Aflatoxin is produced by the fungus Aspergillus flavus when the fungus invades developing corn kernels. Because of its
potent toxicity, the levels of aflatoxin are regulated by the Food and Drug Administration (FDA) in the US, allowing 20
ppb (parts per billion) limits in food, and feed intended for interstate commerce. Currently, aflatoxin detection and
quantification methods are based on analytical tests. These tests require the destruction of samples, can be costly and
time consuming, and often rely on less than desirable sampling techniques. Thus, the ability to detect aflatoxin in a rapid,
non-invasive way is crucial to the corn industry in particular. This paper described how narrow-band fluorescence
indices were developed for aflatoxin contamination detection based on single corn kernel samples. The indices were
based on two bands extracted from full wavelength fluorescence hyperspectral imagery. The two band results were later
applied to two large sample experiments with 25 g and 1 kg of corn per sample. The detection accuracies were 85% and
95% when 100 ppb threshold was used. Since the data acquisition period is significantly lower for several image bands
than for full wavelength hyperspectral data, this study would be helpful in the development of real-time detection
instrumentation for the corn industry.
Fluorescence hyperspectral imaging is increasingly being used for food quality inspection and detection of potential food
safety concerns. The flexible nature of a self-scanning pushbroom hyperspectral imager lends itself to these kinds of
applications, among others. To increase the use of this technique there has been a tendency to use low cost off-the-shelf
hyperspectral sensors which are typically not radiometrically calibrated. To ensure that these systems are optimized for
response and repeatability, it is imperative that the systems be both radiometrically and spectrally calibrated specifically
for fluorescence imaging. Fluorescence imaging provides several challenges such as low signal, stray light and a low
signal dynamic range that are improved with careful radiometric calibration. A radiometric and spectral approach that includes flat fielding and the conversion of digital number responses to radiance for calibrating this imaging system and other types of hyperspectral imagers is described in this paper. Results show that this method can be adopted for calibrating fluorescence and reflective hyperspectral imaging systems in the visible and near infra-red domains.
Aflatoxins are toxic secondary metabolites of the fungi Aspergillus flavus and Aspergillus parasiticus, among others.
Aflatoxin contaminated corn is toxic to domestic animals when ingested in feed and is a known carcinogen associated
with liver and lung cancer in humans. Consequently, aflatoxin levels in food and feed are regulated by the Food and
Drug Administration (FDA) in the US, allowing 20 ppb (parts per billion) limits in food and 100 ppb in feed for
interstate commerce. Currently, aflatoxin detection and quantification methods are based on analytical tests including
thin-layer chromatography (TCL) and high performance liquid chromatography (HPLC). These analytical tests require
the destruction of samples, and are costly and time consuming. Thus, the ability to detect aflatoxin in a rapid, nondestructive
way is crucial to the grain industry, particularly to corn industry. Hyperspectral imaging technology offers a
non-invasive approach toward screening for food safety inspection and quality control based on its spectral signature.
The focus of this paper is to classify aflatoxin contaminated single corn kernels using fluorescence hyperspectral
imagery. Field inoculated corn kernels were used in the study. Contaminated and control kernels under long wavelength
ultraviolet excitation were imaged using a visible near-infrared (VNIR) hyperspectral camera. The imaged kernels were
chemically analyzed to provide reference information for image analysis. This paper describes a procedure to process
corn kernels located in different images for statistical training and classification. Two classification algorithms,
Maximum Likelihood and Binary Encoding, were used to classify each corn kernel into "control" or "contaminated" through pixel classification. The Binary Encoding approach had a slightly better performance with accuracy equals to 87% or 88% when 20 ppb or 100 ppb was used as classification threshold, respectively.
Aflatoxin is a mycotoxin predominantly produced by Aspergillus flavus and Aspergillus parasitiucus fungi that grow
naturally in corn, peanuts and in a wide variety of other grain products. Corn, like other grains is used as food for human
and feed for animal consumption. It is known that aflatoxin is carcinogenic; therefore, ingestion of corn infected with
the toxin can lead to very serious health problems such as liver damage if the level of the contamination is high. The US
Food and Drug Administration (FDA) has strict guidelines for permissible levels in the grain products for both humans
and animals. The conventional approach used to determine these contamination levels is one of the destructive and
invasive methods that require corn kernels to be ground and then chemically analyzed. Unfortunately, each of the
analytical methods can take several hours depending on the quantity, to yield a result. The development of high spectral
and spatial resolution imaging sensors has created an opportunity for hyperspectral image analysis to be employed for
aflatoxin detection. However, this brings about a high dimensionality problem as a setback. In this paper, we propose a
technique that automatically detects aflatoxin contaminated corn kernels by using dual-band imagery. The method
exploits the fluorescence emission spectra from corn kernels captured under 365 nm ultra-violet light excitation. Our
approach could lead to a non-destructive and non-invasive way of quantifying the levels of aflatoxin contamination. The
preliminary results shown here, demonstrate the potential of our technique for aflatoxin detection.
Accurate retrieval of wildland fire temperature from remote imagery would be useful in improving prediction of fire propagation and estimates of fire effects such as burn severity and gas and particle production. The feasibility of estimating temperatures for subpixel fires by spectral unmixing has been established by previous work with
the AVIRIS sensor. However, this unmixing approach can also produce optimizations for temperatures that may not be physically related to the fraction of flaming combustion in a pixel. Furthermore, previous techniques have treated fire as a blackbody and have modeled the mixed pixel transmitted radiance as two blackbody sources. This first order approximation can also affect the temperature retrieval. Knowledge of emissivity and use of a more complex radiance model should improve the accuracy of the temperature estimation. We therefore, propose a technique which improves the previous approach by using the potassium emission to pre-determine pixels that actually contain signal from flaming combustion and a modified mixed pixel radiance model. A non-linear, constrained multi-dimensional optimization procedure which estimates flame emissivity was applied
to the model to estimate fire temperature and its areal extent. Results are shown for AVIRIS data sets acquired over Cuiaba, Brazil (1995) and the San Bernardino Mountains (1999).
The purpose of this paper is to describe a physics based fire model in DIRSIG. The main objective is to utilize research on radiative emissions from fire to create a 3D rendering of a scene to generate a synthetic multispectral or hyperspectral image of wildfire. These synthetic images can be used to evaluate detection algorithms and sensor platforms.
To produce realistic flame structures and realistic spectral emission across the visible and infrared spectrum, we first need to produce 3D time-dependent data describing the fire evolution and its interaction with the environment. Here we utilize an existing coupled atmosphere-fire model to represent the finescale dynamics of convective processes in a wildland fire. Then the grid-based output from the fire propagation model can be used in DIRSIG along with the spectral emission representative of a wildland fire to run the ray-tracing model to create the synthetic scene.
The technical approach is based on a solid understanding of user requirements for format and distribution of the information provided by a high spatial resolution remote sensing system.
Typical existing fire detection algorithms for airborne and satellite based imagers employ the Planckian radiation in the 3.5 -5 μm and 8 - 14 μm spectral regions. These algorithms can have high false alarm rates and furthermore, the issue of validation of subpixel detection is a lingering problem. We present an empirical testing of fire detection algorithms for controlled and uniform burning and hot targets of known area. Image data sets of the targets were captured at different altitudes with the Modular Imaging Spectrometer Instrument (MISI). MISI captures hyperspectral
VNIR and multispectral SWIR/MWIR/LWIR imagery. The known range of target areas ranges from larger than the MISI IFOV to less than 0.5% of the IFOV. The in situ temperatures were monitored with thermocouples and pyrometers. Spectroradiometric data of targets and backgrounds were also collected during the experiment. The data were analysed using existing algorithms as well as novel approaches. The algorithms are compared by determining the minimum resolvable
fire pixel fraction.
Fire detection has been an active research field for many years and a number of algorithms have been proposed. These algorithms, however, are often inflexible in dealing with the spatial and temporal heterogeneity of the environment. Different biomes, seasons, and temperatures usually cause the performance of these algorithms to vary dramatically. In this paper, we propose a new algorithm for fire detection based on the Mahalanobis distance that exploits the statistical properties of multi-spectral images. The distinguishing feature of our algorithm is its robustness. It can effectively differentiate fire from background in various environments, using a single, fixed threshold. We evaluate our algorithm by comparing it to three state-of-the-art existing algorithms: the MODVOLC normalized fire index algorithm, the Arino's threshold algorithm, and the contextual MODIS algorithm. All algorithms are tested using MODIS images taken in different parts of the world as well as at different times. Experimental results demonstrate that our algorithm consistently achieves the best performance, showing a low and constant false alarm rate.