关键词: 棉纤维, 马克隆值, 等级预测, 公检指标, 智能检验

Abstract: The micronaire reflects the fineness and maturity of cotton fibers. Research shows that the maturity level affects the physical properties of cotton fibers, and the micronaire also has a strong correlation with other quality indicators of cotton fibers. Although cotton fiber inspection has gradually become instrumented, there are many indicators, and the process is complex. To make full use of the public inspection data, simplify the inspection process, and improve inspection efficiency, this paper considers the potential linear or nonlinear relationship between the physical performance indicators of cotton fibers and studies a model that reflects the micronaire with other indicators.
This paper first preprocesses the collected data, performs descriptive statistical analysis, and determines the maximum and minimum values in the normalization process. Then, it uses Adaboost, LightGBM, and GBDT algorithms to perform feature selection on the indicators and analyze the importance level. Since there are differences in the analysis results of different methods on each indicator, this paper establishes a matrix to comprehensively analyze the selection results and finally determines that 9 indicators are involved in the establishment of the micronaire grade prediction model. These 9 indicators are Rd, +b, impurity particle number, impurity area percentage, upper half average length, length uniformity index, breaking strength ratio, breaking elongation ratio, and short fiber rate. Finally, this paper uses decision tree, random forest, and LightGBM algorithms to establish the micronaire grade model, and obtains the final result of the model through the evolution process of adjusting parameters and other methods. By comparing the results of the three models, this paper finds that LightGBM has the best result for the micronaire grade prediction.
This paper is to apply the LightGBM algorithm to the micronaire grade prediction of cotton fibers, explore the correlation of multiple physical indicators of cotton fibers by data mining methods, use Adaboost, LightGBM, and GBDT methods to comprehensively determine the 9 indicators as the basic indicators for the micronaire grade prediction, and establish a prediction model with a verification accuracy of 85.7%, which provides a theoretical reference for the intelligent inspection of cotton fibers. The follow-up work can further optimize the cotton fiber inspection indicators, use fewer indicators to achieve the micronaire grade prediction, or choose multiple nonlinear algorithms to analyze and compare the indicators, and further improve the accuracy of the micronaire grade prediction.

Key words: cotton fiber, micronaire, grade prediction, inspection indicators, intelligent inspection