摘要
为解决自然环境下被遮挡目标检测精度低的问题,提出一种基于遮挡标记的目标检测算法,此方法对图像数据进行标注,分析被遮挡目标的图像特征损失,提出遮挡补偿机制;基于YOLO V3模型构建二分类器对目标和背景进行区分,从而完成目标检测任务.实验结果表明:改进后的YOLO V3模型在不同的遮挡目标个数占比下,平均检测精度均明显高于传统的YOLO V3模型,在自然环境下检测被遮挡目标,能达到更好的效果.
To solve the low detection accuracy of occluded objects in natural environment,an object detection algorithm is proposed based on occlusional labels.The occlusional labeling method is introduced to label the image data.The occlusional compensation mechanism is presented for the image feature loss of the occluded objects.The YOLO V3 model is used to construct the binary classifier to distinguish the object and the background to complete the detecting tasks.The experimental results show that the average detection accuracy of the improved model on the fruit dataset is significantly higher than the traditional YOLO V3 model and it can achieve better results to detect the occluded objects in the natural environment.
作者
帖军
宋威
尹帆
郑禄
杨欣
TIE Jun;SONG Wei;YIN Fan;ZHENG Lu;YANG Xin(College of Computer Science,South-Central University for Nationalities,Hubei Provincial Engineering Research Center for Intelligent Management of Manufacturing Enterprise,Wuhan 430074,China)
出处
《中南民族大学学报（自然科学版）》
CAS
2020年第3期302-308,共7页
Journal of South-Central University for Nationalities：Natural Science Edition
基金
湖北省技术创新专项重大项目(2019ABA101)
中央高校基本科研业务费专项资金资助项目(CZQ18016)
中南民族大学研究生学术创新基金资助项目(2019SYCXJJ118)。
关键词
目标检测
遮挡标记
补偿机制
object detection
occlusional label
compensation mechanism
