棉花学报2025，Vol.37Issue(2)：94-105,12.DOI:10.11963/cs20240051
基于RGB图像和随机森林算法的棉种识别
Identification of cultivated cotton species based on RGB images and random forest algorithm
摘要
Abstract
[Objective]Accurate identification of cotton species is of great significance for breeding,cultivation management,and pest control.However,the traditional manual identification method is subjective and inefficient.Therefore,this study aims to develop a rapid classification model based on red,green,and blue(RGB)image and random forest(RF)algorithm to realize automatic recognition of cotton species.[Methods]In this study,Gossypium herbaceum,G.arboreum,G.barbadense,and G.hirsutum lines were planted to collect the RGB images of cotton leaves at the squaring stage and the flowering and boll-setting stages,then the color and morphological feature parameters were extracted.Based on the extracted features,three RF models were constructed:one using only leaf features at the squaring stage,another using only leaf features at the flowering and boll-setting stages,and a comprehensive model combined features from both stages.The classification performance of each model was evaluated,and the key features affecting cotton species were identified through feature importance analysis.To assess the superiority of the RF model,the classification effect of the support vector machine(SVM)and K nearest neighbor(KNN)algorithm was conducted for comparison.[Results]The classification model combining the leaf features of the squaring stage and the flowering and boll-setting stages had the highest accuracy,with an overall accuracy of 97.71％and a Kappa coefficient of 0.95,which was superior to the model based only on leaf features from a single growth stage.Feature importance analysis showed that leaf area and roundness played an important role in cotton species recognition.Additionally,the RF model demonstrated better classification performance than SVM and KNN,exhibiting higher stability and accuracy.[Conclusion]The cotton species identification method based on RGB images and the RF algorithm proposed in this study does not require complex image pre-processing and can provide new insights and technical support for crop precision management and the application of machine learning algorithms in agriculture.关键词
RGB图像/随机森林算法/分类/棉花Key words
RGB image/random forest algorithm/classification/cotton引用本文复制引用
王亚茹,韩迎春,雷亚平,杨北方,熊世武,焦亚辉,马云珍,李亚兵,支晓宇..基于RGB图像和随机森林算法的棉种识别[J].棉花学报,2025,37(2):94-105,12.基金项目
国家自然科学基金(3230151245) （3230151245）
中央级公益性科研院所基本科研业务费专项(1610162023051) （1610162023051）
