Yeyin Shi

Unmanned Aerial Vehicles for High-Throughput Phenotyping and Agronomic Research

Advances in automation and data science have led agriculturists to seek real-time, high-quality, high-volume crop data to accelerate crop improvement through breeding and to optimize agronomic practices. Breeders have recently gained massive data-collection capability in genome sequencing of plants. Faster phenotypic trait data collection and analysis relative to genetic data leads to faster and better selections in crop improvement. Furthermore, faster and higher-resolution crop data collection leads to greater capability for scientists and growers to improve precision-agriculture practices on increasingly larger farms; e.g., site-specific application of water and nutrients. Unmanned aerial vehicles (UAVs) have recently gained traction as agricultural data collection systems. Using UAVs for agricultural remote sensing is an innovative technology that differs from traditional remote sensing in more ways than strictly higher-resolution images; it provides many new and unique possibilities, as well as new and unique challenges. Herein we report on processes and lessons learned from year 1—the summer 2015 and winter 2016 growing seasons–of a large multidisciplinary project evaluating UAV images across a range of breeding and agronomic research trials on a large research farm. Included are team and project planning, UAV and sensor selection and integration, and data collection and analysis workflow. The study involved many crops and both breeding plots and agronomic fields. The project’s goal was to develop methods for UAVs to collect high-quality, high-volume crop data with fast turnaround time to field scientists. The project included five teams: Administration, Flight Operations, Sensors, Data Management, and Field Research. Four case studies involving multiple crops in breeding and agronomic applications add practical descriptive detail. Lessons learned include critical information on sensors, air vehicles, and configuration parameters for both. As the first and most comprehensive project of its kind to date, these lessons are particularly salient to researchers embarking on agricultural research with UAVs.

Damage Mapping of Powdery Mildew in Winter Wheat with High-Resolution Satellite Image

Powdery mildew, caused by the fungus Blumeria graminis, is a major winter wheat disease in China. Accurate delineation of powdery mildew infestations is necessary for site-specific disease management. In this study, high-resolution multispectral imagery of a 25 km2 typical outbreak site in Shaanxi, China, taken by a newly-launched satellite, SPOT-6, was analyzed for mapping powdery mildew disease. Two regions with high representation were selected for conducting a field survey of powdery mildew. Three supervised classification methods—artificial neural network, mahalanobis distance, and maximum likelihood classifier—were implemented and compared for their performance on disease detection. The accuracy assessment showed that the ANN has the highest overall accuracy of 89%, following by MD and MLC with overall accuracies of 84% and 79%, respectively. These results indicated that the high-resolution multispectral imagery with proper classification techniques incorporated with the field investigation can be a useful tool for mapping powdery mildew in winter wheat.

Improvement of a ground-LiDAR-based corn plant population and spacing measurement system

We develop a sensing system to measure corn population and within-row spacing using the LiDAR technology.We develop an algorithm to calculate corn plant population and within-row spacing from the laser point cloud data.Total errors on plant counting were less than 6% in no-weed fields.The RMSE of within-row spacing measurements were less than 1.9cm in no weed field. The variability of corn plant location and within-row spacing has been demonstrated to have a significant correlation with grain and biomass yield. They are included in many yield prediction models which are used to guide mid-season variable-rate fertilizer applications. A prototype sensing system was developed to automatically measure corn plant location and spacing on-the-go based on ground LiDAR technology. The system travelled along crop rows with a ground LiDAR sensor scanning at the bottom section of each corn plant. The possibility of corn stalk identification was increased because each stalk appeared in multiple scans from various view angles of the sensor. The first version of the prototyping system was developed earlier and resulted in a relatively low detecting accuracy. In this paper, an improved version of the prototyping system was presented with substantial additional field evaluation results. The system was improved in terms of the data acquisition platform and the data processing algorithms, specifically, the scan registration and stalk recognition procedures to reduce the misidentification errors. Additional field evaluation was conducted on 200 plants at their V8 growth stage. A total plant counting error of 5.5% and a 1.9cm of root-mean-squared error (RMSE) in spacing measurement were achieved between the sensor measurements and the manually measured ground truth data. The new data processing algorithm was also tested on the data collected with the first version system. The false positive plant counting error was reduced from 24.0% with the first version system to 14.0% with the improved algorithms.

Unmanned aerial systems-based remote sensing for monitoring sorghum growth and development

Unmanned Aerial Vehicles and Systems (UAV or UAS) have become increasingly popular in recent years for agricultural research applications. UAS are capable of acquiring images with high spatial and temporal resolutions that are ideal for applications in agriculture. The objective of this study was to evaluate the performance of a UAS-based remote sensing system for quantification of crop growth parameters of sorghum (Sorghum bicolor L.) including leaf area index (LAI), fractional vegetation cover (fc) and yield. The study was conducted at the Texas A&M Research Farm near College Station, Texas, United States. A fixed-wing UAS equipped with a multispectral sensor was used to collect image data during the 2016 growing season (April–October). Flight missions were successfully carried out at 50 days after planting (DAP; 25 May), 66 DAP (10 June) and 74 DAP (18 June). These flight missions provided image data covering the middle growth period of sorghum with a spatial resolution of approximately 6.5 cm. Field measurements of LAI and fc were also collected. Four vegetation indices were calculated using the UAS images. Among those indices, the normalized difference vegetation index (NDVI) showed the highest correlation with LAI, fc and yield with R2 values of 0.91, 0.89 and 0.58 respectively. Empirical relationships between NDVI and LAI and between NDVI and fc were validated and proved to be accurate for estimating LAI and fc from UAS-derived NDVI values. NDVI determined from UAS imagery acquired during the flowering stage (74 DAP) was found to be the most highly correlated with final grain yield. The observed high correlations between UAS-derived NDVI and the crop growth parameters (fc, LAI and grain yield) suggests the applicability of UAS for within-season data collection of agricultural crops such as sorghum.

Development and testing of a customized low-cost unmanned aircraft system based on multispectral and thermal sensing for precision agriculture applications

The ability to conduct useful science under the framework of precision agriculture is not only dependent upon the collection of high quality usable data of plants, soil, and water, but also dependent upon the type of vehicle the sensors are flown on, properly tuned sensors, and the way in which the vehicle is flown. To achieve this capability requires the proper matching and integration of air vehicle, sensors, mission design, and image processing techniques. Although commercial Unmanned Air Systems are starting to be equipped with autopilots, sensors, and simple data processing software, they are often limited to only one sensor, and often lack cross platform integration expandability. This paper develops methodologies and procedures for a highly integrated fixed-wing Unmanned Air System that is customized for precision agriculture science. It addresses sensor selection, vehicle platform selection, flight planning, and data processing procedures. The approach is validated by assessment of collected imagery and data from flights conducted on actual plots. Results presented in the paper show that by comparison to data collected during earlier flights with a non-integrated system, the approach presented here which matches vehicle characteristics to sensor characteristics and employs proper flight planning, mission design, and auto-triggering of the sensor produces better data quality, and improved mosaicking. The approach is judged to be a promising candidate for improved data collection for precision agriculture.

Detection of multi-tomato leaf diseases ( late blight, target and bacterial spots ) in different stages by using a spectral-based sensor

Several diseases have threatened tomato production in Florida, resulting in large losses, especially in fresh markets. In this study, a high-resolution portable spectral sensor was used to investigate the feasibility of detecting multi-diseased tomato leaves in different stages, including early or asymptomatic stages. One healthy leaf and three diseased tomato leaves (late blight, target and bacterial spots) were defined into four stages (healthy, asymptomatic, early stage and late stage) and collected from a field. Fifty-seven spectral vegetation indices (SVIs) were calculated in accordance with methods published in previous studies and established in this study. Principal component analysis was conducted to evaluate SVIs. Results revealed six principal components (PCs) whose eigenvalues were greater than 1. SVIs with weight coefficients ranking from 1 to 30 in each selected PC were applied to a K-nearest neighbour for classification. Amongst the examined leaves, the healthy ones had the highest accuracy (100%) and the lowest error rate (0) because of their uniform tissues. Late stage leaves could be distinguished more easily than the two other disease categories caused by similar symptoms on the multi-diseased leaves. Further work may incorporate the proposed technique into an image system that can be operated to monitor multi-diseased tomato plants in fields.

Field detection of anthracnose crown rot in strawberry using spectroscopy technology

In-field hyper spectral detection of strawberry anthracnose crown rot.33 vegetation indices were investigated with three classification algorithms.Average in-field detection accuracy was 74% for asymptomatic and symptomatic samples. Anthracnose crown rot (ACR) is one of the major diseases affecting strawberry crops grown in warm climates and causes huge yield losses each year. ACR is caused by the fungus Colletotrichum. Since this airborne disease spreads rapidly, detection at the early stage of infection is critical. The objective of this study was to investigate the feasibility of detecting ACR in strawberry at its early stage under field conditions using spectroscopy technology. Hyperspectral data were collected in-field using a mobile platform on three categories of strawberry plants: infected but asymptomatic, infected and symptomatic, and healthy. As a comparison, indoor data were also collected from the same three categories of strawberry plants under a controlled laboratory setup. Three classification models, stepwise discriminant analysis (SDA), Fisher discriminant analysis (FDA), and the k-Nearest Neighbor (kNN) algorithms, were investigated for their potential to differentiate the three infestation categories. Thirty-three spectral vegetation indices (SVIs) were calculated as inputs using selected spectral bands in the visible (VIS) and near infrared (NIR) regions to train classification models. The mean classification accuracies of in-field tests for the three infestation categories were 71.3%, 70.5%, and 73.6% for SDA, FDA, and kNN, respectively. These accuracies were approximately 1520% lower than those of the indoor tests. The low accuracy (15.4%) of classifying healthy leaves in-field using the kNN model was possibly due to the training datasets being unbalanced. After the adjustment of sample sizes of each category, the accuracies of kNN improved greatly, especially for the healthy and symptomatic categories. Overall, SDA was the optimal classifier for both indoor and in-field tests for detection strawberry ACR. However, kNN performed better for asymptomatic leaves in the field in the case of balanced sample sizes of each category.

Mapping wheat rust based on high spatial resolution satellite imagery

Abstract Timely and accurate disease incidence monitoring and estimation using satellite imagery is critical for the effective control of wheat rust disease. Though studies have been conducted using different spectral and sensing technologies to detect and monitor the wheat rust disease, few studies have been conducted in the application of high resolution multi-spectral satellite imagery in the monitoring of wheat rust to facilitate an operational monitoring over large areas. Since there are not many options for the multispectral data than the hyperspectral data, it is more important to select appropriate vegetation indices for the classification model. However, there have been few literatures about the comparison of these feature selection methods on the application of wheat disease. We proposed a wheat rust disease mapping protocol including removing the non-vegetated and lodged area, calculating and filtering the vegetation indices and mapping the wheat rust. In the process of filtering the proper features, we applied wrapper feature selection instead of traditional filter feature selection combined with the classification methods (support vector machine and random forests). The experiment data was a scene of ZY-3 satellite image of the wheat field in Changge County in Henan Province with a certain portion of rust disease. The classification results can achieve overall accuracy of higher than 90%, ranging from 90.80% to 95.10%. The wrapper feature selection method with the overall accuracy of 93.60% is better than filters feature selection method with the overall accuracy of 92.65%. The random forests method with the overall accuracy of 94.80% is better than support vector machine method with the overall accuracy of 91.45%. The high accuracies thus justified the feasibility of using high-resolution multi-spectral satellite images for mapping wheat rust disease, which is promising for this technology to be applied in the practical wheat production management.

A case study of comparing radiometrically calibrated reflectance of an image mosaic from unmanned aerial system with that of a single image from manned aircraft over a same area

Though sharing with many commonalities, one of the major differences between conventional high-altitude airborne remote sensing and low-altitude unmanned aerial system (UAS) based remote sensing is that the latter one has much smaller ground footprint for each image shot. To cover the same area on the ground, it requires the low-altitude UASbased platform to take many highly-overlapped images to produce a good mosaic, instead of just one or a few image shots by the high-altitude aerial platform. Such an UAS flight usually takes 10 to 30 minutes or even longer to complete; environmental lighting change during this time span cannot be ignored especially when spectral variations of various parts of a field are of interests. In this case study, we compared the visible reflectance of two aerial imagery – one generated from mosaicked UAS images, the other generated from a single image taken by a manned aircraft – over the same agricultural field to quantitatively evaluate their spectral variations caused by the different data acquisition strategies. Specifically, we (1) developed our customized ground calibration points (GCPs) and an associated radiometric calibration method for UAS data processing based on camera’s sensitivity characteristics; (2) developed a basic comparison method for radiometrically calibrated data from the two aerial platforms based on regions of interests. We see this study as a starting point for a series of following studies to understand the environmental influence on UAS data and investigate the solutions to minimize such influence to ensure data quality.

Automated geographic registration and radiometric correction for UAV-based mosaics

Texas A and M University has been operating a large-scale, UAV-based, agricultural remote-sensing research project since 2015. To use UAV-based images in agricultural production, many high-resolution images must be mosaicked together to create an image of an agricultural field. Two key difficulties to science-based utilization of such mosaics are geographic registration and radiometric calibration. In our current research project, image files are taken to the computer laboratory after the flight, and semi-manual pre-processing is implemented on the raw image data, including ortho-mosaicking and radiometric calibration. Ground control points (GCPs) are critical for high-quality geographic registration of images during mosaicking. Applications requiring accurate reflectance data also require radiometric-calibration references so that reflectance values of image objects can be calculated. We have developed a method for automated geographic registration and radiometric correction with targets that are installed semi-permanently at distributed locations around fields. The targets are a combination of black (≈5% reflectance), dark gray (≈20% reflectance), and light gray (≈40% reflectance) sections that provide for a transformation of pixel-value to reflectance in the dynamic range of crop fields. The exact spectral reflectance of each target is known, having been measured with a spectrophotometer. At the time of installation, each target is measured for position with a real-time kinematic GPS receiver to give its precise latitude and longitude. Automated location of the reference targets in the images is required for precise, automated, geographic registration; and automated calculation of the digital-number to reflectance transformation is required for automated radiometric calibration. To validate the system for radiometric calibration, a calibrated UAV-based image mosaic of a field was compared to a calibrated single image from a manned aircraft. Reflectance values in selected zones of each image were strongly linearly related, and the average error of UAV-mosaic reflectances was 3.4% in the red band, 1.9% in the green band, and 1.5% in the blue band. Based on these results, the proposed physical system and automated software for calibrating UAV mosaics show excellent promise.