摘要
为了保障化工厂数显式仪器设备的正常运行及故障及时检测、减少人力巡检的工作量,达到化工厂安全生产的要求,需要对带有数显式仪表的设备进行准确的读数检测。但化工厂仪表盘表面颗粒粗糙、灰尘堆积造成采集到的图像存在大量噪点,加之由于巡检机器人采集图像过程中的位置偏移,造成图像扭曲,对仪表数值的读取产生很大的影响,因此需要提高数显式仪表的数值识别精度。通过构建Hopfield三神经元网络模型进行仿真,并进行随机噪声检验,发现噪声水平在0.1及以下的识别效果最佳。通过对图像进行灰度化、降噪等预处理操作消除噪声,然后对仪表盘的数值区域进行定位、分割、校正、识别,实验结果表明,数值区域的平均识别准确率为90.15%,满足化工厂巡检机器人对数显式仪表的识别要求。
In order to ensure the normal operation and timely detection of digital instrumentation equipment in chemical plants, reduce the workload of human inspection, and meet the requirements of safe production in chemical plants, it is necessary to perform accurate reading detection on equipment with digital display instruments. However, the roughness of the instrument panel surface of the chemical plant and the accumulation of dust cause a lot of noise in the collected image. In addition, due to the positional deviation of the inspection robot during image collection, the image is distorted, which has a great influence on the reading of the instrument value. Therefore, it is necessary to improve the numerical recognition accuracy of digital display instruments. Through the construction of Hopfield three neuron network model for simulation and random noise test, it is found that the recognition effect with noise level of 0.1 or below is the best. By preprocessing operations such as graying and noise reduction on the image, the noise is eliminated, and then the numerical area of the instrument panel is positioned, segmented, corrected and identified. The experimental results show that the average recognition accuracy of the numerical area is 90.15%, which meets the requirements of chemical industry. Factory inspection robots&apos, requirements for the identification of digital display instruments.
作者
蔡佩征
牟星辰
Cai Peizheng;Mu Xingchen(College of Automation and Electronic Engineering,Qingdao University of Science and Technology,Qingdao 266000,China)
出处
《电子测量技术》
2020年第19期107-111,共5页
Electronic Measurement Technology
