基于边缘检测的分液过程监测与终点识别

Process monitoring and end-point identification of unattended liquid phase separation based on edge detection

为解决实验室高耗时操作无人值守问题,提出一种复杂背景透明玻璃仪器内分液过程监测及终点识别算法。利用摄像头采集实验室分液操作画面,采用Canny算子进行边缘检测,获取分液操作的容器区域;为避免分液过程中红绿蓝(RGB)色彩空间多通道特征混合影响,将目标区域画面实时转化为灰度图像;通过全局灰度图像熵随时间变化趋势监测分液进程,并利用图像熵变化曲线最大值判断分液终点。结果表明:基于掩膜裁剪区域一维灰度图像熵终点识别方法与科学家肉眼判断时间平均绝对误差为0.80 s,判断过程更具备稳定性;与基于全画面算法对比,掩膜算法终点判别平均绝对误差降低19.50 s,相对误差降低90.96%,说明复杂背景对终点识别干扰显著;与基于RGB色彩空间图像熵相比,终点判别平均绝对误差降低7.10 s,相对误差降低89.87%,说明纹理信息在终点判别过程中起主要作用。

To address the problem of unmanned and time-consuming laboratory operations,a monitoring and endpoint recognition algorithm for the pipetting process inside a transparent glass instrument with complex background was proposed.The experimental pipetting operation was captured using a camera,and the Canny operator was used for edge detection to obtain the container area for the pipetting operation.To avoid the influence of the multi-channel feature mixing in red-green-blue(RGB)color space during the pipetting,the target area image was converted to a grayscale image in real time.The trend of global grayscale image entropy was monitored over time to detect the progress of the pipetting,and the maximum value of the entropy change curve was used to determine the endpoint of the pipetting.The results show that the endpoint recognition method based on the one-dimensional grayscale image entropy in the masked cutting area has an average absolute error time of 0.80 s compared to the naked eye judgment of a scientist,and the judgment process is more stable.Compared with the whole-image algorithm,the masked algorithm reduces the average absolute error of endpoint discrimination by 19.50 s,and the relative error is reduced by 90.96%,indicating that the complex background significantly interferes with endpoint recognition.Compared with the image entropy based on the RGB color space,the endpoint discrimination based on texture information reduces the average absolute error by 7.10 s and the relative error by 89.87%,indicating that the texture information plays a major role in endpoint discrimination.