Matthew A. Lee

Early Detection of Crop Injury from Glyphosate on Soybean and Cotton Using Plant Leaf Hyperspectral Data

In this paper, we aim to detect crop injury from glyphosate, a herbicide, by both traditionally used spectral indices and newly extracted features with leaf hyperspectral reflectance data for non-Glyphosate-Resistant (non-GR) soybean and non-GR cotton. The new features were extracted by canonical analysis technique, which could provide the largest separability to distinguish the injured leaves from the healthy ones. Spectral bands used for constructing these new features were selected based on the sensitivity analysis results of a physically-based leaf radiation transfer model (leaf optical PROperty SPECTra model, PROSPECT), which could help extend the effectiveness of these features to a wide range of leaf structures and growing conditions. This approach has been validated with greenhouse measured data acquired in glyphosate treatment experiments. Results indicated that glyphosate injury could be detected by NDVI (Normalized Difference Vegetation Index), RVI (Ratio Vegetation Index), SAVI (Soil Adjusted Vegetation Index), and DVI (Difference Vegetation Index) in 48 h After the Treatment (HAT) for soybean and in 72 HAT for cotton, but the other spectral indices either showed little use for separation, or did not show consistent separation for healthy and injured soybean and cotton. Compared with the traditional spectral indices, the new features were more feasible for the early detection of glyphosate injury, with leaves sprayed with a higher rate of glyphosate solution having larger feature values. This trend became more and more pronounced with time. Leaves sprayed with different glyphosate rates showed some separability 24 HAT using the new features and could be totally distinguished at and beyond 48 HAT for both soybean and cotton. These findings demonstrated the feasibility of applying leaf hyperspectral reflectance measurements for the early detection of glyphosate injury using these newly proposed features.

Decision Fusion of Textural Features Derived From Polarimetric Data for Levee Assessment

Texture features derived from Synthetic Aperture Radar (SAR) imagery using grey level co-occurrence matrix (GLCM) can result in very high dimensional feature spaces. Although this high dimensional texture feature space can potentially provide relevant class-specific information for classification, it often also results in over-dimensionality and ill-conditioned statistical formulations. In this work, we propose a polarization channel based feature grouping followed by a multi-classifier decision fusion (MCDF) framework for a levee health monitoring system that seeks to detect landslides in earthen levees. In this system, texture features derived from the SAR imagery are partitioned into small groups according to different polarization channels. A multi-classifier system is then applied to each group to perform classification at the subspace level (i.e., a dedicated classifier for every subspace). Finally, a decision fusion system is employed to fuse decisions generated by each classifier to make a final classification decision (healthy levee versus landslide in this work). The resulting system can handle the high dimensionality of the problem very effectively, and only needs a few training samples for training and optimization.

Applying cellular automata to hyperspectral edge detection

This paper proposes the concept of using cellular automata (CAs) and adapted edge detection algorithms for edge detection in hyperspectral images. The approach consists of an edge detection CA and a post-processing CA (that implements morphological operations for denoising the edges). The CA approach is generalized in that it operates on any three-dimensional data cube, allowing for hyperspectral dimensionality reduction as a pre-processing stage if preferred. The CA approach is designed, implemented, and applied to airborne hyperspectral imagery for qualitative assessment and applied to synthetic imagery, where the precise ground truth of edges are known, for quantitative assessment. Results show the CA method to be very promising for both unsupervised and supervised edge detection in hyperspectral imagery.

Sensitivity of hyperspectral classification algorithms to training sample size

Algorithms that exploit hyperspectral imagery often encounter problems related to the high dimensionality of the data, particularly when the amount of training data is limited. Recently, two algorithms were proposed to alleviate the small sample size problem - one is based on employing a Multi-Classifier Decision Fusion (MCDF) in the raw reflectance domain, and the other employed the MCDF framework in the Discrete Wavelet Transform domain (DWT-MCDF). This paper investigates the sensitivity of conventional single classifier based classification approaches, as well as MCDF and DWT-MCDF to variations in the amount of data employed for training the classification system. The hyperspectral data in this experiment was obtained using an airborne hyperspectral imager used by SpecTIR™. The results of the experimental analysis show that for the given application, the MCDF and DWT-MCDF algorithms are significantly less sensitive than the conventional algorithms to limited training data. PCA consistently results in overall accuracies of about 35%. LDA accuracies are very high, about 75%, when there is an abundance of training data - about 10X (i.e. number of training samples is 10 times the number of spectral bands); remains above 60% for training data abundances of 2X and higher; but dramatically decreases to ∼20% for abundances of 1X. MCDF results in accuracies ranging between 65% and 75% for training data abundance of 3X and higher, but the accuracies drop to ∼60% for 2X and ∼55% for 1X. DWT-MCDF results in high accuracies with the least sensitivity to training data abundance. Its accuracies range between ∼60–65% for abundances of 1X to 10X.

Detection of slump slides on earthen levees using polarimetric SAR imagery

Key results are presented of an extensive project studying the use of synthetic aperture radar (SAR) as an aid to the levee screening process. SAR sensors used are: (1) The NASA UAVSAR (Uninhabited Aerial Vehicle SAR), a fully polarimetric L-band SAR capable of sub-meter ground sample distance; and (2) The German TerraSAR-X radar satellite, also multi-polarized and featuring 1-meter GSD, but using an X-band carrier. The study area is a stretch of 230 km of levees along the lower Mississippi River. The L-band measurements can penetrate vegetation and soil somewhat, thus carrying some information on soil texture and moisture which are relevant features to identifying levee vulnerability to slump slides. While X-band does not penetrate as much, its ready availability via satellite makes multitemporal algorithms practical. Various feature types and classification algorithms were applied to the polarimetry data in the project; this paper reports the results of using the Support Vector Machine (SVM) and back-propagation Artificial Neural Network (ANN) classifiers with a combination of the polarimetric backscatter magnitudes and texture features based on the wavelet transform. Ground reference data used to assess classifier performance is based on soil moisture measurements, soil sample tests, and on site visual inspections.

Glyphosate-resistant and glyphosate-susceptible Palmer amaranth (Amaranthus palmeri S. Wats.): hyperspectral reflectance properties of plants and potential for classification

BACKGROUND Palmer amaranth (Amaranthus palmeri S. Wats.) is a troublesome agronomic weed in the southern United States, and several populations have evolved resistance to glyphosate. This paper reports on spectral signatures of glyphosate-resistant (GR) and glyphosate-sensitive (GS) plants, and explores the potential of using hyperspectral sensors to distinguish GR from GS plants. RESULTS GS plants have higher light reflectance in the visible region and lower light reflectance in the infrared region of the spectrum compared with GR plants. The normalized reflectance spectrum of the GR and GS plants had best separability in the 400-500 nm, 650-690 nm, 730-740 nm and 800-900 nm spectral regions. Fourteen wavebands from within or near these four spectral regions provided a classification of unknown set of GR and GS plants, with a validation accuracy of 94% for greenhouse-grown plants and 96% for field-grown plants. CONCLUSIONS GR and GS Palmer amaranth plants have unique hyperspectral reflectance properties, and there are four distinct regions of the spectrum that can separate the GR from GS plants. These results demonstrate that hyperspectral imaging has potential application to distinguish GR from GS Palmer amaranth plants (without a glyphosate treatment), with future implications for glyphosate resistance management. Published 2014. This article is a U.S. Government work and is in the public domain in the USA.

Use of remote sensing to screen earthen levees

Multi-polarized L-band Synthetic Aperture Radar is investigated for its potential to screen earthen levees for weak points. Various feature detection and classification algorithms are tested for this application, including both radiometric and textural methods such as grey-level co-occurrence matrix and wavelet features.

Determining the Effects of Storage on Cotton and Soybean Leaf Samples for Hyperspectral Analysis

This paper studies the effect of storage techniques for transporting collected plant leaves from the field to the laboratory for hyperspectral analysis. The strategy of collecting leaf samples in the field for laboratory analysis is typically used when ground truthing is needed in remote sensing studies. Results indicate that the accuracy of hyperspectral measurements depends on a combination of storage technique (in a cooler or outside a cooler), time elapsed between collecting leaf samples in the field and measuring in the laboratory, and the plant species. A nonlinear model fitting method is proposed to estimate the spectrum of decaying plant leaves. This revealed that the reflectance of soybean leaves remained within the normal range for 45 min when the leaves were stored in a cooler, while soybean leaves stored outside a cooler remained within the normal range for 30 min. However, cotton leaves stored in a cooler decayed faster initially. Regardless of storage technique, results indicate that up to a maximum of 30 min can elapse between plant leaf sampling in the field and hyperspectral measurements in the laboratory. This study focused on cotton and soybean leaves, but the implication that time elapsing between sampling leaves and measuring their spectrum should be limited as much as possible can be applied to any study on other crop leaves. Results of the study also provide a guideline for crop storage limits when analyzing by laboratory hyperspectral sensing setting to improve the quality and reliability of data for precision agriculture.

Background adaptive division filtering for hand-held ground penetrating radar

The challenge in detecting explosive hazards is that there are multiple types of targets buried at different depths in a highlycluttered environment. A wide array of target and clutter signatures exist, which makes detection algorithm design difficult. Such explosive hazards are typically deployed in past and present war zones and they pose a grave threat to the safety of civilians and soldiers alike. This paper focuses on a new image enhancement technique for hand-held ground penetrating radar (GPR). Advantages of the proposed technique is it runs in real-time and it does not require the radar to remain at a constant distance from the ground. Herein, we evaluate the performance of the proposed technique using data collected from a U.S. Army test site, which includes targets with varying amounts of metal content, placement depths, clutter and times of day. Receiver operating characteristic (ROC) curve-based results are presented for the detection of shallow, medium and deeply buried targets. Preliminary results are very encouraging and they demonstrate the usefulness of the proposed filtering technique.

Differentiating glyphosate-resistant and glyphosate-sensitive Italian ryegrass using hyperspectral imagery

Glyphosate based herbicide programs are most preferred in current row crop weed control practices. With the increased use of glyphosate, weeds, including Italian ryegrass (Lolium multiflorum), have developed resistance to glyphosate. The identification of glyphosate resistant weeds in crop fields is critical because they must be controlled before they reduce the crop yield. Conventionally, the method for the identification with whole plant or leaf segment/disc shikimate assays is tedious and labor-intensive. In this research, we investigated the use of high spatial resolution hyperspectral imagery to extract spectral curves derived from the whole plant of Italian ryegrass to determine if the plant is glyphosate resistant (GR) or glyphosate sensitive (GS), which provides a way for rapid, non-contact measurement for differentiation between GR and GS weeds for effective site-specific weed management. The data set consists of 226 greenhouse grown plants (119 GR, 107 GS), which were imaged at three and four weeks after emergence. In image preprocessing, the spectral curves are normalized to remove lighting artifacts caused by height variation in the plants. In image analysis, a subset of hyperspectral bands is chosen using a forward selection algorithm to optimize the area under the receiver operating characteristic (ROC) between GR and GS plants. Then, the dimensionality of selected bands is reduced using linear discriminant analysis (LDA). Finally, the maximum likelihood classification was conducted for plant sample differentiation. The results show that the overall classification accuracy is between 75% and 80% depending on the age of the plants. Further refinement of the described methodology is needed to correlate better with plant age.