Chongyang Wang

Papaya Tree Detection with UAV Images Using a GPU-Accelerated Scale-Space Filtering Method

The use of unmanned aerial vehicles (UAV) can allow individual tree detection for forest inventories in a cost-effective way. The scale-space filtering (SSF) algorithm is commonly used and has the capability of detecting trees of different crown sizes. In this study, we made two improvements with regard to the existing method and implementations. First, we incorporated SSF with a Lab color transformation to reduce over-detection problems associated with the original luminance image. Second, we ported four of the most time-consuming processes to the graphics processing unit (GPU) to improve computational efficiency. The proposed method was implemented using PyCUDA, which enabled access to NVIDIA’s compute unified device architecture (CUDA) through high-level scripting of the Python language. Our experiments were conducted using two images captured by the DJI Phantom 3 Professional and a most recent NVIDIA GPU GTX1080. The resulting accuracy was high, with an F-measure larger than 0.94. The speedup achieved by our parallel implementation was 44.77 and 28.54 for the first and second test image, respectively. For each 4000 × 3000 image, the total runtime was less than 1 s, which was sufficient for real-time performance and interactive application.

A total suspended sediment retrieval model for multiple estuaries and coasts by Landsat imageries

Based on 119 in-situ data from five estuaries and coasts of South China including Xunwen coast, estuary of Moyangjiang River, estuary and coast of Pearl River, estuary of Hanjiang River and estuary of Yangtze River, this paper aims to develop and establish a TSS retrieval model that applicable in different field conditions. After recalibrating and validating the form with the highest correlation coefficient between reflectance and TSS concentration and other TSS retrieval models that have been successful applied in many places, we found that the quadratic model of the ratio of logarithmic transformation of red band and near infrared band and logarithmic transformation of TSS concentration (QRLTSS) shows the highest performance. QRLTSS model based on Landsat OLI, ETM+ and TM can explained about 71% of the TSS concentration variation (4.3~577.2 mg/L) in the five regions and has a high and acceptable validation accuracy with root mean square error (RMSE) of 21.5-25mg/L and mean relative error (MRE) of 27.2-32.2%. We concluded that QRLTSS model can be used to quantify the TSS concentration of multiple estuaries and coasts of south China which would be helpful to understand the temporal and spatial variation of TSS in a large region. QRLTSS model should be applied to Landsat imagery for further validation in the future. The approach proposed in the paper also could promote the research work of establishing regional and uniform TSS retrieval model forward.

The spatial and temporal variation of total suspended solid concentration in Pearl River Estuary during 1987–2015 based on remote sensing

The movement and migration of total suspended solid (TSS) are the essential component of global material cycling and change. Based on the TSS concentrations retrieved from 112 scenes of Landsat remote sensing imageries during 1987-2015, the spatial and temporal variations of TSS concentration in high flow season and low flow seasons of six sub-regions (west shoal, west channel, middle shoal, east channel, east shoal and Pearl River Estuary Chinese White Dolphin National Nature Reserve and its adjacent waters (NNR)) of Pearl River Estuary (PRE) were analyzed and compared by statistical simulation. It was found that TSS concentrations in east and west shoals were about 23mg/L and 64mg/L higher than that of the middle shoal, respectively. There was a significant decreasing trend of TSS concentration from the northwest (223.7mg/L) to southeast (51.4mg/L) of study area, with an average reduction of 5.86mg/Lperkm, which mainly attributes to unique interaction of runoff and tide in PRE. In high flow season, there existed a significant and definite annual cycle period (5-8years) of TSS concentration change primarily responding to the periodic variation of precipitation. There were five full-fledged period changes of TSS detected in west shoal and west channel (the years of changes in 1988, 1994, 1998, 2003, 2010, 2015), while there were the last four cycle periods found in middle shoal, east channel, east shoal and NNR only. TSS concentrations in shoals and channels of PRE showed a significant decreased trend mainly due to the dam construction at the same time, with an average annual TSS concentration decrease of 5.7-10.1mg/L in high flow season from 1988 to 2015. There was no significant change trend of TSS concentration in NNR before 2003, but the TSS concentration decreased significantly after the establishment of the NNR since June 2003, with an average annual decrease of 9.7mg/L from 2004 to 2015. It was deduced that man-made protection measures had a great influence on the variation trend and intensity of TSS concentration in PRE, but had little effect on the cycle of TSS changes, indicating that the cyclical change is a very strong natural law. In low flow season, there was no significant change trend of TSS concentrations in PRE except that TSS concentrations in west channel and middle shoal showed a weak increasing trend (2.1mg/L and 2.9mg/L, respectively), which is probably because of controlled discharge for avoiding the intrusion of saltwater in PRE. Evidently, the change trend and cycle periods of TSS concentration in high- and low-flow seasons in six sub-regions of PRE had significant difference. The decreasing trend and cycle periods of TSS concentration mainly occurred in high flow season. The change trend and cycle periods of TSS concentration in low flow season was relatively small in PRE. The study shows that long series mapping of Landsat remote sensing images is an effective way to help understanding the spatial and temporal variation of TSS concentrations of estuaries and coasts, and to increase awareness of environmental change and human activity effects.

Spatiotemporal Analysis of MODIS Land Surface Temperature With In Situ Meteorological Observation and ERA-Interim Reanalysis: The Option of Model Calibration

The land surface temperature (<italic>Ts</italic>) is an important parameter in land surface and atmosphere studies. A set of synchronously observed “ground-truth” temperature as training data is required for some empirical/semiempirical statistical and neural network methods for retrieving <italic>Ts</italic> from passive microwave (PMW) remote sensing data. To provide information for the choice of the most suitable dataset in <italic>Ts </italic> retrieval of PMW remote sensing, the spatiotemporal comparison between the moderate-resolution imaging spectroradiometer <italic>Ts</italic> (MODIS <italic>Ts</italic>), the meteorologically observed <italic>Ts</italic> ( <italic>in situ Ts</italic>), the meteorologically observed near-surface air temperature (<italic>in situ Ta</italic>), and European Center for Medium-Range Weather Forecast reanalysis products, the ERA-Interim <italic>Ts</italic> (ERA <italic>Ts</italic>), in South China for each seasons daytime and nighttime is conducted in this paper. Results show that a large discrepancy between the MODIS <italic>Ts</italic> and the <italic>in situ Ts</italic> exists, whereas the discrepancies between the MODIS <italic>Ts</italic>, the <italic>in situ Ta</italic> and the ERA <italic> Ts</italic> are relatively smaller in daytime. For nighttime period, the differences between each dataset are relatively much smaller. Because the MODIS <italic>Ts</italic> is representative at the satellite pixel scale, it has a smaller spatial-scale mismatch with PMW data compared to <italic>in situ</italic> meteorological observation. The MODIS <italic>Ts</italic> is suitable for both the daytime and the nighttime PMW <italic>Ts</italic> model calibration if it is synchronously observed under almost clear-sky condition. By contrast, for the PMW <italic>Ts</italic> model calibration within the daytime period, the synchronously obtained <italic>in situ Ts</italic> is not suitable to be used as training data. If the ground temperature of daytime period derived from PMW is required, but the MODIS <italic> Ts</italic> is unavailable, the <italic>in situ Ta</italic> should be selected as the “ground truth” for the model calibration. However, it should be noticed that the inversion results are the near-surface air temperature rather than the <italic>Ts</italic>. Remarkably, reanalysis products such as the ERA <italic>Ts</italic> presents an alternative choice for both the daytime and the nighttime <italic>Ts</italic> model calibration if there are no MODIS <italic>Ts</italic> products or <italic>in situ</italic> temperature available. After the comparison, an example of PMW <italic>Ts</italic> retrieval for nighttime period was given, showing a promising performance on deriving an applicable PMW <italic>Ts</italic> inversion model based on the selection of an appropriate training dataset.

Machine Learning for Estimating Leaf Dust Retention Based on Hyperspectral Measurements

Hyperspectral sensors provide detailed information for dust retention content (DRC) estimation. However, rich hyperspectral data are not fully utilized by traditional image analysis techniques. We integrated several recently developed machine learning algorithms to estimate DRC on plant leaves using the spectra measured by the ASD FieldSpec 3. The experiments were carried out on three common green plants of southern China. The important hyperspectral variables were first identified by applying the random forest (RF) algorithm. Three estimation models were then developed using the support vector machine (SVM), classification and regression tree (CART), and RF algorithms. The results showed that the increase in dust retention contents on plant leaves enhanced their reflectance in the visible wavelength but weakened their reflectance in the infrared wavelength. Wavelengths in the ranges of 450–500 nm, 550–600 nm, 750–1000 nm, and 1100–1300 nm were identified as important variables using the RF algorithm and were used to estimate the DRC. The comparison of the three machine learning techniques for DRC estimation confirmed that the SVM and RF models performed well because their estimations were similar to the measured DRC. Specifically, the average for SVM and RF model are 0.85 and 0.88. The technical approach of this study proved to be a successful illustration of using hyperspectral measurements to estimate the DRC on plant leaves. The findings of this study can be applied to monitor the DRC on leaves of other plants and can also be integrated with other types of spectral data to measure the DRC at a regional scale.

Monitoring litchi canopy foliar phosphorus content using hyperspectral data

Abstract Phosphorus (P) is an important element to litchi yield and fruit quality in addition to nitrogen (N) and potassium (K). This study was undertaken to explore the ability to predict P content using canopy reflectance. Some published indices and two ratio spectral indices (Ratio of reflectance index, RRI; Ratio of reflectance difference index, RRDI) developed by band interactive-optimization algorithms were investigated to determine their performance in predicting litchi canopy foliar P content. The results showed that optimal spectral indices selected by correlation analyses reached the highest level of accuracy in the retrieval of P content at each growth stage (R2cv = 0.54–0.98, RMSEcv = 0.02–0.03). The particular wavelengths of importance in the significant RRIs and RRDIs changed with the growing stages, cultivars and planting conditions. The sensitive wavebands ranged from the visible to the short-wave infrared (SWIR) regions, which are related to the absorption features of pigments (e.g., anthocyanin, chlorophyll), proteins, nitrogen, starch, sugar, oil, cellulose, and lignin. And the wavebands in SWIR region were used in the optimal RRIs and RRDIs for growth stages. This study demonstrates that the optimal RRDI is useful in predicting litchi foliar P content. The successes of use of SWIR in foliar nutrient monitoring is important for precision agriculture.

Temperature variation and winter planted potato's NDVI change during early 2016's super cold wave in Guangdong province, South China

under the background of climate warming, obvious decrease trend in frequency and intensity of cold wave that invaded Guangdong province, South China have been observed in recent years. However in January of 2016, a named “super cold wave” which caused serious influence on human health and agriculture production invaded Guangdong. In this paper, satellite derived land surface temperature (LST) from GCOM-W1/AMSR2 brightness temperature (TB) were retrieved. Based on the satellite-derived LST, its variation characteristic was analyzed, which showed an apparent decrease trend and increase trend in Guangdong province during the whole “super cold wave” process. Moreover, based on NDVI retrieved from HJ-1 A/B satellites CCD sensors, obvious declination of potatos NDVI after the “super cold wave” can be seen: nearly 89% area of potatos NDVI decreased reflecting great loss on potatos production which was confirmed by in situ investigation. This study demonstrated that multi-sources satellite remote sensing data have the capability of monitoring temperature change in cold wave and assessing crop loss after cold injury.

Establishment of remote sensing monitoring method and standard on winter planting crops: A case study in Leizhou Peninsula of south China

The winter crop planting can bring increased income for peasants, but also improve the soil quality by rotation system. So, the planting area monitoring remote sensing of winter crops have been an important work for agriculture planting structure statistic, management and decision in south China. At present the cloudy and rainy weather is a very serious issue for remote sensing of winter crops in south China. So grasping the phenology of winter crops is of vital significance to establishment of remote sensing monitoring method and standard on winter crops. In order to form the remote sensing monitor method and standard to extract the planting area of winter crops, the paddy field and dryland based multi-temporal spectral angle remote sensing approach and standard were put forward, that includes five Landsat OLI image spectra and corresponding NDVI to compose the reference spectra of 40 bands during autumn and winter season. The mapping result in 12 experimental plots showed that the remote sensing monitoring standard performed well for retrieving the winter planting crops in Guangdong province of south China.

Winter Wheat Leaf Area Index (LAI) Inversion Combining with HJ-1/CCD1 and GF-1/WFV1 Data

The LAI is the key factor which has an important influence on crop growth. LAI inversion from remote sensing is an important work in crop management. While, the accuracy of LAI inversion from remote sensing data is restricted by the limited number of observation. Multiple-sensor method has been proposed by the researchers. In this study, two sensor remote sensing data (HJ-1A/CCD1 and GF-1/WFV1) were collected in the study area. The random forest regression (RFR) was adopted in LAI inversion. The MODIS LAI product and the measured wheat LAI were used to calibrate and validate the LAI inversion model. The four spectral indices (DVI, SR, EVI, and SAVI) based on remote sensing data were calculated to develop the LAI inversion model. The accuracy of inversion of wheat LAI by remote sensing image can be improved by adding observations of angle data. Our data analysis resulted in an accuracy of R2 = 0.36, MAE = 0.467, and RMSE = 0.613 for the measured LAI. And in the validation by MODIS LAI product, an accuracy of R2 = 0.48, MAE = 1.05, and RMSE = 2.72 was found, which was a little greater than the average accuracy of mono-angle data for inversion of LAI. The result indicates that the reasonable combination of multi-sensor data can improve the accuracy of LAI estimation.

Leaf Area Index Estimation of Winter Pepper Based on Canopy Spectral Data and Simulated Bands of Satellite

Leaf area index (LAI) is an important indicator of crop growth status. In this paper, the relationships between canopy reflectance at 400–2500 nm and leaf area index (LAI) in pepper crop were studied. 102 pair of canopy reflectance and LAI of pepper were collected in 2014–2015. Reflectance of canopy were measured in the field over a spectral range of 400–2500 nm. Simultaneously, the LAI were collected by the LAI-2000. Estimation models of LAI were developed based on the whole spectrum range by partial least squares regression (PLSR) and support vector regression (SVR), respectively. Then the field canopy spectra were resampled according to the band response functions of seven satellite sensors. They were the Vegetation and environment monitoring on a new micro-satellite (VENμS), Worldview-2 (WV-2), RapidEye-1 (RE-1), HJ1/CCD1, Sentinel-2, Landsat 8/OLI and GaoFen (GF) 1/WFV1. The values of common used spectral indices were calculated based on the simulated sensor bands, respectively. Prediction models were also developed based on the spectral indices and simulated bands. The results showed that the PLSR model by whole spectrum had the good accuracy of LAI estimation with the R2c = 0.726, RMSEc = 0.462, R2cv = 0.635, RMSEcv = 0.538. For the simulated satellite datasets, the better LAI estimation were obtained by Sentinel-2 and Venμs bands with the R2cv greater than 0.600 and RMSEcv less than 0.557. The Estimation model by simulated WV-2 bands, and RE-1 bands had the lowest performance with the R2cv between 0.50 and 0.55, and RMSEcv between 0.600 and 0.623. The inversion results demonstrated the potential of the multispectral remote sensing data to calibrate the LAI estimation model of winter pepper for the precision agriculture application.