摘要
小目标检测算法通常难以达到理想的检测精度.为此,在分析常规算法基础之上对Faster-RCNN算法网络结构进行修改,得到了面向小目标的多尺度区域神经网络检测算法,该方法是利用高层特征和低层特征相互弥补的方法提升小目标的检测精度.实验结果表明,通过利用高层特征和低层特征有针对性地对图像进行信息提取,能有效提高小目标检测的准确度.
Small object detection algorithm is often difficult to achieve the desired detection accuracy. For this reason,the Faster-RCNN algorithm network structure is modified based on the analysis of conventional algorithms,and a multi-scale area neural network detection algorithm for small targets is obtained. This method uses high-level features and low-level features to complement each other to improve small targets The detection accuracy. Experimental results show that by using high-level features and low-level features to extract information from images in a targeted manner,the accuracy of small target detection can be effectively improved.
作者
易星
YI Xing(College of Information Engineering,Shenyang University of Chemical Technology,Shenyang 110142,Liaoning,China)
出处
《喀什大学学报》
2021年第3期65-69,共5页
Journal of Kashi University
