YOLO-K模型多目标检测算法研究

针对图像中目标过多、相似以及有遮挡的情况,提出了一种改进传统YOLO v3算法的新算法.首先,为了使改进的算法可以更准确的检测小尺度目标,在原算法的网络中增加浅层特征提取层,然后,为了提高大尺度目标的检测精度,在大尺度特征提取层上增加输出层,得到改进的YOLO-K模型,并通过数据集进行测试验证.实验结果表明,YOLO-...

Research On Multi-target Detection Algorithm Based On YOLO-K Model

In view of the situation that there are too many,similar and occluded objects in the image,the traditional YOLO v3 algorithm is accordingly improved.First,in order to make the improved algorithm more accurate in detecting small-scale targets,a shallow feature extraction layer was added to the network of the original algorithm.Then,in order to improve the detection accuracy of large-scale targets,an output layer was added to the large-scale feature extraction layer to obtain the improved YOLO-K model and verified by data set.The experimental results show that compared with YOLO v3 algorithm,the detection accuracy of YOLO-K is significantly improved when detecting objects with occluded,multi-target and dark images.