基于SSD神经网络的图像自动标注及应用研究

Automatic Image Annotation and Applied Research based on SSD Deep Neural Network

针对目前图片标注成本高的问题,提出了融合自动标注半监督学习的协同训练算法:先基于SSD算法训练图片检测模型,融合半监督学习算法,在用自动标注的图片训练模型时叠加手动标注的数据,最终生成目标检测自动标注模型。实验结果表明,模型在经过6次迭代后自动标注生成的位置坐标与手动标注的真实位置坐标之间的平均IoU达到了80%以上,测试结果说明该算法在实际应用中有较大的应用前景。

At present,image annotation costs a lot of manpower.Aiming at the above problems,this paper proposes a collaborative training algorithm that integrates automatic labeling and semi-supervised learning.First,the image detection model is trained based on SSD algorithm.Then,the algorithm idea of semi-supervised learning is integrated.When training with the automatically annotated images,part of manually annotated data is superimposed at the same time,finally generate a usable target detection model.The experimental results show that after 6 iterations,the IoU between the position coordinates generated by automatic annotation and the real position coordinates manually annotated reached more than 80%.The test results show that this algorithm has a great application prospect in practical application.