An unsupervised method for plant species identification was developed which uses computer extracted individual whole leaves from color images of crop canopies. Green canopies were isolated from soil/residue backgrounds using a modified Excess Green and Excess Red separation method. Connected components of isolated green regions of interest were changed into pixel fragments using the Gustafson-Kessel fuzzy clustering method. The fragments were reassembled as individual leaves using a genetic optimization algorithm and a fitness method. Pixels of whole leaves were then analyzed using the elliptic Fourier shape and Haralick's classical textural feature analyses. A binary template was constructed to represent each selected leaf region of interest. Elliptic Fourier descriptors were generated from a chain encoding of the leaf boundary. Leaf template orientation was corrected by rotating each extracted leaf to a standard horizontal position. This was done using information provided from the first harmonic set of coefficients. Textural features were computed from the grayscale co-occurrence matrix of the leaf pixel set. Standardized leaf orientation significantly improved the leaf textural venation results. Principle component analysis from SAS (R) was used to select the best Fourier descriptors and textural indices. Indices of local homogeneity, and entropy were found to contribute to improved classification rates. A SAS classification model was developed and correctly classified 83% of redroot pigweed, 100% of sunflower 83% of soybean, and 73% of velvetleaf species. An overall plant species correct classification rate of 86% was attained.
Fuzzy excess green (ExG) crisp indices and clustering algorithms such as the Gustafson-Kessel (GK) have been
successfully used for unsupervised classification of hidden and prominent regions of interest (ROI’s), namely green
plants in crop color images against bare clay soil, corn residue and wheat residue, typical of the Great Plains. Each
process can be enhanced with Zadeh (Z) and Gath-Geva (GG) fuzzy enhancement techniques. Enhanced indices and
clusters can be then sorted by final degree of fuzziness, and recombined into labeled, false-color class images, which
can be used as templates for further shape and textural analyses. ROI’s with the lowest degree of fuzziness were
consistently found to be plant clusters according to foveated or prominence of the region size within the image.
Clustering performance according to partition densities and hyper volume was also evaluated. These latter measures can
be used to select the number of clusters and evaluate the computational time needed to find plant ROI’s with complex
backgrounds under different lighting conditions. Enhanced GK clustering methods have performed very well and have
identified plants in bare soil, corn residue plants , and wheat straw plants, well into the high 90 percentages, depending
on plant age category and the relative proportion of plant size within the image. Improved clustering algorithms with
textural finger printing could be potentially useful for unsupervised remote sensing, mapping, crop management, weed,
and pest control for precision agriculture.
This paper summarizes the theory of fuzzy inference systems and its application to plant and weed detection. Two simple examples are presented, both of which discriminate between plants and soil and residue backgrounds in color images based on derived excess green and excess red color indices. The first example shows that a numerical excess red model can be readily replaced with a fuzzy inference system, based on training of red, green, and blue inputs and excess red. The second example shows an arbitrary system of fuzzy inference with excess red and excess green using human preselection, which also gives satisfactory discrimination results.
Machine vision based on classical image processing techniques has the potential to be a useful tool for plant detection and identification. Plant identification is needed for weed detection, herbicide application or other efficient chemical spot spraying operations. The key to successful detection and identification of plants as species types is the segmentation of plants form background pixel regions. In particular, it would be beneficial to segment individual leaves form tops of canopies as well. The segmentation process yields an edge or binary image which contains shape feature information. Results indicate that red-green-blue formats might provide the best segmentation criteria, based on models of human color perception. The binary image can be also used as a template to investigate textural features of the plant pixel region, using gray image co-occurrence matrices. Texture features considers leaf venation, colors, or additional canopy structure that might be used to identify various type of grasses or broadleaf plants.
A significant reduction in the amount of pesticides applied in agricultural and biological systems could be achieved using spot spray technology. To accomplish this, advanced plant sensor systems must be developed that can accurately locate and identify weeds from crop plants in the field. Currently, both public and commercial efforts have concentrated on single element optical sensors based on key reflective elements of the plant and soil system. Machine vision or image analysis is being investigated as another possible tool in plant sensing. It may provide valuable optical design information for less expensive single-element sensors. Moreover, shape features and textural analysis already provides simple broadleaf-grass classification based on staged plant images. These have not been thoroughly field tested. Another approach is to test image analysis algorithms, using three-dimensional rendering of weed and plant canopy architecture under complex lighting regimes. What was essentially done was to extract plant shape and textural information, along with essential physiological data from actual photographic images and then reassemble them as a virtual plant in the computer. A dissection program was written in C and efficiently extracts and stores irregular leaf shape and texture data. A canopy architecture program was written in C and Media Cybernetics HALOR graphics routines under DOS Expanded Memory on a personal computer. The plant simulation model consists of a three dimensional space where simulated light rays are generated as diffuse or speculative illumination. Plant surfaces are simulated with actual textural maps. The virtual plant is then manipulated to generate images that would be seen with machine vision. Computer simulated weed images were used to generate and test different fields of view sizes for evaluating how single element optical sensors would respond to composite leaf-soil reflectance.
Optical plant sensors constructed from red and near-infrared (NIR) filtered photodetector pairs were used in conjunction with the normalized difference index (NDI) to detect plants. Plants must occupy a minimum of 8% of the photodetector pair field-of-view (FOV) to be detected. Thus, knowing the size and location of the FOV is crucial. Since the NDI requires red and NIR reflectance measurements from coincident areas, it is equally important to know the coincident area of a red-NIR detector pair and its location. Reflectance measurements taken every 1 cm from a 60 cm X 60 cm surface can be graphically viewed to determine the size and location of the FOV of a plant sensor. The surface has low reflectance and contains a 20 cm X 20 cm highly reflective checkerboard pattern in the center. From individual FOVs of red-NIR pairs, the coincident area can be found.
Crop surface temperature under a radiant heated greenhouse was measured using a portable infrared thermometer. Plants were arranged so that each plant occupied a grid cell of 30 cm X 30 cm (1 ft X 1 ft). Data collected were analyzed for their spatial distribution. Geostatistical software was used to characterize the spatial variability of the plant surface temperature. The shape of the empirical semi-variogram suggested that a spherical model was best fitted to the empirical semi-variogram. This model indicated that the nugget effect was estimated at 1.2, the sill at 3.3 and the range at 1.65 m. This model was used in block kriging to estimate plant surface temperature for unsampled locations.
Shape parameters such as aspect, roundness, and the ratio of thickness to perimeter were used to describe plant shape and are different according to the species that they represent. Color slide images of several species of plants were digitized for computer analysis. Three optical methods were tested to separate target plants from the soil and residue background. The separation method that provided the best contrast was the normalized difference index. Subtracting the blue or the red raster from the green raster also provided good separation on soils with little residue. Once the plant image had been isolated from the background, leaf edges were automatically traced using a commercial software package. Analysis of the shape of the plant outline was then performed, resulting in the plant shape parameters. Grasses and broadleaf plants had similar values for each shape parameter during the first ten days after emergence. After this period, differences occurred between grasses and broadleaf plants. The parameter that best discriminated grasses from broadleaf plants was the aspect (major axis length/minor axis length). However, when a grass sends out more than one shoot radially from the stem, the aspect will be similar to broadleaf plants. This study contributes to the design of a system that can determine weed populations and identify plant species without the use of human intervention.
Growing plants, soil types, and surfaces and residues on a soil surface have distinct natural light reflectances. These reflectance characteristics have been determined using current spectroradiometry technology. Detection of plants is possible based upon the distinct reflectance characteristics of plants, soil, and residues. An optical plant reflectance sensor was developed which utilizes a pair of red and near infrared sensitive photodetectors to measure the radiancy from the plant and soil. Another pair of sensors measures radiancy from a highly radiant reference surface to accommodate varying intensities of the natural light. The ratio of the target and reference radiancies is the target reflectance. Optical filters were used to select the spectral bandwidth sensitivities for the red and NIR photodetectors. The reflectance values were digitized for incorporation into a normalized difference index in order to provide a stronger indication that a live plant is present within the field of view of the sensor. This sensor system was combined with a microcontroller for activating a solenoid controlled spray nozzle on a single unit prototype spot agricultural sprayer.
Understanding the process of sprinkler droplet formation and behavior while in flight is crucial to the improvement of water and chemical application by sprinkler irrigation. A relatively simple and inexpensive method is presented which allows in-flight characterization of irregularly shaped liquid droplets and ligaments produced by irrigation sprinkler devices. A high-speed photographic image acquisition probe was constructed to allow nonintrusive, direct investigation of water jet breakup, drop formation and behavior, size, and shape of monodispersed drops at various locations from the nozzle. Image contrast and motion freezing was accomplished by backlighting the liquid particles with a 1.5 microsecond(s) duration stroboscope. High resolution film allowed analysis over a wide range of particle sizes (0.3 mm to 50 mm). Image processing and analysis was performed on digitized images using commercially available software. This software provided geometrical size and shape parameters of the breakup fragments and allowed discrimination between in and out-of-focus drops by post analysis. Applications of the method included: studying the effect of wind direction and speed, nozzle type, and pressure on sprinkler jet breakup.
Various vision methods for inspecting the growth and quality of poinsettia plants are discussed in this paper . The visible and near-infrared vision approaches are based on previous spectral reflectance measurements . Low (0 ppm) nitrogen plants grown in a greenhouse showed an increase in red (0. 7 - 0. 75 rim) and a decrease in near-infrared ( 0 . 8 - 1 . 1 im) reflectance over high ( 256 ppm) nitrogen levels . Growth chamber plants showed similar reflectance in the red but different NIR reflectance than with greenhouse plants . NIR reflectance was affected by vegetative density and not by leaf nitrogen content. Thermal imaging techniques (12 - 14 im) improve canopy temperature measurements . The usefulness of image methods depends on reflectivity analog-digital sensitivity and background lighting quality. An electronic plant doctor based on a database of images would be a useful tool for the grower to perform visual diagnostics. 1.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.