Browsing by Author "Song, Xiaopeng"
Now showing 1 - 6 of 6
- Results Per Page
- Sort Options
Item Application of unmanned aerial systems and deep learning in high-throughput plant phenotyping(2022-08) Lin, Zhe; Guo, Wenxuan; Ritchie, Glen; Kelly, Brendan; Song, XiaopengPlant phenotyping plays an essential role in decision support in precision agriculture and plant breeding. However, traditional phenotyping methods are typically through manually measuring the target traits, which is time-consuming and labor-intensive with sampling bias. Technological innovations in unmanned aerial systems (UAS) with various sensors provide high-resolution data for high-throughput plant phenotyping. UAS image-based machine learning algorithms have been applied in plant phenotyping. However, only limited studies have evaluated the performance of using deep learning and UAS imaging in cotton or plant phenotyping for breeding. Therefore, the goal of this study was to evaluate the performance of applications of using UAS images and deep learning algorithms in sorghum (Sorghum bicolor L. Moench) and cotton (Gossypium hirsutum L.) phenotyping. The objectives were to 1) develop a deep learning CNN image segmentation algorithm using UAS imagery to detect and quantify sorghum panicles; 2) to assess the application of MobileNet and CenterNet models in cotton stand counting at the seedling stage; 3) to develop an algorithm for detecting and counting open cotton bolls and assess the performance in relation to image acquisition altitudes and camera angles; 4) to develop a method for cotton boll detection using point cloud data derived from LiDAR and RGB data and to compare the performance of RGB images and LiDAR point in cotton boll detection. A set of 1000 UAS images, acquired at 10 m height, were randomly selected, and a mask was developed for each by manually delineating sorghum panicles for training the U-Net model for sorghum panicle detection. The algorithm performed the best with 1000 training images, with an accuracy of 95.5% and a root mean square error (RMSE) of 2.5, which showed the accuracy had a general increasing trend with the number of training images. The results indicated that the integration of image segmentation and the U-Net CNN model is an accurate and robust method for sorghum panicle counting and offers an opportunity for enhanced sorghum breeding efficiency and accurate yield estimation. UAS images were collected at 20 m flight height at the seedling stages on two dates in 2020 for cotton stand counting. The CenterNet model had a better overall performance for cotton plant detection and counting with 900 training images. More training images are required when applying object detection models on images with different dimensions from training datasets. Both MobileNet and CenterNet models have the potential to accurately and timely detect and count cotton plants based on high-resolution UAS images at the seedling stage. UAS imagery was collected from two angles, 60° to the land surface at 15 m and 90° at 10 m and 15 m flight height for open cotton boll detection in 2020. Open cotton bolls were detected with UAS images using the CenterNet model. Cotton boll count had more accurate predictions with UAS images taken with 60° camera angles. Models trained with images containing leaves performed more accurately than models trained with images only containing open bolls. UAS imagery acquired at oblique angles is effective for boll counting and offers an opportunity for enhanced cotton breeding efficiency and accurate yield estimation. LiDAR data was collected at 12 m and UAS imagery was collected with 90° and 60° cameras at 20 m height in 2021 for open cotton boll detection. Open cotton bolls were detected in LiDAR and UAS RGB-based point cloud data using the density-based spatial clustering of applications with noise (DBSCAN) algorithm. LiDAR-based point cloud data performed more accurately on cotton boll detection than RGB image-based point cloud data. Also, cotton boll count had more accurate predictions with small-sized plants. Point cloud data from LiDAR and UAS RGB imagery offer an opportunity for enhanced cotton breeding efficiency and accurate yield estimation. This study offers useful guidance for choosing appropriate deep learning models, remote sensing sensors, and the quantity of training data for computer vision tasks in agricultural applications. The proposed algorithms support decision-making in precision agriculture and plant breeding. Additional research is required to evaluate how image resolution, flight height, and environmental factors like soil background and light conditions affect plant phenotyping.Item Application of unmanned aerial systems in precision agriculture and plant breeding(2022-08) Gu, Haibin; Guo, Wenxuan; Ritchie, Glen; Mills, Cory; Lewis, Katie; Song, XiaopengUnmanned aerial systems (UAS) are a promising remote sensing technology in smart agriculture. The overall objective of this study was to evaluate the effectiveness of UAS technology in the application of precision agriculture and plant breeding. UAS remote sensing was applied in estimating surface soil water content (SWC), correcting cotton yield monitor data, assessing water stress of cotton cultivars, and predicting cotton yield. SWC is a major determinant of crop production, and accurately retrieving SWC plays a crucial role in effective water management. The objective of chapter II was to develop an algorithm to retrieve SWC by integrating soil texture into a vegetation index derived from UAS multispectral and thermal images. The Normalized Difference Vegetation Index (NDVI) and surface temperature (Ts) derived from the UAS multispectral and thermal images were employed to construct the Temperature Vegetation Dryness Index (TVDI) using the trapezoid model. Soil texture was incorporated into the trapezoid model based on the relationship between soil texture and the lower and upper limits of SWC to form the Texture Temperature Vegetation Dryness Index (TTVDI). The results showed that the TTVDI had better performance in estimating SWC compared to the TVDI, with an increase in R2 by 14.5% and 14.9%, and a decrease in RMSE by 46.1% and 10.8% for the 2019 and 2020 samples, respectively. The yield monitor is one of the most adopted precision agriculture technologies. The objective of chapter III was to evaluate the application of UAS images in improving yield monitor data for cotton (Gossypium hirsutum L.) yield estimation. The results showed that the estimated yield using cotton unit coverage (CUC) from UAS images was correlated with the yield monitor data (R2 = 0.57 and 0.75) in two fields. The multiple linear regression (MLR) models had better performance in correcting the yield monitor data compared to the simple linear regression (SLR) model, with an increase in R2 by 10.3% to 14.3%, a decrease in root-mean-square deviation (RMSE) by 9.7%, and 29.4%, and a decrease in relative Absolute Error (RAE) by 13.5% and 24.7% in two fields. After correction, the spatial autocorrelation of cotton yield increased in two fields, as indicated by the increase of Moran`s I value by 4.5% and 19.7% in the two fields, respectively. Efficiently monitoring and quantifying the response of crop cultivars to water stress is critical in developing high-performance genotypes in water-limited environments. The objective of chapter IV was to assess water stress in cotton using UAS images to select water stress-resistant cultivars in plant breeding. Vegetation indices (VIs) and crop water stress index (CWSI) derived from UAS images were applied to assess water stress in eight cotton cultivars. The VIs effectively differentiated cultivars in the middle and late seasons, while CWSI detected cultivar differences in the late growing season. VIs had a strong positive relationship with cotton yield starting from the mid-growing season in two years (R2 ranged from 0.90 to 0.95). The successful classification of cultivars using UAS images provides critical information for selecting drought-resistant cultivars under low-level irrigation rates (e.g., 30% ET), while it is a good indicator for selecting high-yielding cultivars under minimal water stress (e.g., 90% ET) in cotton breeding. Accurate prediction of crop yield is critical to improving the selection of water stress-resistant cultivars. The objective of chapter V was to assess the relationship between cotton lint yield and VIs to determine the optimal period for cotton yield prediction over the growing season. Individual VIs and multi-temporal VIs derived from UAS images were applied to build the relationship with cotton yield. The results showed that multi-temporal VIs has a stronger relationship with cotton yield. The analysis based on multi-temporal VIs showed that EVI performed better in yield prediction than other VIs. The optimal period for cotton yield prediction was in the mid-late stages of cotton growth, as suggested by high R2 (0.92), low RMSE (229.8 kg ha-1), and RAE% (37.4%). This study demonstrated the applicability of UAS images in estimating surface SWC and improving the yield monitor data in precision agriculture. In addition, UAS images have great potential in assessing the performance of cotton cultivars and predicting yield in response to water stress, which helps to improve breeding efficiency in selecting water stress-resistant cotton cultivars.Item Identifying and spatio-temporal analysis of dust point sources in southwestern United States and applying robotics in wind erosion studies(2020-12) Kandakji, Tarek; Lee, Jeffrey A.; Mulligan, Kevin R.; Song, Xiaopeng; Ardon-Dryer, Karin; Gill, Thomas E.; Van Pelt, Robert S.; Zhu, ZheThis dissertation is composed of three studies; two of them were published as scientific articles in two peer-reviewed academic journals. The first study is entitled “Identifying and Characterizing Dust Point Sources in the Southwestern United States using Remote Sensing and GIS” and it is published in Geomorphology. Appendix A is a screenshot of the first page of the published article showing Volume number and the related Digital Object Identifier (DOI). This study aimed to detect dust point sources in the Southern Great Plains and the Chihuahuan Desert of the United States (U.S.) between the years 2001 and 2016 using satellite images. The detected dust points were later characterized based on the geomorphology and the land use/land cover type of their emission source. A total of 1508 dust point sources were detected. The results showed that ephemeral lakes (i.e. playas) geomorphic class produce the most dust sources in proportion to their area. Cultivated croplands enclose 43% of the dust points, while shrublands and grasslands, combined, enclose 45% of the points. Results from this study confirm the importance of playas as a dynamic source of dust in southwestern U.S. Moreover, this study suggests that anthropogenic factors may be playing a major role in dust emission within southwestern U.S. The results of this study highlights the need to perform further spatio-temporal and quantitative analysis on the detected dust point sources to quantify the contribution of different land cover types on dust emission in the region. This suggestion was fulfilled in the second study of this dissertation. The second study is entitled “Drought and Land Use/Land Cover Impact on Dust Sources in Southern Great Plains and Chihuahuan Desert of the US: Inferring Anthropogenic Effect” and it is published in the Science of The Total Environment. Appendix B is a screenshot of the first page of the published manuscript showing the Volume number and the related DOI. The objective of this research was using the dust point sources dataset produced in the first study to test the hypothesis that there is a statistically significant association between drought level and land cover that may contribute to dust emission in the Southern Great Plains and Chihuahuan Desert regions of the U.S. To reach this objective, several analysis were conducted including spatio-temporal analysis and statistical analysis. The chi-square analysis results showed a significant association between land use type and drought level on dust emission (χ2 (6) = 45.54, р < 0.001), thus confirming the proposed hypothesis. Results from this study indicate that human activities in dust-prone regions will worsen the negative impacts of drought by changing land cover and making the soil more erodible in multiple ways. The third study of this dissertation is entitled “In Situ Measurement of Wind Shear Stress using the Rhex Hexapod Robot Platform”. In this work we used the rigid hexapod robot (RHex) to carry 3D sonic anemometers at two near surface heights to collect wind data. The rationale behind using a mobile senor in the field is to model wind and dust emission using a robust technique rather than using the traditional wind tunnels. The real conditions in the field are hard to simulate in a wind tunnel since it utilizes artificial wind stream flow that is restrained by the walls of the wind tunnel. Moreover, wind tunnel experiments fail to replicate air temperature and moisture content that affect momentum flux. Momentum flux within the boundary layer is an important factor in generating shear stress on the surface. Solar radiation in the open field can cause heating of the air at the surface, which will affect convective overturn and increase wind velocity. The robot scanned a 10 m × 10 m study area in an open field and collected wind and shear stress data as affected by four types of artificial roughness elements simulating a solid object like tree trunk and natural shrubs with different porosity and density. The shear stress and the wind speed data collected by the robot appear to be randomly distributed and does not follow any pattern, which reflects the complexity of real world conditions. These conditions are controlled by many factors, like wind gusts, surface heating, and resulting turbulence from convective overturn. However, the shading effect of solid object was observed, and its effect extends up to 0.5 m downwind of the object at 0.2 m height above the ground (midpoint between ground surface and anemometer mounted at 0.4 m). This study provides an important guideline for future studies, and highlights the potential of using RHex in further studies. Researchers in successive studies should try to fix the limitations of using RHex and improve the study design based on this work.Item Leveraging NAIP,LiDAR and Sentinel data for accurate multiclass mapping of heterogenous grassland landscapes(2022-08) Subedi, Mukti; Portillo-Quintero, Carlos; Kahl, Samantha; Cox, Robert D.; McIntyre, Nancy; Song, XiaopengWith the advancement in remote sensing (RS), sensors, platforms and data processing, RS data have significantly contributed to science and policy. However, data processing capability has not fully matured to handle high spatial and temporal resolution data to make management decisions. Using high-spatial-resolution (NAIP), temporal (Sentinel), and light detection and ranging (LiDAR) data, I derived land use land cover (LULC) maps using machine learning and data fusion while accounting for spatial autocorrelation in the sample data. Chapter I offers a brief overview of the development and current state-of-the-art practices in accurate LULC mapping using high-spatial-resolution remote sensing data, then introduces a list of questions that were tackled in this dissertation. Finally, Chapter I presents a brief synopsis of subsequent chapters and summarize the limitations. The chapter outlines the process of data ingestion, pre-processing and machine learning to produce accurate LULC mapping. The chapter describes a methodological workflow that can be adapted to regular computing resources available in most office settings. Chapter III evaluated the efficacy and effectiveness of surface features derived from LiDAR data in improving the mapping of grass- and shrub-dominated landscapes. Chapter IV integrated the time-series Sentinel 2A data overhigh-resolution NAIP data in heterogeneous landscapes using stacking ensemble machine learning. Finally, chapter V summarizes each chapter and discusses each case study's overall significance. With data fusion and supervised machine learning, I showed a practical approach to producing accurate land use land cover maps of grass-and-shrub-dominated landscapes of Texas.Item Mapping urban growth of Dallas-Fort Worth metropolitan area from 1984 to 2019 using Landsat data(2020-12) Li, Shu; Song, Xiaopeng; Lee, Jeffrey A.; Cao, Guofeng; Mulligan, Kevin R.The Dallas-Fort Worth (DFW) Metropolitan Area is one of the fastest growing metropolitan areas in the U.S. Its rapid growth requires research investigation. Various studies have been conducted to understand the urbanization patterns and the impacts of urban expansion in this region using satellite data. Although the emphasis of previous studies was different, each of them discussed urban impervious surface change at different spatial and temporal scales. All previous studies agreed that the DFW metroplex experienced fast urban development and rapid increase in impervious surface cover over the recent decades, with the expansion of fringe areas around the DFW city cores. In this study, I applied an empirical, machine learning method to retrieve the long-term impervious surface cover for the DFW Metropolitan Area. I used the high-resolution planimetric maps obtained from the municipalities and the National Agriculture Imagery Program (NAIP) images as reference data for training and evaluation. I used Landsat data to generate annual continuous maps of impervious surface cover. The Landsat images were composited to summer and winter predictor variables according to vegetation seasonality. Composited seasonal images were able to reduce the variation and noise caused by vegetation phenology, atmospheric effect and cloud contamination. I trained a classification and regression tree (CART) model to predict impervious surface cover. The resultant maps were per-pixel continuous representation of impervious surface cover at a spatial resolution of 30-m annually from 1984 to 2019. I found that the area of impervious surface of DFW metroplex grew from 1,194 square kilometers to 2,880 square kilometers over the 35-year span. The counties of Dallas, Tarrant, Denton and Collin had the largest urban growth during the study time period. I also found that the quantified urban impervious surface increase at the county scale had high correlations with population growth over the same time period, with an r2 ranged from 0.83 to 0.96. The empirical method I applied can reliably map and monitor annual impervious surface cover change over a long period. The method can be potentially applied to other land cover types such as forest and cropland in other regions.Item The land use history, economic drivers, and future trends of urban growth in Saudi Arabia(2022-08) Aljaddani, Amal H.; Song, Xiaopeng; Lee, Jeffrey A.; Mulligan, Kevin R.; Ludwig, JulianThis dissertation investigates the land use history, economic drivers, and future trends of urban growth in Saudi Arabia. It consists of five chapters, including Introduction and Conclusions as the first chapter and last chapter, respectively. Chapter two aimed to use satellite observations to produce a long-term dataset at 30-meter spatial resolution for the 13 capital cities in Saudi Arabia between 1985 and 2019. This chapter has been published as a peer-reviewed article in the journal of Remote Sensing. In this chapter, I downloaded all available Landsat data, including Thematic Mapper (TM 4-5), Enhanced Thematic Mapper Plus (ETM+ 7), and Operational Land Imager (OLI 8). After that, urban and non-urban training samples were collected and fed into a Change detection and classification algorithm (CCDC) that used a random forest classifier (RFC). The CCDC algorithm was used to produce the annual classification maps for the 13 capital cities in the first month of July of each year. Then, historical maps of urban growth were compiled between 1985 and 2019 to monitor the changes in urban growth. I implemented a stratified random sampling design to assess the annual classification maps and multi-temporal urban change maps, which also provided the area estimation and uncertainties. Higher overall accuracy was found in the annual classification maps and the multi-temporal urban change maps. In 1985, 2000, and 2019, the urban area occupied 13.23, 14.96, and 27.43% of the total area, respectively. Chapter three examined the Granger-causality relationship between urban growth and economic variables in Saudi Arabia, such as real GDP, inflation, merchandise imports and exports, and oil rents. First, I implemented the Augmented-Dickey-Fuller (ADF), Phillips & Perron (PP), and Kwiatkowski-Phillips-Schmidt-Shin (KPSS) on the initial time series. Then, the first difference of the log was implemented when the time series confirmed non-stationarity. After that, the bivariate Granger-causality was performed on the stationary urban economic time series. The results showed one bidirectional relationship between the urban growth and the real GDP growth, a unidirectional relationship between the urban growth and merchandise imports growth, and a unidirectional relationship between merchandise exports growth and urban growth. There is no Granger causality between urban growth and inflation and between the urban growth and oil rents growth. Chapter four aimed to predict the future of urban growth in Riyadh, Saudi Arabia. In this chapter, the prediction models were based on the driving forces that produced the land suitability maps under three different scenarios: business as usual (BAU), rapid economic growth (REG), and Integrated environmental sustainability (IES). I used cellular automata, Markov chain (CA-Markov), and a multilayer perceptron (MLP) neural network that integrated the driving forces of each scenario. Next, the validation process was implemented for each scenario between the prediction maps of 2019 and the actual classification maps of 2019. The results showed that the Kappa standard of CA-Markov and MLP neural networks was moderate and above 65%. The urban area of the CA-Markov showed a substantial increase over 2019, 2030, and 2050 compared with MLP neural network. Overall, studies included in this dissertation provided a comprehensive understanding of the land use history, the economic drivers, and the likely future scenarios of urban growth in Saudi Arabia. The findings of the dissertation provide the scientific basis for public policy-making to improve urban planning design and environmental sustainability.