Background Unmanned aerial vehicle (UAV)-based remote sensing provides a flexible, low-cost, and efficient approach to monitor crop growth status at fine spatial and temporal resolutions, and has a high potential to accelerate breeding process and improve precision field management. Method In this study, we discussed the use of lightweight UAV with dual image-frame snapshot cameras to estimate aboveground biomass (AGB) and panicle biomass (PB) of rice at different growth stages with different nitrogen (N) treatments. The spatial–temporal variations in the typical vegetation indices (VIs) and AGB were first investigated, and the accuracy of crop surface model (CSM) extracted from the Red Green Blue (RGB) images at two different stages were also evaluated. Random forest (RF) model for AGB estimation as well as the PB was then developed. Furthermore, variable importance and sensitivity analysis of UAV variables were performed to study the potential of improving model robustness and prediction accuracies. Results It was found that the canopy height extracted from the CSM (Hcsm) exhibited a high correlation with the ground-measured canopy height, while it was unsuitable to be independently used for biomass assessment of rice during the entire growth stages. We also observed that several VIs were highly correlated with AGB, and the modified normalized difference spectral index extracted from the multispectral image achieved the highest correlation. RF model with fusing RGB and multispectral image data substantially improved the prediction results of AGB and PB with the prediction of root mean square error (RMSEP) reduced by 8.33–16.00%. The best prediction results for AGB and PB were achieved with the coefficient of determination (r 2 ), the RMSEP and relative RMSE (RRMSE) of 0.90, 0.21 kg/m 2 and 14.05%, and 0.68, 0.10 kg/m 2 and 12.11%, respectively. In addition, the result confirmed that the sensitivity analysis could simplify the prediction model without reducing the prediction accuracy. Conclusion These findings demonstrate the feasibility of applying lightweight UAV with dual image-frame snapshot cameras for rice biomass estimation, and its potential for high throughput analysis of plant growth-related traits in precision agriculture as well as the advanced breeding program. Electronic supplementary material The online version of this article (10.1186/s13007-019-0418-8) contains supplementary material, which is available to authorized users.
Remote estimation of flower number in oilseed rape under different nitrogen (N) treatments is imperative in precision agriculture and field remote sensing, which can help to predict the yield of oilseed rape. In this study, an unmanned aerial vehicle (UAV) equipped with Red Green Blue (RGB) and multispectral cameras was used to acquire a series of field images at the flowering stage, and the flower number was manually counted as a reference. Images of the rape field were first classified using K-means method based on Commission Internationale de l’Éclairage (CIE) L*a*b* space, and the result showed that classified flower coverage area (FCA) possessed a high correlation with the flower number (r2 = 0.89). The relationships between ten commonly used vegetation indices (VIs) extracted from UAV-based RGB and multispectral images and the flower number were investigated, and the VIs of Normalized Green Red Difference Index (NGRDI), Red Green Ratio Index (RGRI) and Modified Green Red Vegetation Index (MGRVI) exhibited the highest correlation to the flower number with the absolute correlation coefficient (r) of 0.91. Random forest (RF) model was developed to predict the flower number, and a good performance was achieved with all UAV variables (r2 = 0.93 and RMSEP = 16.18), while the optimal subset regression (OSR) model was further proposed to simplify the RF model, and a better result with r2 = 0.95 and RMSEP = 14.13 was obtained with the variable combination of RGRI, normalized difference spectral index (NDSI (944, 758)) and FCA. Our findings suggest that combining VIs and image classification from UAV-based RGB and multispectral images possesses the potential of estimating flower number in oilseed rape.
We investigated the feasibility and potentiality of determining firmness, soluble solids content (SSC), and pH in kiwifruits using hyperspectral imaging, combined with variable selection methods and calibration models. The images were acquired by a push-broom hyperspectral reflectance imaging system covering two spectral ranges. Weighted regression coefficients (BW), successive projections algorithm (SPA) and genetic algorithm–partial least square (GAPLS) were compared and evaluated for the selection of effective wavelengths. Moreover, multiple linear regression (MLR), partial least squares regression and least squares support vector machine (LS-SVM) were developed to predict quality attributes quantitatively using effective wavelengths. The established models, particularly SPA-MLR, SPA-LS-SVM and GAPLS-LS-SVM, performed well. The SPA-MLR models for firmness (R pre = 0.9812, RPD = 5.17) and SSC (R pre = 0.9523, RPD = 3.26) at 380–1023 nm showed excellent performance, whereas GAPLS-LS-SVM was the optimal model at 874–1734 nm for predicting pH (R pre = 0.9070, RPD = 2.60). Image processing algorithms were developed to transfer the predictive model in every pixel to generate prediction maps that visualize the spatial distribution of firmness and SSC. Hence, the results clearly demonstrated that hyperspectral imaging has the potential as a fast and non-invasive method to predict the quality attributes of kiwifruits.
Near-infrared (874–1734 nm) hyperspectral imaging (NIR-HSI) technique combined with chemometric methods was used to trace origins of 1200 Chinese wolfberry samples, which from Ningxia, Inner Mongolia, Sinkiang and Qinghai in China. Two approaches, named pixel-wise and object-wise, were investigated to discriminative the origin of these Chinese wolfberries. The pixel-wise classification assigned a class to each pixel from individual Chinese wolfberries, and with this approach, the differences in the Chinese wolfberries from four origins were reflected intuitively. Object-wise classification was performed using mean spectra. The average spectral information of all pixels of each sample in the hyperspectral image was extracted as the representative spectrum of a sample, and then discriminant analysis models of the origins of Chinese wolfberries were established based on these average spectra. Specifically, the spectral curves of all samples were collected, and after removal of obvious noise, the spectra of 972–1609 nm were viewed as the spectra of wolfberry. Then, the spectral curves were pretreated with moving average smoothing (MA), and discriminant analysis models including support vector machine (SVM), neural network with radial basis function (NN-RBF) and extreme learning machine (ELM) were established based on the full-band spectra, the extracted characteristic wavelengths from loadings of principal component analysis (PCA) and 2nd derivative spectra, respectively. Among these models, the recognition accuracies of the calibration set and prediction set of the ELM model based on extracted characteristic wavelengths from loadings of PCA were higher than 90%. The model not only ensured a high recognition rate but also simplified the model and was conducive to future rapid on-line testing. The results revealed that NIR-HSI combined with PCA loadings-ELM could rapidly trace the origins of Chinese wolfberries.
The detection of Thermal Power Plants (TPPs) is a meaningful task for remote sensing image interpretation. It is a challenging task, because as facility objects TPPs are composed of various distinctive and irregular components. In this paper, we propose a novel end-to-end detection framework for TPPs based on deep convolutional neural networks. Specifically, based on the RetinaNet one-stage detector, a context attention multi-scale feature extraction network is proposed to fuse global spatial attention to strengthen the ability in representing irregular objects. In addition, we design a part-based attention module to adapt to TPPs containing distinctive components. Experiments show that the proposed method outperforms the state-of-the-art methods and can achieve 68.15% mean average precision.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.