摘要
在硬糖制造过程中,由于糖浆温度不稳定、搅拌不均等原因导致硬糖中出现气泡缺陷,影响其外观和口感,因此需要对含气泡的硬糖进行检测剔除。针对糖果气泡缺陷小且位置不固定的问题,提出了一种基于YOLOv7的糖果气泡缺陷识别算法。首先,采集糖果的彩色图像和深度图像,并对其彩色图像进行直方图均衡化和中值滤波,构建糖果气泡缺陷数据库;然后,构建基于YOLOv7的气泡缺陷识别模型,采用SimOTA和ReOrg操作提高模型运算速度和识别精度;最后,将构建的糖果气泡缺陷识别模型与现有的其他深度学习目标检测算法进行对比,实验结果表明所提出的识别模型综合性能最优。
During the manufacturing process of hard candies,problems such as unstable syrup temperatures and uneven stirring can result in the formation of air bubble defects in the candies,which can negatively impact their appearance and taste.Therefore,it is necessary to detect and remove hard candies that contain air bubbles.To tackle the issue of small and randomly positioned air bubble defects in candies,this paper introduces a candy air bubble defect recognition algorithm based on YOLOv7.Firstly,colored images and depth images of the candies are collected.Histogram equalization and median filtering are then applied to the colored images in order to establish a database of candy air bubble defects.A YOLOv7-based model for recognizing air bubble defects is constructed,incorporating SimOTA and ReOrg operations to achieve a balance between processing speed and recognition accuracy.Finally,the constructed model for recognizing candy air bubble defects is compared with existing deep learning object detection algorithms.Experimental results demonstrate that the proposed recognition model exhibits the best overall performance.
作者
方晓东
朱婷婷
吴旻
李星佑
薛胜
FANG Xiao-dong;ZHU Ting-ting;WU Min;LI Xing-you;XUE Sheng(School of Mechatronic Engineering,Nanjing Forestry University,Nanjing Jiangsu 210037,China)
出处
《林业机械与木工设备》
2024年第3期19-23,共5页
Forestry Machinery & Woodworking Equipment
