An image processing and machine learning solution to automate Egyptian cotton lint grading
An image processing and machine learning solution to automate Egyptian cotton lint grading
复制标题
DOI：
10.1177/00405175221145571
复制
发表时间：
2022-12-20
影响因子：
2.3
通讯作者：
Emaish, Haitham H.
中科院分区：
文献类型：
--
作者：
Fisher, Oliver J.;Rady, Ahmed;Emaish, Haitham H.
埃及棉花是埃及经济最重要的商品之一，以其质量享誉全球，主要通过人工检查进行评估和分级。这种分级有几个缺点，包括劳动要求高，检查效率低，以及受光线和人的主观性等检查条件的影响。这项工作提出了一种低成本的解决方案，用分类模型取代人工检查，使用电荷耦合设备相机捕获的图像来对埃及棉花皮棉进行分级。虽然这种方法已经被评估用于美国和中国陆地棉短纤维的分类，但它还没有在埃及棉花上进行测试，埃及棉花具有独特的特征和分级要求。此外，开发这些分类模型的方法已经扩展到包括图像处理技术，该技术消除了垃圾对颜色测量的影响，并提取了捕捉棉花样品的样本内方差的特征。评估了三种不同的监督机器学习算法：人工神经网络、随机森林和支持向量机。精度最高的模型(82.13-90.21%)使用了随机森林算法。模型的准确性受到与标记用于开发分类模型的棉花样本相关的人为错误的限制。无监督机器学习方法，包括k-均值聚类、层次聚类和高斯混合模型，被用来指示标记错误发生的位置。
Egyptian cotton is one of the most important commodities for the Egyptian economy and is renowned globally for its quality, which is largely assessed and graded by manual inspection. This grading has several drawbacks, including significant labor requirements, low inspection efficiency, and influence from inspection conditions such as light and human subjectivity. This work proposes a low-cost solution to replace manual inspection with classification models to grade Egyptian cotton lint using images captured by a charge-coupled device camera. While this method has been evaluated for classifying US and Chinese upland cotton staples, it has not been tested on Egyptian cotton, which has unique characteristics and grading requirements. Furthermore, the methodology to develop these classification models has been expanded to include image processing techniques that remove the influence of trash on color measurements and extract features that capture the intra-sample variance of the cotton samples. Three different supervised machine learning algorithms were evaluated: artificial neural networks; random forest; and support vector machines. The highest accuracy models (82.13-90.21%) used a random forest algorithm. The models' accuracy was limited by the human error associated with labeling the cotton samples used to develop the classification models. Unsupervised machine learning methods, including k-means clustering, hierarchical clustering, and Gaussian mixture models, were used to indicate where labeling errors occurred.