基于机器学习算法的牡蛎肥满度分类

Classification of plumpness in Crassostrea gigas based on the machine learning

中文关键词:  牡蛎  肥满度分类  数字图像处理  外部形态特征提取  机器学习  最大信息系数

英文关键词:Crassostrea gigas  plumpness classification  digital image processing  external morphology features extraction  machine learning  maximum information coefficient

中文摘要:

      目的 基于机器学习算法, 对同样重量范围下的牡蛎按照肥满度高低进行分类。方法 首先利用数字图像处理技术提取牡蛎外部形态特征, 获得牡蛎的粗糙度、伸长率、紧密度、长轴、短轴、面积等特征指标作为参数。然后利用机器学习算法在数据分析上的强大功能, 采用随机森林(random forest, RF)算法与梯度提升决策树(gradient boosting decision tree, GBDT)算法分别构建肥满度识别模型。最后, 将模型用于不同重量范围的牡蛎样本, 对牡蛎进行肥满度识别分类。结果 对于0~50 g的牡蛎, RF算法能取得较好的效果, 肥满度识别率达到79.3%, 50~100 g的牡蛎, GBDT算法的肥满度识别率达到86.4%。结论 相对于传统的按照重量对其肥满度分类而言, 本方法能够快速有效地识别出相同重量范围下牡蛎肥满度的高低, 为牡蛎分类提供了新的方法。

英文摘要:

      Objective To classify the Crassostrea gigas with the same weight range according to their plumpness based on machine learning algorithm. Methods Firstly, the external morphological features of Crassostrea gigas were extracted by digital image processing technology, and the roughness, elongation, compactness, major axis, minor axis and area of Crassostrea gigas were obtained as parameters. Secondly, with the powerful function of machine learning algorithm in data analysis, random forest (RF) and gradient boosting decision tree (GBDT) were used to build the plumpness degree recognition model respectively. Finally, the model as applied to Crassostrea gigas samples with different weight ranges to identify and classify Crassostrea gigas plumpness. Results For Crassostrea gigas of 0?50 g, RF algorithm could achieve good results, and the plumpness recognition rate was 79.3%, and for 50?100 g, the plumpness recognition rate of GBDT was 86.4%. Conclusion Compared with the traditional classification of its plumpness by weight, the method can quickly and effectively identify the plumpness of Crassostrea gigas in the same weight range, and provides a new method for Crassostrea gigas classification.