Raindrop Images Research Articles

Frequency-oriented hierarchical fusion network for single image raindrop removal.

Single image raindrop removal aims at recovering high-resolution images from degraded ones. However, existing methods primarily employ pixel-level supervision between image pairs to learn spatial features, thus ignoring the more discriminative frequency information. This drawback results in the loss of high-frequency structures and the generation of diverse artifacts in the restored image. To ameliorate this deficiency, we propose a novel frequency-oriented Hierarchical Fusion Network (HFNet) for raindrop image restoration. Specifically, to compensate for spatial representation deficiencies, we design a dynamic adaptive frequency loss (DAFL), which allows the model to adaptively handle the high-frequency components that are difficult to recover. To handle spatially diverse raindrops, we propose a hierarchical fusion network to efficiently learn both contextual information and spatial features. Meanwhile, a calibrated attention mechanism is proposed to facilitate the transfer of valuable information. Comparative experiments with existing methods indicate the advantages of the proposed algorithm.

Mutual channel prior guided dual-domain interaction network for single image raindrop removal

Removing raindrops adhering to lenses or glass is challenging since raindrops have more complex internal structures and optical effects than rain streaks. Existing raindrop removal methods typically employ single-channel masks for raindrop representation. However, these methods ignore the complementarity of priors between different channels, thereby losing the color and transparency information of raindrops. Moreover, simply applying the same processing across different pixels is insufficient to deal with spatially-varying degradation. To address these limitation, we propose a novel Mutual channel prior guided Dual-domain Interaction Network (MDINet) for single image raindrop removal. Specifically, to effectively guide the restoration of the raindrop region, we propose a mutual channel prior by modeling the multi-channel raindrop soft mask, in which the enriched color and transparency information of raindrops are represented and learned by our model. To tackle the spatially-varying degradation, we design the dual-domain interaction block to capture global low-frequency content and local high-frequency details. Simultaneously, a spatially adaptive modulation block is devised to handle complicated and diverse raindrops. Moreover, we propose a contrastive discrimination strategy to motivate the restored image to approach the clean image and stay away from the raindrop image in the representation space. Extensive experiments on benchmark datasets and real-world images indicate that the proposed method achieves qualitative and quantitative superiority against state-of-the-art methods.

Generative Adversarial Network for Image Raindrop Removal of Transmission Line Based on Unmanned Aerial Vehicle Inspection

In the process of UAV line inspection, there may be raindrops on the camera lens. Raindrops have a serious impact on the details of the image, reducing the identification of the target transmission equipment in the image, reducing the accuracy of the target detection algorithm, and hindering the practicability of UAV line inspection technology in cyber‐physical energy systems. In this paper, the principle of raindrop image formation is studied, and a method of raindrop removal based on generation countermeasure network is proposed. In this method, the attention recurrent network is used to generate the raindrop attention map, and the context code decoder is used to generate the raindrop image. The experimental results show that the proposed method can remove the raindrops in the image and repair the background image of raindrop coverage area and can generate a higher quality raindrop removal image than the traditional method.

A Survey on Video Dehazing Using Deep Learning

With the fast development of object recognition and detection in autonomous driving and video monitoring, image with haze or raindrop can affect the result a lot. As deep learning develops, the hazed and raindrop image can lower the accuracy of object recognition and detection significantly. While hazed images cannot be managed using other image refining process. Since the noise in hazed image is signal-dependent. The object degradation in hazed image is related to object depth. So, the dehazing process depends on the input image. This paper provides a survey on single image and video dehazing methods, from end-to-end system to distributed system. General methods based on deep learning of state-of-art papers from 2010 to 2018 are summarized and compared, accompanied with their datasets of the current progress in this field. The application of these methods and relationship between these methos are also discussed in this paper.

Single Image Rain Removal Based on Deep Learning and Symmetry Transform

Rainy, as an inevitable weather condition, will affect the acquired image. To solve this problem, a single image rain removal algorithm based on deep learning and symmetric transformation is proposed. Because of the important characteristics of wavelet transform, such as symmetry, orthogonality, flexibility and limited support, wavelet transform is used to remove rain from a single image. The image is denoised by using wavelet decomposition, threshold value and wavelet reconstruction in wavelet transform, and the rain drop image is transformed from RGB space to YUV (luma chroma) space by using deep learning to obtain the brightness component and color component of the image. the brightness component and residual component of the raindrop source image and the ideal recovered image without raindrop are extracted. The residual image and brightness component are overlapped again, the reconstructed image is restored to RGB space by YUV inverse transformation, and the final color raindrop free image is obtained. After training the network, the optimal parameters of the network are obtained, and finally the convolution neural network which can effectively remove the rain line is obtained. Experimental results show that compared with other algorithms, the proposed algorithm achieves the highest value in both peak signal-to-noise ratio (PSNR) and structural similarity, which shows that the image effect of the algorithm is better after rain removal.

Automatic Precipitation Measurement Based on Raindrop Imaging and Artificial Intelligence

Rainfall measurement is subjected to various uncertainties due to the complexity of measurement techniques and atmosphere characteristics associated with weather type. Thus, this article presents a video-based disdrometer to analyze raindrop images by introducing artificial intelligence technology for the rainfall rate. First, a high-speed CMOS camera is integrated into a planar LED as a backlight source for appropriately acquiring falling raindrops in different positions. The falling raindrops can be illuminated and used for further image analysis. Algorithms developed for raindrop detection and trajectory identification are employed. In a field test, a rainfall event of 42 continuous hours has been measured by the proposed disdrometer that is validated against a commercial PARSIVEL2 disdrometer and a tipping bucket rain gauge at the same area. In the evaluation for 5-min rainfall images, the results of the trajectory identification are within the precision of 87.8%, recall of 98.4%, and F1 score of 92.8%, respectively. Furthermore, the performance exhibits that the rainfall rate and raindrop size distribution (RSD) obtained by the proposed disdrometer are remarkably consistent with those of PARSIVEL2 disdrometer. The results suggest that the proposed disdrometer based on the continuous movements of the falling raindrops can achieve accurate measurements and eliminate the potential errors effectively in the real-time monitoring of rainfall.

Application research on improved CGAN in image raindrop removal

Rainy weather can greatly reduce the image quality and hinder the subsequent processing of the image. In order to achieve raindrop removal on rainy images, the single image raindrop removal method based on conditional generative adversarial networks (CGAN) is proposed. In this method, CGAN is used as the basic framework. The network receives the raindrop image as an additional condition information and adds Lipschitz constraint on the network. The network model is trained by the combination of condition adversarial loss, content loss, and perception loss to repair the raindrop area and reconstruct the image. The experimental results show that the proposed method has better raindrop removal effect than the existing algorithm and can avoid image blurring on the basis of ensuring the raindrop removal effect.

An efficient raindrop parameter estimation using image processing

Nowadays, image processing algorithms play a key role in the rain drop distribution estimation. This paper deals with number of drops and drop size distribution and its volume in particular time. We are using the raindrop image to calculate the amount of rainfall in a particular time. The proposed image processing based rain drop parameter estimation (IPRDPE) by using double-density dual-tree DWT (DDDT DWT) and thresholding based segmentation. By using effective image fusion technique, rain drop images from different angles are fused and using segmentation and morphological operations raindrop parameters estimated. To get better fused output max-based effective image fusion rules are used. The system using advanced image fusion technique and estimation for rain drop parameter, produce more accuracy and error free system compared to the existing techniques and also achieved better accuracy with respect to the real-time measurement.

High-Speed Optical Disdrometer for Rainfall Microphysical Observations

AbstractA high-speed optical disdrometer (HOD) that was developed for measuring rainfall microphysical quantities, including raindrop shape, size distribution, and fall velocity/acceleration, is the subject of this paper. The main components of the HOD are a high-speed video camera, a light-emitting diode (LED) light, and a sensing unit to detect raindrops passing through the camera view frame. The high-speed video camera is directed at the LED light to capture the silhouettes of the backlit drops when triggered by a raindrop that is detected within a specified focal depth by the sensing unit. The use of a sensing unit enables the confinement of a measurement volume around the camera focal plane. This innovative operation principle ensures the capture of sharp images of raindrop silhouettes, which are then digitally processed to provide accurate information on various raindrop characteristics. The measurement capabilities of the HOD were evaluated through both laboratory and field tests. In the laboratory tests, high-precision spherical lenses with known diameters and water drops of different sizes generated for a known volume of water were used. In the field tests, the HOD was evaluated against a reference rain gauge in a number of rain events. These tests demonstrated the precise accuracy of HOD measurements and the HOD’s technological readiness for field deployment for various applications. It is expected that the HOD will play an important role in generating new insights on raindrop dynamics and related research through its unique measurement capability of providing sequential high-speed images of raindrops.

A Holographic In-Line Imaging System for Meteorological Applications

A digital holographic in-line imaging instrument aimed for meteorological measurements is presented. The system is designed to capture detailed information of macroscopic hydrometeors in a large imaging volume. We suggest the use of holographic imaging to overcome the depth of field limitations of conventional macroimaging systems in meteorological imaging applications. The first low magnification holographic imaging prototype built for measuring free falling hydrometeors in the size range of a few tens of micrometers to several millimeters is described. The design of the imaging setup is discussed in detail, calibration and characterization measurements of magnification, position accuracy, and resolving power are presented. Field measurements during snow falls and rain were made and reconstructed images of snowflakes and raindrops are shown and analyzed. The results of laboratory tests and the field tests show high quality and resolution of reconstructed holograms to enable detailed image analyses on snowflakes, raindrops, and other types of hydrometeors.

Low‐cost laser disdrometer with the capability of hydrometeor imaging

AbstractPrecision of microwave precipitation remote sensing is ensured by studying different hydrometeor classes (raindrops, snowflakes, graupel, and so on), shapes, and drop size distribution (DSD). Recently, DSD estimates and hydrometeor classification algorithms using polarimetric radars were proposed. These algorithm developments require a validation tool providing in situ hydrometeor measurement. We are working toward the development of a low‐cost, light‐weight, and disposable ballooning hydrometeor imaging probe with DSD measurement capabilities. This paper describes the performance of a new line‐scanning laser hydrometeor imaging probe (called PROBE) and validates the DSD using a commercially available impact‐type disdrometer. PROBE captures raindrop images with 0.125 mm horizontal resolution and more than six slices at the 33‐kHz slice rate, providing a reasonable raindrop diameter. DSD validation shows that PROBE acquires adequate DSD shapes during every 2‐min interval. These results demonstrate PROBE's performance in hydrometeor imaging and DSD measurement capabilities in a rainfall. © 2012 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

Application of a histogram modification algorithm to the processing of raindrop images

Automatic processing of digital images of falling raindrops is complicated by a less than ideal grayscale image histogram. These his- tograms do not display a bimodal shape and lack an easily defined mini- mum, making it difficult to choose a threshold for creating a binary im- age. To help identify peaks in these histograms, and simplify threshold selection, a histogram modification technique originally developed by Peleg is used. This method was modified slightly and then applied to raindrop image processing. Its performance is quantified. © 2008 Society of

Edge detection methods applied to the analysis of spherical raindrop images

Optical imaging of raindrops provides important information on the statistical distribution of raindrop size and raindrop shape. These distributions are critical for extracting rainfall rates from both dual- and single-polarization radar signals. A large number of raindrop images are required to obtain these statistics, necessitating automatic processing of the imagery. The accuracy of the measured drop size depends critically on the characteristics of the digital image processing algorithm used to identify and size the drop. Additionally, the algorithm partially determines the effective depth of field of the camera/image processing system. Because a large number of drop images are required to obtain accurate statistics, a large depth of field is needed, which tends to increase errors in drop size measurement. This trade-off between accuracy and depth of field (dof) is also affected by the algorithm used to identify the drop outline. In this paper, eight edge detection algorithms are investigated and compared to determine which is best suited for accurately extracting the drop outline and measuring the diameter of an imaged raindrop while maintaining a relatively large depth of field. The algorithm which overall gave the largest dof along with the most accurate estimate of the size of the drop was the Hueckel algorithm [J. Assoc. Comput. Mach. 20, 634 (1973)].

The existence of vortices in the wakes of simulated raindrops

Flow visualizations are presented of the wakes behind water droplets traveling at terminal velocity in a vertically directed airflow. These images, obtained by levitating water drops in a vertical wind tunnel, reveal vortices in the wake behind the drop. It has long been postulated that such wakes trigger transverse oscillations in raindrops. Although such transverse oscillations were not observed here, the images do show a canting of the oblate drop which appears to be connected to the vortex shedding process. These are the first images of simulated raindrops that visualize both the shape of the drop and the vortices in the wake immediately behind the drop.

Buoyancy of Convective Clouds in TOGA COARE

Abstract The buoyancy of convective clouds in TOGA COARE was calculated from the NCAR Electra in situ measurements of temperature, humidity, liquid water content, and two-dimensional images of raindrops. Most of the measurements were made at 700 mb (10°C) although some were made at 850 (18°C) and 600 mb (2°C). The temperature was measured with the Ophir radiometer, which does not have the wetting problem that has degraded many previous measurements of in-cloud temperature in warm clouds. On average, the in-cloud virtual temperature excess was found to be less than the adiabatic value by about 2 K, while the negative influence of total water content on buoyancy was less than 0.5 K. Furthermore, the total water content was highly variable and much smaller than the adiabatic value. The authors conclude, therefore, that entrainment and mixing was usually a much larger factor in reducing the buoyancy than water loading. The average buoyancy in downdrafts was positive and similar to the value in updrafts.

Measurements of Raindrop Size Distributions over the Pacific Warm Pool and Implications forZ–RRelations

Raindrop images obtained on research flights of the NCAR Electra aircraft in the Tropical Oceans Global Atmosphere Coupled Ocean–Atmosphere Response Experiment (TOGA COARE) are analyzed. The drop size distributions, based on the images obtained in 6-s samples (about 750 m of flight track), are used to calculate both radar reflectivity Z and rain rate R. Airborne radar data from the NOAA P-3 aircraft flying in coordination with the Electra are used to categorize the particle-image data according to whether the drop images were obtained in regions of convective or stratiform precipitation. Within stratiform precipitation, the same rain rate could be produced by a drop spectrum dominated by numerous small drops (lower reflectivity) or by a few large drops (higher reflectivity). The reflectivity values varied by as much as 9 dB for a given rain rate. Reflectivity data from the airborne radar and flight-level data reveal that the stratiform regions often contain fallstreaks of about 0.1–2 km in horizontal dimension. The fallstreaks are associated with large-drop spectra and local maxima in reflectivity up to approximately 40 dBZ and in rain rates up to 25 mm h−1. The fallstreaks extend downward from the melting band and bend with the low-level wind shear, but do not usually reach the surface. Thus, although relatively more uniform than convective regions, stratiform regions can be variable in reflectivity and rain rate at fine spatial scales in both the horizontal and vertical directions. Stratiform regions are therefore best characterized by their ensemble properties rather than the values of individual high-resolution measurements. The variability of stratiform drop size spectra arises primarily from the occurrence of fallstreaks and the discontinuous nature of regions favoring aggregation of snow crystals, and it implies that Z–R distributions associated with convective and stratiform precipitation are not statistically distinct. Thus, separate Z–R relations for convective and stratiform precipitation are not justified, and techniques to distinguish between convective and stratiform precipitation based solely on the characteristics of drop size distributions are not likely to be accurate. The variability of the drop size spectra in tropical precipitation makes an exponential fit to the Z–R relation sensitive to the spatial scale over which Z and R are determined. This sensitivity can be avoided by using a probability-matched Z–R relation. The probability-matched Z–R relation for all the raindrop image data from the Electra collected between altitudes of 2.7 and 3.3 km in TOGA COARE is similar to the Z–R relation obtained at the sea surface in the Global Atmospheric Research Program Atlantic Tropical Experiment.

Ellipse-Fitting Techniques Applied to 2D-PMS Raindrop Images

An ellipse-fitting technique is introduced in this paper to process the images of raindrops sampled by a 2D-PMS (Particle Measuring Systems, Inc.) probe. Algorithms to estimate size, shape, and orientation parameters of partial raindrop images are presented. This technique is evaluated by fitting full raindrop images and comparing the results with other procedures such as Fourier descriptor and moment descriptor methods. It is also evaluated by applying this pressure to partial raindrop images constructed by truncating the top part or the bottom part of full raindrop images, while at the same time the estimates obtained from the full images are used for comparison against the estimates from partial images. It is shown that the ellipse-fitting technique is a viable tool for processing complete 2D-PMS raindrop images, and more importantly, this technique works successfully for processing partial raindrop images. Results of 2D-PMS data analysis of side-looking raindrop images collected during the CAPE (Convection and Precipitation/Electrification) field program are presented.

Fourier and Moment Methods Applied to Two-Dimensional Raindrop Images

Abstract Two-dimensional PMS precipitation probes mounted with horizontal optical axis have been previously used to study the shapes of hydrometeors such as raindrops and graupel. Fourier and moment descriptors have been applied to such images for the purposes of parameter estimation (axis ratio, canting angle) of raindrops, and for classifying raindrops in a raindrop/graupel mixture. Simulations have been used to evaluate these techniques. Our results show that axis ratios of raindrops can be accurately estimated using both Fourier and moment methods. We also show that the canting angle of the raindrop image can be accurately estimated using the moment method. Two image classification techniques were applied to data from below the melting level in a convective storm for classifying raindrops and graupel. The potential usefulness of such techniques is demonstrated in this paper.

Axis Ratios and Oscillations of Raindrops

Abstract Axis ratios were determined for about 3500 raindrop images recorded in summertime rainshowers by an instrumented aircraft. These ratios were used to determine the mean axis ratios and oscillation amplitudes of raindrops. A filtering algorithm using Fourier descriptors was applied to the raindrop images to reduce the quantization noise and the systematic errors, and simulations were used to estimate the standard errors of the measurement procedure. Drops with diameters <4 mm were observed to be slightly more spherical than would be expected for drops in equilibrium. Oscillation amplitudes were found to be typically ±10% in axis ratio for light to moderate rainfall rates, and such oscillations can account for the departures from equilibrium values. The effects of these axis ratios and oscillations on the differential radar reflectivity of rain are calculated and discussed.

Analysis of aircraft hydrometeor spectra and differential reflectivity (ZDR) radar measurements during the Cooperative Convective Precipitation Experiment

Radar measurements of ZDR are compared with ZDR deduced from aircraft graupel spectra during the Cooperative Convective Precipitation Experiment. On August 4, 1981, the University of Wyoming's Super King Air made two penetrations (at altitudes of 3 km and 1.8 km above mean sea level) through a convective cell located 90 km NE of the CHILL radar. The aircraft was equipped with two two‐dimensional PMS probes located orthogonally so as to view the falling hydrometeors along their symmetry axis (V probe) and along an axis orthogonal to it (H probe). Elliptical raindrop images were observed below cloud base (1.8 km altitude), and conical graupel images were observed at 3 km altitude. Because of beam blockage at the lowest elevation scans, radar data were not available in the rain region. Aircraft raindrop spectra were analyzed in detail, and it is shown that ZDR deduced from such spectra should be an excellent estimator of D0 provided the correct form for the drop size distribution is assumed. ZDR and ZH deduced from the graupel spectra are shown to be in the range measured by the radar.