An object detection method for heavy fog scenes based on image defogging and sample enhancement
1
Citation
13
Reference
10
Related Paper
Citation Trend
Abstract:
Deep convolutional neural network has achieved superior recognition performance on many public object detection datasets. However, under the weather conditions of rain or fog, the scarcity of samples has always been the problems restricting the accuracy of detection and identification. To solve this problem, this paper proposed an object detection method for heavy fog scenes based on image defogging and sample enhancement. Firstly, generative adversarial network (GAN) is adopted to remove the fog from images, and then achieve sample enhancement by a style transfer network, which keeps the image content basically unchanged and transform the style of image texture. Fog-free dataset after sample enhancement can reduce the influence of the texture information on the network model and make it pay more attention to the contour information of the object shape. The experimental results on I-HAZE and REISDE dataset show that our proposed method can effectively improve the object detection precision and the mAP (mean average precision) can be improved by up to 15%.Keywords:
Sample (material)
Generative adversarial network
Texture (cosmology)
Haze
Image restoration,based on TV model,performs well for non-texture image,but small texture can be ground easily for texture image.In this paper,texture is extracted from original image and inpainted,then the inpainted texture is compensated to the denoising image outputed by TV model.The experimental results show that the method proposed by this paper can improve the image visual effect well.
Texture (cosmology)
Texture filtering
Cite
Citations (0)
Texture is one of the most important characteristics used in computer vision and image processing applications. A new texture classification and retrieval method is proposed for texture analysis applications. The technique makes use of the random neural network model. The main aim is to represent textures with parameters which are the random neural network weights and classify and retrieve textures using this texture definition. The network has neurons that correspond to each image pixel, and the neurons are connected according to neighboring relationship between pixels. The method is tested on images produced using the Brodatz album and texture blocks cut from remotely sensed images.
Texture (cosmology)
Texture compression
Texture filtering
Contextual image classification
Cite
Citations (5)
Haze or dust haze,mainly consisting of PM10 or PM2.5,can cause a decrease of visibility in the air.In recent years,with the rapid development of urbanization,the urban air pollution becomes more and more serious,and the number of days that dust haze occurs has increased.The formation of dust haze,the chemical component and the damage of dust haze weather are introduced.The study of the effects of dust haze weather on human body from the epidemiology is also reviewed.
Haze
Visibility
Atmospheric Dust
Cite
Citations (0)
A new method of colour texture modelling based on Gaussian distribution mixtures is discussed. We estimate the local statistical properties of the monospectral version of the target texture in the form of a Gaussian mixture of product components. The synthesized texture is obtained by means of a step-wise prediction of the texture image. In order to achieve a realistic colour texture image and to avoid possible loss of high-frequency details we use optimally chosen pieces of the original colour source texture in the synthesis phase. In this sense the proposed texture modelling method can be viewed as a statistically controlled sampling. By using multispectral or mutually registered BTF texture pieces the method can be easily extended also for these textures.
Texture (cosmology)
Texture compression
Texture filtering
Texture Synthesis
Bidirectional texture function
Texture atlas
Projective texture mapping
Texel
Cite
Citations (9)
Texture aspects are encountered in most domains that use image processing. Despite the decades of research on texture image information, texture characterization remains a competition for the scientific community because a rigorous mathematical definition of texture was not yet decided. Therefore, texture analysis and classification needs depth research and continuous improvement. Beside common fractal properties, this paper uses a modern fractal property which characterizes different texture descriptors for image analysis in order to provide additional information and to extract new texture features.
Texture (cosmology)
Texture filtering
Texture compression
Contextual image classification
Cite
Citations (14)
According to the haze data at 7 weather stations from 1971 to 2010 in Shiyan of Hubei province,the characteristics of haze change were analyzed using the methods of a linear trend analysis,a EOF and a M-K and a wavelet analysis,The results showed that in Shiyan haze days was maximum in December and minimum in July to September,the haze days in autumn and winter accounted for more than 70 percent of annual total haze days.In general,the number of annual haze days represent the obvious increasing trend,especially increasing rapidly after the 1990s,the haze days is Increased with a tendency of 5.9 days per 10 years.The change of annual haze days could be divided into five stages,among which haze days from middle to late 1990's and beginning of the 21 century were increasing rapidly.However the number of the haze days in Shiyan is still more and the annual average is 23.3d.From regional distribution,haze days were most in Danjiangkou,and next calne Yunxian,while Fangxian had fewest haze days.In the recent 40 years,the haze show remarkable interannual and interdecadal variability.There is remarkable haze period of 8 years.The most concentrated period of haze mutations was in 2002.
Haze
Cite
Citations (0)
A heavy haze episode that occurred in Beijing from 20 September to 27 September, 2011 was observed to explore the secondary processes of the haze episode. During the haze episode, the relatively stable synoptic conditions and regional transport from polluted areas in the south and southwest of Beijing favored the formation of haze. Significant increases of PM2.5/PM10 ratio was observed during haze period, which implied that the haze was caused by fine particles. Additionally, the presence of secondary inorganic pollutants (SO42–, NO3– and NH4+) sharply increased during the haze episode, which indicated that secondary processes significantly strengthened the haze episode. The sulfur oxidation ratio (SOR) sharply increased from a non-haze episode with a highest value of 0.11 to a haze episode with a highest value of 0.62. Low correlations between SOR and O3 and the temperature were found, whereas a high correlation between SOR and RH was found during the haze episode, which implied that sulfate was mainly produced by the aqueous-phase oxidation of SO2 rather than the gas-phase conversion of SO2 to sulfate in haze episode in Beijing. Furthermore, a fine linear relationship between SOR and the surface area (dS) of particles smaller than 1 µm confirmed the heterogeneous processes of sulfate formation in haze episode. The nitrogen oxidation ratio (NOR) also sharply increased from a non-haze episode with a highest value of 0.03 to a haze episode with a highest value of 0.26, which indicated more intense secondary formation of nitrate in haze episode. Nitrate was found to be mainly produced by a homogenous reaction under ammonium-rich conditions. Higher RH in haze episode reduced the thermodynamic equilibrium constant Ke', and favored the thermodynamic equilibrium reaction of HNO3(g) + NH3(g) ↔ NH4NO3(s, aq) to formed nitrate, which might help explain the enhanced homogenous production of nitrate in haze episode. In addition, a good empirical fit (R2 = 0.70) between NOR and dS was found, which indicated that the particle surface area significantly contributed to the intense homogeneous production of nitrate in haze episode.
Haze
Cite
Citations (82)
Haze
Visibility
Cite
Citations (53)
A new method of colour texture modelling based on Gaussian distribution mixtures is discussed. We estimate the local statistical properties of the monospectral version of the target texture in the form of a Gaussian mixture of product components. The synthesized texture is obtained by means of a step-wise prediction of the texture image. In order to achieve a realistic colour texture image and to avoid possible loss of high-frequency details we use optimally chosen pieces of the original colour source texture in the synthesis phase. In this sense the proposed texture modelling method can be viewed as a statistically controlled sampling. By using multispectral or mutually registered BTF texture pieces the method can be easily extended also for these textures.
Texture (cosmology)
Texture compression
Texture filtering
Texture Synthesis
Projective texture mapping
Bidirectional texture function
Texture atlas
Texel
Cite
Citations (14)
Texture is one of the important characteristics of image.For the further analysis on the characteristic of textures,considering from the direction of texture is needed.The relationship between image edges and texture direction is analyzed and an algorithm for the main direction of texture based on image edge information is proposed.With this method,the main direction can be evaluated properly.Experiments show that the method is effective.
Texture (cosmology)
Texture filtering
Texture compression
Projective texture mapping
Bidirectional texture function
Cite
Citations (2)