Abstract:

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.

Key words: edge detectopm, liquid phase separation, end-point recognition, image entropy, Canny edge detector