<div> <p>Calibration of crop models is an expensive and time intensive procedure, which is essential to accurately predict the possible crop yields given changing climate conditions. One solution is the utilization of unmanned aircraft systems (UAS) deployed with Red Green Blue Composite (RGB), and multispectral sensors, which has the potential to measure and collect in field biomass and yield in a cost and time effective manner. The objective of this project was to develop a relationship between remotely sensed data and crop indices, similar to biomass, to improve the ability to parametrize crop models for local conditions, which in turn could potentially improve the quantification of the effect of hydrological extremes on predicted yield. An experiment consisting of 750 plots (350 varieties) was planted in 2018, and a subset of 18 plots (9 varieties) were planted in 2019. The in-situ above ground biomass along with multispectral and RGB imagery was collected for both experiments throughout the growing season. The imagery was processed through a custom software pipeline to produce spectrally corrected imagery of individual plots. A model was fit between spectral data and sampled biomass resulting in an R-square of 0.68 and RMSE of 160 g when the model was used to estimate biomass for multiple flight dates flights. The VIC-CropSyst model, a coupled hydrological and agricultural system model, was used to simulate crop biomass and yield for multiple years at the experiment location. Soybean growth was parametrized for the location using CropSyst’s Crop Calibrator tool. Biomass values generated from UAS imagery, along with the in-situ collected biomass values were used separately to parametrize soybean simulations in CropSyst resulting in very similar parameter sets that were distinct from the default parameter values. The parametrized crop files along with the default files were used separately to run the VIC-CropSyst model and results were evaluated by comparing simulated and observed values of yield and biomass values. Both parametrized crop files (using in-situ samples and UAS imagery) produced approximately identical results with a max difference of 0.03 T/Ha for any one year, compared to a base value of 3.6 T/Ha, over a 12-year period in which the simulation was ran. The parametrized runs produced yield estimates that were closer to in-situ measured yield, as compared to unparametrized runs, for both bulk varieties and the run experiments, with the exception of 2011, which was a flooding year. The parametrized simulations consistently produced simulated yield results that were higher than the measured bulk variety yields, whereas the default parameters produced consistently lower yields. Biomass was only assessed for 2019, and the results indicate that the biomass after parametrization is lower than the default, which is attributed to the radiation use efficiency parameter being lower in the parametrized files, 2.5 g/MJ versus 2.25 g/MJ. The improved accuracy of predicting yield is evidence that the UAS based methodology is a suitable substitute for the more labor intensive in-situ sampling of biomass for soybean studies under similar environmental conditions.</p> </div><p> <br></p>

In order to accurately conduct the preliminary and detailed design of solar powered Unmanned Aerial Systems (UAS), it is necessary to have a thorough understanding of the systems involved. In particular, it is desirable to have mathematical models and analysis tools describing the energy income and expenditure of the vehicle. Solar energy income models may include available solar irradiance, photovoltaic array power output, and maximum power point tracker efficiency. Energy expenditure models include battery charging and discharging characteristics, propulsion system efficiency, and aerodynamic efficiency. In this thesis, a series of mathematical models were developed that characterize the performance of these systems. Several of these models were then validated against test data. Testing was conducted on specific components used by a solar UAS designed and built by students at the California Polytechnic State University, San Luis Obispo, which completed a six-hour flight relying only on solar energy in May 2019. Results indicate that, while some models accurately predicted test outcomes, others still need further improvement. While these models may be useful during the preliminary and detailed design phases of a solar powered UAS, specific component testing should be conducted to converge on the most desired design solution.

There is a significant amount of ongoing research on developing multi-agent algorithms for mobile robots. Moving those algorithms beyond simulation and into the real world requires multi-robot testbeds. However, there is currently no easily accessible source of information for guiding the creation of such a testbed. In this thesis, we describe the process of creating a testbed at Purdue University involving a set of unmanned aerial vehicles (UAVs). We discuss the components of the testbed, including the software that is used to interface with the UAVs. We also describe the challenges that we faced during the setup process, and evaluate the UAV platforms that we are using. Finally, we demonstrate the implementation of a multi-agent task allocation algorithm on our testbed.

Unmanned aerial systems (UAS) are effective for surveillance and monitoring, but struggle with persistent, long-term tracking, especially without GPS, due to limited flight time. Persistent tracking can be accomplished using multiple vehicles if one vehicle can effectively hand off the tracking information to another replacement vehicle. This work presents a solution to the moving-target handoff problem in the absence of GPS. The proposed solution (a) a nonlinear complementary filter for self-pose estimation using only an IMU, (b) a particle filter for relative pose estimation between UAS using a relative range (c) visual target tracking using a gimballed camera when the target is close to the handoff UAS, and (d) track correlation logic using Procrustes analysis to perform the final target handoff between vehicles. We present hardware results of the self-pose estimation and visual target tracking, as well as an extensive simulation result that demonstrates the effectiveness of our full system, and perform Monte-Carlo simulations that indicate a 97% successful handoff rate using the proposed methods.

No description available.

Unmanned aircraft systems (UASs) pose a potential threat to general aviation/commercial aircraft as UASs are increasingly incorporated into the National Airspace System. This overarching research is aimed at addressing the severity of a UAS mid-air collision with another aircraft. This study is primarily focused on the development of a finite element (FE) model of a ~4 lb fixed-wing UAS (FW-UAS) which will be used to evaluate the severity of small UAS mid-air collisions to manned aircraft. A series of impact tests were performed at the University of Dayton Research Institute - Impact Physics Lab, to study the impact behavior of the high-density components of the FW-UAS (i.e., motor, and battery). For each of the tests, a simulation was set up with the same initial conditions, and boundary conditions as the physical test and the same output parameters were correlated with the test results. A series of numerical stability checks were also performed using the validated FW-UAS FE model to ensure the stability of the explicit dynamic procedures. Simulated impacts between the FW-UAS FE model and targets (deformable flat plate, rigid flat plate, and rigid knife-edge) were performed as stability checks. The FW-UAS FE model developed in this work facilitated the evaluation of the severity of FW-UAS mid-air collision to commercial and business jet airframes performed at and in conjunction with National Institute for Aviation Research. A series of worst-case scenarios involving impacts between the FW-UAS and commercial narrow-body transport and business jet airframes were simulated. For each simulated impact, an impact severity index value was assigned to characterize the relative threat to a given aircraft. In addition, a UAS frangibility study was performed to assess key UAS design features that result in reduced damage to target air vehicles. A “pusher” engine configuration was modeled where the high-density motor is located aft of the UAS’s forward fuselage. Positioning the high-density motor in the aft fuselage played an important role in reducing the impact damage severity.

Operator responsiveness to critical alarm/alert display systems must rely on faster and safer behavioral responses in order to ensure mission success in complex environments such as the operator station of an Unmanned Aerial System (UAS). An important design consideration for effective UAS interfaces is how to map these critical alarm/alert display systems to an appropriate sensory modality (e.g., visual or auditory) (Sarter, 2006). For example, if an alarm is presented during a mission in a modality already highly taxed or overloaded, this can result in increased response time (RT), thereby decreasing operator performance (Wickens, 1976). To overcome this problem, system designers may allow the switching of the alarm display from a highly-taxed to a less-taxed modality (Stanney et al., 2004). However, this modality switch may produce a deleterious effect known as the Modality Shifting Effect (MSE) that erodes the expected performance gain (Spence & Driver, 1997). The goal of this research was to empirically examine a technique called bridging which allows the transitioning of a cautionary alarm display from one modality to another while simultaneously counteracting the Modality Shifting Effect. Sixty-four participants were required to complete either a challenging visual or auditory task using a computer-based UAS simulation environment while responding to both visual and auditory alarms. An approach was selected which utilized two 1 (task modality) x 2 (switching technique) ANCOVAs and one 2 (modality) x 2 (technique) ANCOVA, using baseline auditory and visual RT as covariates, to examine differences in alarm response times when the alert modality was changed abruptly or with the bridging technique from a highly loaded sensory channel to an underloaded sensory channel. It was hypothesized that the bridging technique condition would show faster response times for a new unexpected modality versus the abrupt switching condition. The results indicated only a marginal decrease in response times for the auditory alerts and a larger yet not statistically significant effect for the visual alerts; results were also not statistically significant for the analysis collapsed across modality. Findings suggest that there may be some benefit of the bridging technique on performance of alarm responsiveness, but further research is still needed before suggesting generalizable design guidelines for switching modalities which can apply in a variety of complex human-machine systems.

Anomaly detection aims to identify the data samples that do not conform to a known normal (regular) behavior. As the definition of an anomaly is often ambiguous, unsupervised and semi-supervised deep learning (DL) algorithms that primarily use unlabeled datasets to model normal (regular) behaviors, are popularly studied in this context. The unmanned aerial system (UAS) can use contextual anomaly detection algorithms to identify interesting objects of concern in applications like search and rescue, disaster management, public security etc. This thesis presents a novel multi-stage framework that supports detection of frames with unknown anomalies, localization of anomalies in the detected frames, and validation of detected frames for incremental semi-supervised learning, with the help of a human operator. The proposed architecture is tested on two new datasets collected for a UAV-based system. In order to detect and localize anomalies, it is important to both model the normal data distribution accurately as well as formulate powerful discriminant (anomaly scoring) techniques. We implement a generative adversarial network (GAN)-based anomaly detection architecture to study the effect of loss terms and regularization on the modeling of normal (regular) data and arrive at the most effective anomaly scoring method for the given application. Following this, we use incremental semi-supervised learning techniques that utilize a small set of labeled data (obtained through validation from a human operator), with large unlabeled datasets to improve the knowledge-base of the anomaly detection system. / Master of Science / Anomaly detection aims to identify the data samples that do not conform to a known normal (regular) behavior. As the definition of an anomaly is often ambiguous, most techniques use unlabeled datasets, to model normal (regular) behaviors. The availability of large unlabeled datasets combined with novel applications in various domains, has led to an increasing interest in the study of anomaly detection. In particular, the unmanned aerial system (UAS) can use contextual anomaly detection algorithms to identify interesting objects of concern in applications like search and rescue (SAR), disaster management, public security etc. This thesis presents a novel multi-stage framework that supports detection and localization of unknown anomalies, as well as the validation of detected anomalies, for incremental learning, with the help of a human operator. The proposed architecture is tested on two new datasets collected for a UAV-based system. In order to detect and localize anomalies, it is important to both model the normal data distribution accurately and formulate powerful discriminant (anomaly scoring) techniques. To this end, we study the state-of-the-art generative adversarial networks (GAN)-based anomaly detection algorithms for modeling of normal (regular) behavior and formulate effective anomaly detection scores. We also propose techniques to incrementally learn the new normal data as well as anomalies, using the validation provided by a human operator. This framework is introduced with the aim to support temporally critical applications that involve human search and rescue, particularly in disaster management.

Breeding compact cotton plants is desirable for many reasons, but current research for this is restricted by manual data collection. Using unmanned aircraft system imagery shows potential for high-throughput automation of this process. Using multispectral orthomosaics and ground truth measurements, I developed supervised models with a wide range of hyperparameters to predict three compactness traits. Extreme gradient boosting using a feature matrix as input was able to predict the height-related metric with R2=0.829 and RMSE=0.331. The breadth metrics require higher-detailed data and more complex models to predict accurately.

Nitrogen (N) is the most critical fertilizer applied nutrient for supporting plant growth. It is a critical part of photosynthesis as a component of chlorophyl, hence it is a key indicator of plant health. In recent years, rapid development of multispectral sensing technology and machine learning (ML) methods make it possible to estimate leaf chemical components such as N for predicting yield spatially and temporally. The objectives of this study were to compare the relationships between canopy reflectance and corn (Zea mays L.) leaf N concentration acquired by two multispectral sensors: red-edge multispectral camera mounted on the Unmanned Aerial Vehicle (UAV) and crop circle ACS-430. Four fertilizer N rates were applied, ranging from deficient to excessivein order to have a broad rangein plant N status. Spectral information was collected at different phenological stages of corn to calculate vegetation indices (VIs) for each stage. Moreover, leaf samples were taken simultaneously to determine N concentration. Different ML methods (Multi-Layer Perceptron (MLP), Support Vector Machines (SVMs), Random Forest regression, Regularized regression models, and Gradient Boosting) were used to estimate leaf N% from VIs and predict yield from VIs. Random Forest Regression was utilized as a feature selection method to choose the best combination of variables for different stages and to interpret the relationships between VIs and corn leaf N concentration and grain yield. The Canopy Chlorophyll Content Index (SCCCI) and Red-edge Ratio Vegetation Index (RERVI) were selected as the most efficient VIs in leaf N estimation and SCCCI, Red-edge chlorophyll index (CIRE), RERVI, Soil Adjusted Vegetation Index (SAVI), and Normalized Difference Vegetation Index (NDVI) were chosen as the most effective VIs in predicting corn grain yield. The results derived from using a red-edge multispectral camera showed that the SCCCI was the most proper index for predicting yield at most of the phenological stages and Gradient Boosting was the best-fitted model to estimate leaf N% with an 80% coefficient of determination. Using a Crop Circle ACS-430 showed that the Support Vector Regression (SVR) model achieved the best performance measures than other models tested in the prediction of leaf N concentration.