Automatic titration detection method of organic matter content based on machine vision
Data files
May 27, 2025 version files 922.93 KB
-
code.zip
27.84 KB
-
README.md
2.27 KB
-
rgb_values.csv
892.82 KB
Abstract
This article proposes an automatic titration algorithm for organic matter content detection based on machine vision, which addresses the disadvantages of high risk factor, strong odor, significant pollution to laboratory environment, and slow efficiency of manual titration in organic matter detection. Firstly, by analyzing the color change characteristics during the titration process, machine learning techniques are used to classify the titration speed, and a titration experiment state recognition model is constructed to divide the titration speed into four categories and improve titration efficiency; Secondly, through a large number of titration experiments to collect relevant data and extract key feature parameters, an efficient titration algorithm based on histogram similarity was designed to accurately identify titration endpoints and improve detection accuracy. This study not only solves the limitations of manual operation in traditional titration methods, but also provides new ideas and methods for the automation and intelligence of chemical titration. The test results show that the titration error of the device is less than 0.2ml, and it has higher efficiency than manual titration, with good recognition rate and control accuracy.
Project Overview
This project aims to optimize the titration speed through machine learning models and establish an automatic titration algorithm by setting up an HSV model to accurately determine the endpoint. We used three different machine learning models: Decision Tree, K- K-nearest neighbor ( KNN ) , and support vector machine ( SVM ), and compared their performance. The specific instructions for how to reproduce are included in the README in each model folder.
Target
-Train and evaluate the performance of decision tree, KNN and SVM models on the titration state prediction task.
Compare the accuracy, precision, recall, and F1 score of these models.
-Select the best model to use for the HSV model in the automatic titration algorithm .
method
- Data preparation :
- Collect and preprocess titration experiment data.
- Split the dataset into training and testing sets.
- In this project, a file named “rgb_values.csv” (titration features labeled by titration speed/state, possibly indicating how fast the titration reaction is progressing (e.g., Fast Speed, Slow Speed, Medium Speed or End Point)) serves as an example of data collection. This file contains RGB values and related information obtained during titration experiments.
- Model training :
- Train decision tree, KNN and SVM models using training set data .
-
Performance Evaluation :
-Evaluate the performance of each model on the test set.
- Compare the models’ accuracy, precision, recall, and F1 score.
-
Model selection :
-Select the best performing model based on the evaluation results.
In conclusion
By training and evaluating the three machine learning models, we found that [ Decision Tree ] performed best on the titration state prediction task, with better accuracy, precision, recall, and F1 score than other models. Therefore, we will choose [Decision Tree] for the HSV model in the automatic titration algorithm .
Future Work
-Further tune the parameters of the selected model to improve performance.
-Explore the possibilities of other machine learning algorithms.
-Collect more data to improve the generalization ability of the model.