基于机器视觉技术的田间籽棉品级抽样分级模型研究

【目的】客观评价田间籽棉质量。【方法】依据中国籽棉品级分级标准，基于机器视觉技术选取棉花尺寸、色泽特征建立田间籽棉品级抽样分级模型。【结果】相关分析表明：亮度修正后，图像特征与籽棉品级之间相关显著。贝叶斯判别分析结果表明：基于10折交叉验证建立的籽棉品级判别模型的识别率在75.00％～92.86％之间，模型的平均识别率达83.20％。基于“1个标准误差”规则选取较好的贝叶斯判别模型，它在独立数据集上的泛化精度达89.11％，其中，前3级籽棉的识别率均达到100％。【结论】基于机器视觉技术识别籽棉品级是可行的，有利于提高籽棉品级抽样分级模型精度。

Researches of Grading Model of Field Sampling Cotton Based on Machine Vision Technology

In order to assess the quality of seed cottons objectively, sorting classifiers were designed based on machine vision technologies to grade 305 seed cottons with 7 grades based on their size and adjusted colors according to Chinese government grading standards. Fisher-criterion based canonical discriminants show that size and impurity contributed much more for cotton grades, and the distances among high-grades centroids were long while the ones among low-grades centroids were short. Total samples were divided into the train set and the test set. Cross-validation and Bayes-criterion based classifiers selections on the train set show that various classifiers were selected on 10-fold validation set with accuracies from 75% to 93%, and the approximate optimized classifiers were selected according to their average accuracy of 83%. Classifier performances evaluations on the test set show that the optimized classifiers can classify cottons into 7 grade categories with an accuracy of nearly 88%, and the high-grades cottons from 1 to 3 can be discriminated with an accurary of 100%. It is feasible to classify cotton grades using machine vision technologies and it helps to improve the yield of high-quality cottons.