Modeling of spectral monitoring of cotton aphid population based on feature bands
Cotton aphid is the most common and harmful insect in the process of cotton growth and development, and it is the most serious pest that restricts the high yield and high quality of cotton in China and even in the world. The field-scale experiments were conducted for two years to estimate the popul...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
AcademicPres
2024-12-01
|
| Series: | Notulae Botanicae Horti Agrobotanici Cluj-Napoca |
| Subjects: | |
| Online Access: | https://www.notulaebotanicae.ro/index.php/nbha/article/view/14099 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Cotton aphid is the most common and harmful insect in the process of cotton growth and development, and it is the most serious pest that restricts the high yield and high quality of cotton in China and even in the world. The field-scale experiments were conducted for two years to estimate the population of cotton aphids model on cotton leaves using remote sensing. Hyperspectral data from single leaves and aphid populations were obtained at different growth stages of various cotton varieties. Preprocessing techniques including savitzky-golay (SG), first derivative (FD), second derivative (SD), logarithm transformation (LG), and reciprocal transformation (RG) were applied to the original spectral data. The successive projections algorithm (SPA) combined with the Pearson correlation coefficient (Pearson) method were used to construct three types of hyperspectral monitoring models: linear regression (LR), Extreme gradient boosting (XGBoost), and partial least squares regression (PLSR). Results indicated that LG significantly improved model accuracy, while SPA effectively reduced the number of bands required for analysis. Among the three models constructed using selected feature bands, the XGBoost model outperforms LR and PLSR models in terms of prediction accuracy. LG-Person-XGBoost identified ten feature bands with a coefficient of determination (R2), root mean square error (RMSE), and relative percent deviation (RPD) values reaching 0.76, 65.74 heads/leaf, and 0.91 respectively for the modeling set; whereas predicted R2, RMSE, and RPD values were found to be 0.36, 57.62 heads/leaf, and 0.99 respectively. LG-SPA-XGBoost achieved superior prediction accuracy by selecting eight feature bands with R2, RMSE, and RPD values reaching 0.87, 46.7 heads/leaf, and 1.43 respectively for the modeling set; whereas predicted R2, RMSE, and RPD values were found to be 0.77, 42.66 heads/leaf, and 1.27 respectively. This indicates that the hyperspectral remote sensing model can be used to estimate the population of cotton aphids on cotton leaves based on the selection of feature bands providing are reference for the nondestructive monitoring of cotton aphids in cotton fields and offering a significant supplement to the traditional methods of crop pest quantity monitoring.
|
|---|---|
| ISSN: | 0255-965X 1842-4309 |