pistachios' occlusion and deformation of pistachios in different frames because open-mouth pistachios that move and roll on the transportation line may appear as closed-mouth in some frames and open-mouth in other frames. Our novel model first is trained on the RetinaNet object detector network using our dataset to detect different types of pistachios in video frames. After gathering the detections, we apply them to a new counter algorithm based on a new tracker to assign pistachios in consecutive frames with high accuracy. Our model is able to assign pistachios that turn and change their appearance (e.g., open-mouth pistachios that look closed-mouth) to each other so does not count them incorrectly. Our algorithm performs very fast and achieves good counting results. The computed accuracy of our algorithm on six videos (9486 frames) is 94.75%.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑