先验知识指导生成虚拟样本在指针式仪表识别上的应用

为解决现有指针式仪表识别方法依赖于预处理的有效性且泛化能力不足的问题,提出一种基于深度卷积神经网络与虚拟样本结合的识别方法.该方法利用深度卷积神经网络自适应地提取仪表图像关键特征,避免无关信息的干扰;采用先验知识构建指针式仪表虚拟样本生成模型,解决深度卷积神经网络面临的小样本难题.仿真数据、实验数据和现场实际应用结果表明,文中方法是可行有效的,且比传统的指针定位方法识别效果更好,尤其在更换仪表、局部信息缺失等复杂情况下具有很好的鲁棒性.

Generating Virtual Samples Based on Prior Knowledge in Pointer Meter Recognition

In order to solve the problems that the existing method of pointer meter recognition depended on the validity of preprocessing and the lack of generalization ability,a hybrid method of automatic recognition on pointer meters combining deep convolutional neural networks with virtual samples is proposed.The deep convolutional neural networks are used to adaptively extract the key features of the instrument image to avoid the interference of irrelevant information.The prior knowledge is used to construct the virtual sample generation model of the pointer meter to solve the small sample problem faced by the deep convolutional neural networks.Simulation data,experimental data and practical application were applied to validate the method.The results show that the method achieves better effect and good robustness with its application on different instruments under various scenes.