一种改进的YOLO V3目标检测方法

An Improved YOLO V3 Object Detection

目标检测是当今计算机视觉领域较为热门和流行的研究方向,在国防、安全和医疗保障等领域应用广泛。然而小目标的检测准确度一直不高,针对这一问题,提出了一种基于YOLO V3网络模型的改进方法,通过增强小目标的检测准确度来提高网络整体的检测成功率。由于小目标在图像中所占像素很少,经过多层卷积之后提取得到的特征不明显。改进方法通过将原网络模型中经2倍降采样的特征图进行卷积分别叠加到第二及第三个残差块的输入端,以此增强浅层特征信息。同时,在第一个8倍降采样的特征图后连接RFB模块,增强特征提取能力。用改进后的网络模型在PASCAL VOC数据集上与原网络进行对比实验。结果表明,改进之后的网络模型有效提高了小目标的检测准确率。

Object detection is a hot and popular research direction in the field of computer vision,which is widely used in the fields of national defense,security and medical security. However,the detection accuracy of small targets is not high. To solve this problem,we propose an improved method based on YOLO V3 network model,which improves the detection accuracy of the whole network by enhancing the detection accuracy of small targets. Because the small target occupies very few pixels in the image,the features extracted after the multi-layer convolution are not obvious. The improved method enhances the shallow feature information by convolving the feature maps of the original network model by two times down sampling onto the input ends of the second and third residual blocks respectively. In addition,the RFB module is connected after the first 8 times down sampling feature map to enhance the feature extraction ability. The improved network is compared with the original network on the PASCAL VOC data set. The results show that the improved network effectively improves the detection accuracy of small targets.