Estimation Of Vegetation Fraction Research Articles

Fine-tuning convolutional neural network with transfer learning for semantic segmentation of ground-level oilseed rape images in a field with high weed pressure

Image processing technology has gained considerable attention in agricultural proximal sensing applications, including plant disease detection, vegetation fraction estimation, monitoring of the crop growth status, and image-based site-specific management. Image segmentation is the first and crucial step to process complex infield images. However, image segmentation by either hand engineered-based or deep learning-based methods that train the entire system from scratch is a daunting task and needs several hundreds of labeled images that may be difficult to obtain in practice. The recent development of transfer learning has shown the potential of transferring the learned feature detectors of a pre-trained convolutional neural network to a new image dataset. This study was thus aimed to evaluate three transfer learning methods using a VGG16-based encoder net for semantic segmentation of oilseed rapes images in a field with high-density weeds. Three different transfer learning approaches using a VGG16-based encoder model were proposed, and their performances were compared to a VGG19-based encoder net. Relying on the intensive use of data augmentation and transfer learning, we showed that such networks could be trained end-to-end using a few annotated training images. The highest accuracy of 96% was obtained by the VGG16-based encoder net in which the fine-tuned model was only used for feature extraction and the segmentation was performed using shallow machine learning classifiers (MLCs). Transfer learning demonstrated to be efficient and presented a robust performance in segmenting plants amongst high-density weeds. The implementation of MLCs is reasonable for real-time applications with the segmentation time less than 0.05 s/image.

Estimation of vegetation fraction using RGB and multispectral images from UAV

ABSTRACTThe vegetation fraction (VF) monitoring in a specific area is a very important parameter for precision agriculture. Until a few years ago, high-cost flights on aeroplanes and satellite imagery were the only option to acquire data to estimate VF remotely. Recently, Unmanned Aerial Vehicles (UAVs) have emerged as a novel and economic tool to supply high-resolution images useful to estimate VF. VF is usually estimated by spectral indices using red-green-blue (RGB) and near-infrared (NIR) bands data. For this study, a UAV equipped with both kinds of sensors (RGB and NIR) was used to obtain high-resolution imagery over a maize field in progressive dates along the mid-season and the senescence development stages. The early-season stage was also monitored using only RGB spectral indices. Flights were performed at 52 m over the terrain, obtaining RGB images of 1.25 cm pixel−1 and multispectral images of 2.10 cm pixel−1. Three spectral indices in the visible region, Excess Green (ExG), Colour Index of Vegetation (CIVE), and Vegetation Index Green (VIg), and three NIR-based vegetation indices, Normalized Difference Vegetation Index (NDVI), Green NDVI (GNDVI), and Normalized Green (NG), were evaluated for VF estimation. Otsu’s method was applied to automatically determine the threshold value to classify the vegetation coverage. Results show that ExG presents the higher mean accuracy (85.66%) among all the visible indices, with values ranging from 72.54% to 99.53%, having its best performance in the earlier development stage. Nevertheless, GNDVI mean accuracy (97.09%) overcomes all the indices (visible and multispectral), ranging in value from 92.71% to 99.36%. This allowed comparing the accuracy difference gained by using a NIR sensor, with a higher economic cost than required using a simple RGB sensor. The results suggest that ExG can be a very suitable option to monitor VF in the early-season growth stage of the crop, while later stages could require NIR-based indices. Thus, the selection of the index will depend on the objectives of the study and the equipment capacity.

Remote Estimation of Vegetation Fraction and Flower Fraction in Oilseed Rape with Unmanned Aerial Vehicle Data

This study developed an approach for remote estimation of Vegetation Fraction (VF) and Flower Fraction (FF) in oilseed rape, which is a crop species with conspicuous flowers during reproduction. Canopy reflectance in green, red, red edge and NIR bands was obtained by a camera system mounted on an unmanned aerial vehicle (UAV) when oilseed rape was in the vegetative growth and flowering stage. The relationship of several widely-used Vegetation Indices (VI) vs. VF was tested and found to be different in different phenology stages. At the same VF when oilseed rape was flowering, canopy reflectance increased in all bands, and the tested VI decreased. Therefore, two algorithms to estimate VF were calibrated respectively, one for samples during vegetative growth and the other for samples during flowering stage. The results showed that the Visible Atmospherically Resistant Index (VARIgreen) worked most accurately for estimating VF in flower-free samples with an Root Mean Square Error (RMSE) of 3.56%, while the Enhanced Vegetation Index (EVI2) was the best in flower-containing samples with an RMSE of 5.65%. Based on reflectance in green and NIR bands, a technique was developed to identify whether a sample contained flowers and then to choose automatically the appropriate algorithm for its VF estimation. During the flowering season, we also explored the potential of using canopy reflectance or VIs to estimate FF in oilseed rape. No significant correlation was observed between VI and FF when soil was visible in the sensor’s field of view. Reflectance at 550 nm worked well for FF estimation with coefficient of determination (R2) above 0.6. Our model was validated in oilseed rape planted under different nitrogen fertilization applications and in different phenology stages. The results showed that it was able to predict VF and FF accurately in oilseed rape with RMSE below 6%.

Spatial Mapping of Vegetation Based on Out-of-Box Vegetation Indices

The infrastructures of power grid can be easily damaged by fire disasters. Monitoring the spatial variations in the vegetation fraction is significant to predicate the probability of a fire disaster. While each vegetation indices (VI) computing method tends to suit some problem better than others, and typically have different parameters and configurations to be adjusted before achieving optimal performance on a dataset, we focus on conjunction with many types of traditional VIs' computing methods to improve their performance. Experimental results show that the accuracy of vegetation fraction estimation based on our method is higher than by using any single vegetation indices.

Remote estimation of crop fractional vegetation cover: the use of noise equivalent as an indicator of performance of vegetation indices

Many algorithms have been developed for the remote estimation of vegetation fraction in terms of combinations of spectral bands, derivatives of reflectance spectra, neural networks, inversion of radiative transfer models, and several multispectral statistical approaches. The most widespread type of algorithm used is the mathematical combination of visible and near-infrared reflectance, in the form of spectral vegetation indices. The general objective of this study is to evaluate different vegetation indices for the remote estimation of the fractional vegetation cover in two crop types, maize and soybean, with contrasting canopy architectures and leaf structures. The noise equivalent of vegetation indices was used as an indicator of sensitivity and accuracy of vegetation fraction estimation. Among the indices tested, the enhanced vegetation index (EVI2), wide dynamic range vegetation index (WDRVI), green- and red-edge normalized difference vegetation index (NDVI) were found to be accurate in estimating vegetation fraction. These results were obtained using reflectance data acquired with close-range sensors (i.e. spectroradiometers mounted 6 m above the top of canopy). WDRVI was able to estimate vegetation fraction in both crops with no re-parameterization with RMSE below 6% and mean normalized bias below 2%.

Estimation and vicarious validation of urban vegetation abundance by spectral mixture analysis

Both moderate and high spatial resolution imagery can be used to quantify abundance and distribution of urban vegetation for urban landscape management and to provide inputs to physical process models. Estimation of vegetation fraction from Landsat ETM+ and Quickbird allows for operational monitoring and reconnaissance at moderate resolution with calibration and vicarious validation at higher resolution. Establishing a linear correspondence between ETM-derived vegetation fraction and Quickbird-derived vegetation fraction facilitates the validation task by extending the spatial scale from 30 × 30 m to a more manageable 2.8 × 2.8 m. A comparative analysis indicates that urban reflectance can be accurately represented with a three component linear mixture model for both Landsat ETM+ and Quickbird imagery in the New York metro area. The strong linearity of the Substrate Vegetation Dark surface (SVD) mixture model provides consistent estimates of illuminated vegetation fraction that can be used to constrain physical process models that require biophysical inputs related to vegetation abundance. When Quickbird-derived 2.8 m estimates of vegetation fraction are integrated to 30 m scales and coregistered to Landsat-derived 30 m estimates, median estimates agree with the integrated fractions to within 5% for fractions > 0.2. The resulting Quickbird-ETM+ scatter distribution cannot be explained with estimate error alone but is consistent with a 3% to 6% estimation error combined with a 17 m subpixel registration ambiguity. The 3D endmember fraction space obtained from ETM+ imagery forms a ternary distribution of reflectance properties corresponding to distinct biophysical surface types. The SVD model is a reflectance analog to Ridd's V–I–S land cover model but acknowledges the fact that permeable and impermeable surfaces cannot generally be distinguished on the basis of broadband reflectance alone. We therefore propose that vegetation fraction be used as a proxy for permeable surface distribution to avoid the common erroneous assumption that all nonvegetated surfaces along the gray axis are completely impermeable. Comparison of mean vegetation fractions to street tree counts in New York City shows a consistent relationship between minimum fraction and tree count. However, moderate and high resolution areal estimates of vegetation fraction provide complementary information because they image all illuminated vegetation, including that not counted by the in situ street tree inventory.

Vegetation and soil lines in visible spectral space: A concept and technique for remote estimation of vegetation fraction

The goal of this study is to investigate the information content of reflectance spectra of crops in the visible and near infrared range of the spectrum and develop a technique for remote estimation of vegetation fraction ( VF ). For four wheat species with VF =100% in a wide range of pigment contents and compositions, a high degree of covariance was found for paired reflectances ( R ) at 550 nm versus 700 nm ( R 550 versus R 700) and 500 nm versus 670 nm ( R 500 versus R 670). Both relationships, defined as 'vegetation lines', were linear with determination coefficients r 2 >0.9 and the plotted points were tightly clustered. Using the same coordinates to plot reflectances for a variety of soils, a high degree of covariance ( r 2 >0.94) and a distinct 'soil line' were found. The vegetation and soil lines define a two-dimensional spectral construct within which canopy reflectances, regardless of VF, may be located. Based on these optical properties of vegetation and soils, an attempt was made to estimate VF remotely for selected plant canopies. It is suggested that the coordinate location within the constructs, as defined by reflectances at 500 nm and 670 nm as well as at 550 nm and 700 nm, be used to measure VF . Algorithms for VF assessment in wheat for a wide range of soil brightness were devised and validated. The root mean square error (RMSE) of VF prediction was less than 10%. The technique was also validated by means of independent datasets taken above cornfields in Nebraska. The RMSE of VF prediction did not exceed 9.7%.

Novel algorithms for remote estimation of vegetation fraction

Spectral properties of a wheat canopy with vegetation fraction (VF) from 0% to 100% in visible and near-infrared (NIR) ranges of the spectrum were studied in order to devise a technique for remote estimation of VF. When VF was <60%, from emergence till middle of the elongation stage, four distinct, and quite independent, spectral bands of reflectance existed in the visible range of the spectrum: 400 to 500 nm, 530 to 600 nm, near 670 nm, and around 700 nm. When VF was between 60% and 100%, reflectance in the NIR leveled off or even decreases with an increase of VF. The decreased reflectance in the NIR, occurring at or near the midseason, can be a limiting factor in the use of that spectral region for VF estimation. It was found that for VF>60%, the information content of reflectance spectra in visible range can be expressed by only two independent pairs of spectral bands: (1) the blue from 400 to 500 nm and the red near 670 nm; (2) the green around 550 nm and the red edge region near 700 nm. We propose using only the visible range of the spectrum to quantitatively estimate VF. The green (as well as a 700-nm band) and the red (near 670 nm) reflectances were used in developing new indices, which were linearly proportional to wheat VF ranging from 0% to 100%. The Atmospherically Resistant Vegetation Index (ARVI) concept was used to correct indices for atmospheric effects. Visible Atmospherically Resistant Index in the form VARI=( R green− R red)/( R green+ R red− R blue) was found to be minimally sensitive to atmospheric effects allowing estimation of VF with an error of <10% in a wide range of atmospheric optical thickness. Validation of the newly suggested technique was carried out using wheat independent data sets and reflectance data obtained for cornfields in Nebraska. Predicted green VF was compared with retrieved from digital images. Despite the fact that the reflectance contrast among the visible channels is much smaller than between the visible and NIR, the sensitivity of suggested indices to moderate to high values of VF is much higher than for the Normalized Difference Vegetation Index (NDVI), and the error in VF prediction did not exceed 10%. Suggested indices will complement the widely used NDVI, ARVI, Soil Adjusted Vegetation Index (SAVI) and others, which are based on the red and the NIR bands in VF estimation, and also Green Atmospherically Resistant Index (GARI), which is based on the green and the NIR bands.

Sub-pixel Model for Vegetation Fraction Estimation based on Land Cover Classification

Sub-pixel Model for Vegetation Fraction Estimation based on Land Cover Classification