Monitoring defoliation rate and boll-opening rate of machine-harvested cotton based on UAV RGB images

• Published in: European journal of agronomy Vol. 151; p. 126976
• Main Authors: Ma, Yiru, Chen, Xiangyu, Huang, Changping, Hou, Tongyu, Lv, Xin, Zhang, Ze
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2023
• Subjects: Boll-opening rate; Comprehensive evaluation of defoliation effect; Defoliation rate; Machine-harvested cotton; RGB image; UAV; MIC; PSO-ELM; Boll-opening rate; Cor; ELM; KRR; UAV; RMSE; RGB; MSR; GLCM; Comprehensive evaluation of defoliation effect; Machine-harvested cotton; RF; ANOVA; RGB image; Defoliation rate
• ISSN: 1161-0301; 1873-7331
• DOI: 10.1016/j.eja.2023.126976

Defoliation and artificial ripening are critical steps before mechanical harvesting of cotton. The defoliation effect is a key factor in determining the optimal harvesting time of cotton, as an improper machine-harvesting time can reduce the yield and quality of cotton. Therefore, rapid and accurate estimation of the defoliation rate and boll-opening rate is critical for evaluating the effect of defoliation on machine-harvested cotton. This study aimed to test whether high-resolution RGB images using UAVs, different characteristics screening methods, and modelling techniques can effectively monitor the defoliation rate and boll-opening rate for machine-harvested cotton. The hypothesis was that the combination of vegetation index, colour space component, and textural features, along with different methods, could improve the defoliation rate and boll-opening rate of machine-harvested cotton. Cotton defoliant treatment field experiments were conducted in Shihezi, Xinjiang, China, during 2019–2020, after defoliant treatment to acquire the defoliation rate and boll-opening rate. Correlation coefficient (Cor), maximum information coefficient (MIC), and random forest (RF) was used to screen the visible light vegetation indices, colour space parameters, and texture features. Multivariate stepwise regression (MSR), kernel ridge regression (KRR), extreme learning machine (ELM), and particle swarm optimized extreme learning machine (PSO–ELM) algorithms were used to construct monitoring models for the defoliation rate and boll-opening rate of machine-harvested cotton. Using the principal components analysis algorithm, comprehensive evaluation indices of defoliation effects were constructed to provide theoretical and methodological support for determining the time of machine-harvested cotton. The results showed that (1) a model based on RF_PSO-ELM achieved optimal assessment of the defoliation rate (R²=0.59, RMSE=19.37%, rRMSE=34.54%) and boll-opening rate (R²=0.73, RMSE=19.11%, rRMSE=46.40%), (2) a comprehensive evaluation index of the machine-harvested cotton defoliation effect was constructed based on the defoliation rate, boll-opening rate, and yield, which was named PCA1. (3) PCA1 was used as the standard to determine the harvesting time, when a PCA1 value of > 1.3225 indicated the optimal harvesting time. This index can accurately judge the best harvest time of machine-harvested cotton in large-scale production. •RF is more effective in screening features to monitoring defoliation and boll-opening rate.•The monitoring model constructed by PSO-ELM has relatively good accuracy.•Using model results can construct a defoliation effect comprehensive evaluation index.•The index of defoliation effect can fast judge the harvesting time of cotton.