Bubble analysis during titration: A hybrid method study based on adaptive thresholding and watershed transformation
Abstract
In automated titrations in chemical experiments, syringe pump injection and rotor stirring often generate numerous random bubbles within the solution, obscuring true color and destabilizing features, leading to false endpoint detection. To address this issue, this paper proposes a hybrid segmentation framework combining adaptive thresholding and improved watershed filtering. First, illumination normalization and bilateral filtering are used as preprocessing methods. First, illumination normalization and bilateral filtering are used as preprocessing to mitigate uneven illumination. Then, an adaptive thresholding method is combined with bubble extraction, and a threshold difference feedback mechanism is introduced to ensure the accuracy of the threshold difference. Then, an adaptive thresholding method is combined with bubble extraction, and a threshold difference feedback mechanism is introduced to ensure convergence. Finally, optimized labeling is generated for the binary image and embedded in a Vincent submerged watershed to effectively mitigate over-segmentation. Experimental results show that this method achieves a PA of 0.97, IoU of 0.82, Dice of 0.9, Recall of 0.96, and Precision of 0.84, outperforming other segmentation methods. Ablation experiments verify the effective complementarity of these modules. This method provides robust, real-time technical support for accurate analysis of the segmentation methods. This method provides robust, real-time technical support for accurate analysis of titrations and other bubble-containing scenarios.