Image classification tasks leverage CNN to yield accurate results that supersede their predecessor human-crafted algorithms. Applicable use cases include Autonomous, Face, Medical Imaging, and more. Along with the growing use of AI image classification applications, we see emerging research on the robustness of such models to adversarial attacks, which take advantage of the unique vulnerabilities of the Artificial Intelligence (AI) models to skew their classification results. While not visible to the Human Visual System (HVS), these attacks mislead the algorithms and yield wrong classification results. To be incorporated securely enough in real-world applications, AI-based image classification algorithms require protection that will increase their robustness to adversarial attacks. We propose replacing the commonly used Rectifier Linear Unit (ReLU) Activation Function (AF), which is piecewise linear, with non-linear AF to increase their robustness to adversarial attacks. This approach has been considered in recent research and is motivated by the observation that non-linear AF tends to diminish the effect of adversarial perturbations in the DNN layers. To gain credibility of the approach, we have applied Fast Sign Gradient Method (FGSM), and Hop-Skip- Jump (HSJ) attacks to a trained classification model of the MNIST dataset. We then replaced the AF of the model with non-linear AF (Sigmoid, GeLU, ELU, SeLU, and Tanh). We concluded that while attacks on the original model have a 100% success rate, the attack success rate is dropped by an average of 10% when non-linear AF is used.
The BGU CubeSat satellite is from a class of low-cost, compact satellites. Its dimensions are 10×10×30 cm. It is equipped with a low resolution 256×320 pixels short wave infrared (SWIR) camera at the 1.55-1.7mm wavelength band. Images are transmitted in bursts of tens of images at a time with few pixel shifts from the first image to the last. Each image burst is suitable for Multiple Image Super Resolution (MISR) enhancements. MISR can construct a high-resolution (HR) image from several low-resolution (LR) images yielding an image that can resolve more details that are crucial for research in remote sensing. In this research, we verify the applicability of SOTA deep learning MISR models that were developed following the publication of the PROBA-V MISR satellite dataset at the visible red and near IR. Our SWIR multiple images differ from PROBA-V by the spectral band and by the method of collecting multiple images of the exact location. Our imagery data is acquired by a burst of very close temporal images. PROBA-V revisits the satellite at a period smaller than 30 days, assuming the soil dryness is about the same. We compare the results of Single Image Super-Resolution (SISR) and MISR techniques to "off-the-shelf" products. The quality of the super-resolved images is compared by nonreference metrics suitable for remote sensing applications and by experts' visual inspection. Unlike remarkable achievements by the GAN technique that can achieve very appealing results that are not always faithful to the original ground truth, the super-resolved images should preserve the original details as much as possible for further scientific remote sensing analysis.
It is anticipated that in some extreme situations, autonomous cars will benefit from the intervention of a ״Remote Driver״. The vehicle computer may discover a failure and decide to request remote assistance for safe roadside parking. In a more extreme scenario, the vehicle may require a complete remote-driver takeover due to malfunctions or an inability to resolve unknown decision logic. In such cases, the remote driver will need a sufficiently good quality real-time video stream of the vehicle cameras to respond quickly and accurately enough to the situation at hand. Relaying such a video stream to the remote Command and Control (C&C) center is especially challenging when considering the varying wireless channel bandwidths expected in these scenarios. This paper proposes an innovative end-to-end content-sensitive video compression scheme to allow efficient and satisfactory video transmission from autonomous vehicles to the remote C&C center.
A recent paper by Belen'kii proposed an inner scale turbulence MTF theory model to explain one of the many types of aerosol MTF experimental results by Dror, Sadot, and Kopeika. A broad comparison is made here between Belen'kii's inner scale turbulence MTF and our practical instrumentation-based aerosol MTF models and experiments. Belen'kii's model is strongly contradicted by those experimental results themselves, as well as by the many other published results not considered by Belen'kii. The Dror, Sadot, Kopeika experiments all contain comparisons of practical instrumentation-based aerosol MTF modeling which was calculated from aerosol size distributions actually measured during the experiments. These aerosol MTF calculations validated rather accurately the practical aerosol MTF model measurements. A summary of 9 independent different types of measurements and analyses which show we measured aerosol MTF and not any form of turbulence MTF is included, as well as specific contradictions between Belen'kii's model and our actual measurements. These also indicate our measurements of turbulence MTF were indeed correct, and that a broad system engineering approach to atmospheric optics should be encouraged instead of the narrow pure turbulence, pure aerosol, or pure absorbing atmospheric models often used.
The recently developed atmospheric Wiener filter, which corrects for turbulence and aerosol blur and path radiance simultaneously, is implemented in digital restoration of AVHR imagery over the five wavelength bands of the satellite instrumentation. Restoration is most impressive for higher optical depth situations which cause more blur, with improvement in regard to both smallness of size of resolvable detail and contrast. Turbulence modulation transfer function (MTF) is calculated from meteorological data. Aerosol MTF is consistent with optical depth, measured with a sum-photometer. The product of the two yields atmospheric MTF which is implemented in the atmospheric Wiener filter. Turbulence blur, aerosol blur, and path radiance contrast loss are all corrected simultaneously, as if there were no intervening atmosphere. Image restorations with accompanying atmospheric MTF curves are presented. However, restoration results using a simple inverse atmospheric MTF filter were quite similar. This indicates the satellite images were characterized by very low noise and that turbulence jitter was very limited which, in turn, indicates that the turbulence MTFs integrated upwards over the path length wee not significant when compared to aerosol MTFs. Restorations are shown for various wavelength bands and are quite apparent even under clear weather conditions.
The recently developed atmospheric Wiener filter, which corrects for turbulence and aerosol blur and path radiance simultaneously, is implemented in digital restoration of AVHRR imagery over the five wavelength bands of the satellite instrumentation. Restoration is most impressive for higher optical depth situations, with improvement with regard to both smallness of size of resolvable detail and contrast Turbulence modulation transfer function (MTF) is calculated from meteorological data. Aerosol MTF is calculated from optical depth, measured with a sun-photometer. The product of the two yields atmospheric MiT which is implemented in the atmospheric Wiener filter. Image restorations with accompanying atmospheric MTF curves are presented. However, restoration results using a simple inverse MTF filter were quite similar. This indicates the satellite images were characterized by very low noise and that turbulence jitter was very limited which, in turn, indicates that the twbulencc MTFs integrated upwards over the path length were small compared to aerosol MTFs.
Imaging quality of optical systems in a turbid environment is influenced not only by the contents of the turbid layer between the object and the optical receiver but also by the inhomogeneity of that medium. This is important particularly when imaging is performed through clouds, non homogeneous layers of dust, or over vertical or slant paths through the atmosphere. Forward small angle scattering influences more severely image quality and blur when the scattering layer is closer to the receiver. In this study the influence of the position of the scattering layer along the optical axis on the image quality and modulation transfer function (MTF) is investigated. The scattering layer was in controlled laboratory experiments consisted of calibrated polystyrene particles of known size and quantity. A point source was imaged by a computerized imaging system through a layer containing polystyrene particles and the point spread function (PSF) was recorded. The scattering MTF was calculated using the measured PSF. The MTF was measured as a function of the relative distance of the layer from the receiver. The experimental results were compared to theoretical models based on the solution of the radiative transfer theory under the small angle approximation.
Restoration for actual atmospherically blurred images is performed using a Wiener filter which corrects simultaneously for both turbulence and aerosol blur by enhancing the image spectrum primarily at those high frequencies least affected by the jitter or randomness in turbulence MTF. Correction is based upon predicted rather than measured atmospheric MTF. Both turbulence and aerosol MTFs are predicted using meteorological parameters measured with standard weather stations at the time and location where the image was recorded. A variety of weather conditions are considered. Past results have shown good correlation between measured and predicted atmospheric MTFs. Corrections are shown here for turbulence blur alone, for aerosol blur alone, and for both together. Since recorded images suffer frequently from poor contrast because of atmospheric path radiance, a simple image contrast improvement is also considered for the clarity of the atmospheric deblurring effect.
The overall atmospheric modulation transfer function (MTF) is essentially the product of the turbulence and aerosol MTFs. Models describing meteorological dependences of both C2n (where Cn is the refractive index structure coefficient) and the coarse aerosol size distribution have been developed previously. Here, they are used to predict the turbulence MTF, aerosol MTF, and overall atmospheric MTF according to the weather. Comparison of predictions with measurements yields very low mean squared normalized error and suggests that such prediction models can also be very useful in image restoration based on weather data at the time and general location in which the image was recorded. An interesting aspect of this work is that measurements of the aerosol MTF with different imaging instrumentation are very different, as expected from theory developed previously concerning the practical aerosol MTF actually recorded in the image. This is dependent on instrumentation parameters. This experimental verification with two different imaging systems supports the model that the "practical" aerosol MTF is very dependent on instrumentation.
Overall atmospheric modulation transfer function (MTF) is essentially the product of turbulence and aerosol MTF. Models describing meteorological dependences of both Cn2 and coarse aerosol size distribution have been developed previously. Here, they are used to predict turbulence MTF, aerosol MTF, and overall atmospheric MTF according to weather. Comparison of predictions to measurements yields very high correlations and suggests that such prediction models can also be very useful in image restoration based on weather data at the time and general location in which the image was recorded. An interesting aspect of this work is that measurements of aerosol MTF with different imaging instrumentation are very different, as expected from theory developed previously concerning the practical aerosol MTF actually recorded in the image. This is dependent upon instrumentation parameters. This experimental verification supports the model that the `practical' aerosol MTF is very dependent upon instrumentation.
Predictions of atmospheric transmittance in desert aerosol environments using MODTRAN code diverge significantly from measured data. Good prediction of the desert particulate size distribution is required in order to predict atmospheric scattering and absorption parameters. It is also essential to the prediction of the aerosol atmospheric modulation transfer function which is often the dominant component of the overall atmospheric MTF. Recently an effort to predict statistics but not size distribution according to simple weather parameters has been made for coarse desert aerosols. A quantitative analysis of the desert particulate size distribution models was also performed. In this research the size distribution parameters measured by optical counters are related to weather parameters. Known statistical and analytical models such as MODTRAN relate the size distribution parameters only to relative humidity for continental atmospheres. Although humidity has a significant role in the prediction of aerosol size statistics, other weather parameters are seen here to strongly influence also the size distribution parameters. Comparisons such as the above can be used to predict under which conditions the MODTRAN aerosol models have good or poor accuracy. It is also hoped that they will lead to improvements in MODTRAN, improving the accuracy of the humidity dependence as well as by incorporating other meteorological parameters into the MODTRAN prediction models.
In this paper, the incorporation of atmospheric aerosol and turbulence blur and motion blur into visible, near infrared, and thermal infrared target acquisition modeling is considered. Here, we show how the target acquisition probabilities and, conversely, the ranges at which objects can be detected are changed by the inclusion of these real-life environmental effects whose blur is often significantly greater than that of imaging system hardware. It is assumed that images are contrast-limited rather than noise-limited, as is indeed the case with most visible, near infrared (IR), and thermal IR sensors. For short focal lengths with low angular magnification, such environmental blur effects on target acquisition are negligible. However, for longer focal lengths with large angular magnification, resolution is limited by them and this has a strong adverse effect on target acquisition probabilities, times, and ranges. The considerable improvement possible with image correction for such environmental blur automatically in a fraction of a second is significant for contrast-limited imaging, and is discussed here too. Knowledge of such environmental MTF is essential to good system design and is also very useful in image restoration for any type of target or object.
Predictions of atmospheric transmittance in desert aerosol environments using MODTRAN code diverge significantly from measured data. Good prediction of the desert particulate size distribution is required in order to predict atmospheric scattering and absorption parameters. It is also essential to the prediction of the aerosol atmospheric modulation transfer function which is often the dominant component of the overall atmospheric MTF. Recently, an effort to predict statistics but not size distribution according to simple weather parameters has been made for coarse desert aerosols. A quantitative analysis of the desert particulate size distribution models was also performed. In this research the size distribution parameters measured by optical counters are related to weather parameters. Known statistical and analytical models such as MODTRAN relate the size distribution parameters only to relative humidity for continental atmospheres. Although humidity has a significant role in the prediction of aerosol size statistics, other weather parameters are seen here to strongly influence also the size distribution parameters. Comparisons such as the above can be used to predict under which conditions the MODTRAN aerosol models have good or poor accuracy. It is also hoped that they will lead to improvements in MODTRAN, improving the accuracy of the humidity dependence as well as by incorporating other meteorological parameters into the MODTRAN prediction models.
An effort to quantify effects of weather in the northern Negev desert on airborne particle concentration and on cross-sectional area per unit volume, so as to permit prediction according to weather, has begun. Correlations of prediction with measurement are on the order of 94% and 91%, respectively. Humidity is the dominant weather parameter, as expected, but it is not the only parameter. There is statistical significance too to solar flux, air temperature, and wind speed. The empirical model here provides a means of comparing theoretical models of effects of weather on airborne particle statistics with real-world phenomena. It is suggested that these models may well be applicable elsewhere, since the newly airborne particles are only a small fraction of the total airborne amount.
An effort to predict desert coarse aerosol statistics but not size distribution according to simple weather parameters has been made. A quantitative analysis of the desert particulate size distribution models was also performed. In this research the size distribution parameters measured by optical counters are related to weather parameters. Known statistical and analytical models such as MODTRAN relate the size distribution parameters only to relative humidity for continental atmospheres. Although humidity has a significant role in the prediction of aerosol size statistics, other weather parameters can also strongly influence the size distribution parameters. Comparisons such as the above can be used to predict under which conditions the MODTRAN aerosol models have good or poor accuracy. It is also hoped that they will lead to improvements in MODTRAN, improving the accuracy of the humidity dependence as well as by incorporating other meteorological parameters into the MODTRAN prediction models.
A method of calculating numerically the optical transfer function appropriate to any type of image motion and vibration, including random ones, has been developed. We compare the numerical calculation method to the experimental measurement; the close agreement justifies implementation in image restoration for blurring from any type of image motion. In addition, statistics regarding the limitation of resolution as a function of relative exposure time for low-frequency vibrations involving random blur are described. An analytical approximation to the probability density function for random blur has been obtained. This can be used for the determination of target acquisition probability. A comparison of image quality is presented for three different types of motion: linear, acceleration, and high-frequency vibration for the same blur radius. The parameter considered is the power spectrum of the picture.
Path integrated atmospheric transmittance over a 5.5 km horizontal path is measured using black target contrast. Measurements of on-line particulate distributions by Particulate Measuring System instrumentation (PMS) and of meteorological parameters are also made. The extinction coefficients, primarily scattering, of aerosols are calculated using the PMS data, and those arising from molecular absorption are calculated by LOWTRAN7. Both extinction coefficients, the directly measured path integrated and those calculated from particulate distribution and meteorological parameters near the receiver, are compared. Good agreement exists especially when relative humidity is low, despite the fact that the second method involves aerosol size distribution by data collected from a single point along the atmospheric path. Disagreement between both methods under high values of relative humidity can be explained by classification errors of the PMS instrumentation because of changes in the index of refraction of particles in a humid environment.
A method of calculating numerically the optical transfer function appropriate to any type of image motion and vibration, including random ones, has been developed. Here, the numerical calculation method is compared to experimental measurement, and the close agreement justifies implementation in image restoration for blurring deriving from any type of image motion. In addition, statistics regarding limiting resolution as a function of relative exposure time for low frequency vibrations involving random blur are described. An analytical approximation to the probability function has been obtained. This can be implemented in target acquisition probability. Comparison of image quality is presented for three different kinds of motion: linear, acceleration, and high frequency vibration for the same blur radius. The parameter considered is the power spectrum of the picture.
A method of calculating numerically the optical transfer function appropriate to any type of image motion and vibration, including random ones, has been developed. Here the numerical calculation method is compared to experimental measurement, and the close agreement justifies implementation in image restoration for blurring deriving from any type of image motion. In addition, statistics regarding limiting resolution as a function of relative exposure time for low frequency vibrations involving random blur are described. This can be implemented in target acquisition probabilities.
Prediction of wave propagation parameters through an aerosol medium such as scattering and absorption coefficients and scattering phase function is possible if the particulate size distribution is known. In this paper an effort is made to relate the desert particulate size distribution parameters to simple meteorological parameters. All of the size distribution curves showed clear `knees' in their characteristic indicating that the source of the particles is more than one. The particulate size distributions were well fitted to a Trimodal log normal distribution. This good agreement suggests that the particulate size distribution is composed from three separate sources. The first and second sources are due to the local particles which exist in the location of the measurement, contributing to the smaller radii of the size distribution. The third source is large particles which are transferred from large deserts like the Sahara desert by dust carrying winds. These particles contribute to the larger radii end of the distribution. The size distribution parameters were related to meteorological parameters. The most dominant parameter was relative humidity. Using this model it should be possible to predict the particulate size distribution from meteorological data. Prediction of particulate size distribution allows the prediction of aerosol Modulation Transfer Function which is crucial in the prediction of image quality propagating through aerosols.
An imaging system with a narrow field of view was used to measure the overall atmospheric MTF of an horizontal path and the turbulence MTF at the same time and over the same optical path over the visible and near IR. Results suggest that the atmospheric MTF is composed not only of turbulence MTF alone. Image quality is found to be degraded by aerosols, although the spatial frequencies of degradation were up to 5 tp 15 cycles per mrad rather than radian.
A new method of numerical calculation of MTF is presented here for image motion in one- dimension. The method is applicable in principle to any type of motion and can be expanded to two-dimensional motion. It is applied here to uniform velocity motion and to sinusoidal vibrations. Comparison to known analytical methods is made where possible, and agreement is excellent. This supports its implementation to any kind of random motion, particularly where no unique analytical MTF is possible.
The limits of image quality through the atmosphere depend on the overall atmospheric modulation transfer function (MTF) cutoffs. The first spatial frequency cutoff, or 'knee,' of the overall atmospheric MTF curve depends on aerosol MTF. The second spatial frequency cutoff, limited by threshold contrast required in the output image, also depends on turbulence. Measurements of atmospheric MTF over a 5.5 km horizontal path near the ground were made for a large range of spatial frequencies at several wavelengths in the visible and near-IR spectrum along with measurements of turbulence by a passive edge wander technique. On-line measurements of particulate size distribution were made using a Particle Measurement System, Inc. (PMS) probe, and on-line meteorological data (air temperature, relative humidity, wind speed, wind direction, and solar flux) were obtained from a weather station a few meters away from the imager. The experimentally-derived MTF curve is compared to well-known models for turbulence MTF and aerosol MTF. The determination of aerosol MTF from macroscale parameters is crucial to prediction of image quality through the atmosphere and can be implemented in image restoration.
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.