Search

Underwater images have been highly required in many application scenarios, such as marine resources exploration, reservoir safety evaluation, and underwater aquaculture monitoring. But the captured images under water are usually degraded, due to the refraction, absorption, and scattering of light by the suspended particles in water. Some limitations of images can be low contrast, blurred details, and color distortion. Therefore, it is urgent for underwater image restoration. Much effort has been made on a variety of underwater image restoration using processing technology so far. However, land image restoration cannot be simply transformed the underwater images using background light estimation. The best restoration cannot be achieved in the current case, such as unnatural restoration, uneven brightness, and reddening color tone of the main body. In this study, an underwater image restoration was proposed using the adaptive fusion of double background lights and accurate estimation of the transmission map. First, two kinds of background lights of the underwater images were obtained using the maximum intensity prior background light estimation method based on underwater light attenuation and Quadtree background light estimation method based on background light flatness. Then, an adaptive fusion algorithm of double background lights was created to realize the accurate estimation of background lights in different scenes. Second, an accurate estimation of the transmission map was established to combine the new underwater dark channel prior, Reversed Saturation Map, and three-channel spectral attenuation coefficient. Furthermore, the restored underwater image was constructed using the obtained fused background light and accurate transmission map. The brightness of the image was adjusted to obtain the final underwater restored image. Finally, the experiment was carried out to verify the model using the aquaculture dataset of Guangdong Tilapia Breeding Farm and Underwater Image Enhancement Benchmark Dataset. The result showed that the improved estimation was better adapted to different underwater environments, thereby improving the contrast and color of the restored underwater image. A better restoration was achieved than the rest, such as dark channel prior, the maximum intensity prior, restoration using blur and light absorption, restoration using blue-green channel defogging, restoration using background light statistics model, and optimization of transmission map. Furthermore, the restoration was evaluated to deal with the current underwater image distortion, dim brightness, and reddish tone. The best value was also achieved among the seven objective evaluation indexes. The Peak Signal to Noise Ratio, Structural Similarity, Mean-Square Error, and Visual Information Fidelity increased by 0.52%, 2.1%, 3.4%, and 0.86%, respectively, in the evaluation indexes of reference image quality using the best underwater image restoration, compared with the second best. Correspondingly, the Natural Image Quality Evaluator, and Natural Image Quality Measure index were improved by 2.4% and 7.4%, respectively, in the evaluation index without reference image quality. Two datasets demonstrate that the proposed restoration can be expected to treat the unnatural restoration, uneven brightness, and reddish color in the traditional underwater image restoration. The finding can also provide a technical reference for underwater image restoration. [ABSTRACT FROM AUTHOR]

Pore boundary is generally blur resulted from the partial volume in the soil CT image. This phenomenon has inevitably posed a great influence on the accuracy of soil pore topology. This study aims to propose a novel Sequence information Generative Adversarial Network (SeqGAN) to realize the Super-Resolution reconstruction of soil CT images. Therefore, the SeqGAN was selected to improve the clarity and accuracy of soil CT images, particularly for the high resolution and feature boundaries. Two improvements of SeqGAN were utilized, including the Sequential Convolution block (SeqConv) structure, and Beginning-to-End Residuals Connection block (BE-Resblock). SeqConv structure was involved two convolution block structure. The first convolution block was used to extract the feature of target image, while the second was used to extract the sequence information of the next and previous image in the sequence, thereby to realize the extraction of sequence information. In the BE-Resblock, more than 8 residual blocks were connected in series to extract the image information. At the same time, the residual blocks of the beginning and end were also connected, where the input information was introduced to reduce the probability of overfitting. Furthermore, twice up-sampling blocks were used to improve the resolution of images, where the final output was 4x high-definition Super-Resolution image. The experimental samples soil was taken from Keshan Farm in the northwest of Keshan County, Qiqihar City, Heilongjiang Province (125°23′57.4224″E, 48°18′37.8828″N). Soil samples were collected with a cutting ring and stored in a plexiglass tube. A submerging test was also conducted to obtain the soil samples. A spiral CT scanner was then used to capture soil CT images. The test datasets were finally taken as the 440 soil CT sequence images with high sequence Structural Similarity (SSIM). Two datasets were obtained after preprocessed, including the high- and low-resolution images with twice difference in resolution. Specifically, the low-resolution image dataset contained 220 soil CT images, where each image presented a resolution of 128×128 pixels and a size of 62.17 mm×62.17 mm. At the same time, the high-resolution image dataset (original image datasets) contained 220 soil CT images, where each image presented a resolution of 256×256 pixels and a size of 62.17 mm×62.17 mm. Three common models were selected to compare with the improved model. Qualitative experiments showed that the improved model well performed a higher resolution, and lower gray difference, thereby to reconstruct most soil pores in detail. Quantitative experiments showed that the Mean Square Error (MSE) of improved model was 77%, lower than that of Bicubic linear interpolation, 85% lower than that of Convolution Neural Network (CNN), and 25% lower than that of Generative Adversarial Network. In addition, the Peak Signal to Noise Ratio (PSNR) of improved model was 8.5, higher than that of the Bicubic linear interpolation, 10.2 higher than that of Convolution Neural Network, and 1.4 higher than that of Generative Adversarial Network. SSIM of superand high-resolution image was 3.9%, higher than that of Bicubic linear interpolation, 2% higher than that of Convolution Neural Network, and 0.2% higher than that of Generative Adversarial Network. Consequently, the SeqGAN can be expected to realize the super-resolution reconstruction of soil CT images with high accuracy and high definition. The finding can also provide a potential data reliability and benefit to follow-up research on soil pore segmentation and soil skeletonization. [ABSTRACT FROM AUTHOR]

Hyperspectral remote sensing image includes hundreds of narrow contiguous spectral bands with high spectral resolution, and can provides a contiguous spectral curve for each pixel, which is an important tool for the cropland monitoring in rapid and large-scale way. However, due to the contradiction between spectral resolution and spatial resolution, hyperspectral remote sensing image usually possesses relative low spatial resolution. Therefore, it is very important to mix various vegetation and soil at one pixel point for spectral decomposition in hyperspectral image and spectral unmixing of farmland. The first step in spectral unmixing is usually endmember extraction. Endmember extraction from a hyperspectral remote sensing image is to find some pixels (endmembers) and regard them as pure spectral reflectance of the vegetation and soil in the image to get the best accuracy of spectral unmixing results. So endmember extraction from hyperspectral remote sensing image can be regarded as a typical discrete optimization problem, which can be solved by swarm intelligence optimization algorithm. Before optimization, the spectral dimension of the image should be reduced by deep learning. In order to solve the problem of spectral unmixing of hyperspectral images, a method of extracting farmland endmember based on deep learning and shuffled frog leaping algorithm (SFLA) is proposed in this paper. Firstly, deep learning model named stacked auto encoders (SAE) was used to extract spectral features . SAE performs a non-linear transfer from the original spectral signals to a form with significant features and less dimensions. In the low-dimension space, the candidate endmembers were selected as the input of the SFLA. The purpose of extracting the candidate endmembers is to simplified the computational complexity in the next step. Secondly, the endmember extraction of hyperspectral image was transferred into the combinatorial optimization and the objective function was constructed, then the SFLA was used to optimize the objective functionto get the best combination of endmembers. The objective function in study was designed as the RMSE (root mean square error) between the real hyperspectral remote sensing image and the simulated hyperspectral remote sensing image using the endmembers and abundance after endmember extraction and spectral unmixing. Thirdly, 2 groups experiments were carried out on the synthetic hyperspectral datasets with 3 different SNR (signal to noise ratio, 20 , 30 and 40 dB) and the real AVIRIS hyperspectral remote sensing dataset in Salinas region, respectively. In experiments, the experimental results of the proposed method was compared with that of 3 traditional methods for endmember extraction including the sequential maximum angle convex cone (SMACC), N-FINDR, Vertex Component Analysis (VCA). The results were evaluated by RMSE and spectral angle. The results showed that the RMSE was 0.050 8, 0.015 9, 0.005 1, 0.006 7 for 20, 30 and 40 dB dataset, and the real dataset, respectively. The average spectral angle was 0.106 88, 0.030 32, 0.009 94 for 20, 30 and 40 dB dataset respectively. The proposed method was better than traditional methods in terms of extraction accuracy, which had wide potential applications on cropland monitoring using hyperspectral remote sensing. The method proposed in this paper reduced the influence of the non-linear factors and the noise, better endmember extraction and spectral unmixing results (both less spectral angle and less RMSE) could be obtained, and the proposed method was robust when the noise of the image increased sharply. In conclusion, the endmember extraction method proposed by this study is of significant importance for the cropland monitoring using hyperspectral remote sensing and has a prosperous future for the application on the remote sensing of agriculture. [ABSTRACT FROM AUTHOR]

As apple harvesting needs large amount of labor, and the seasonality is strong, the night operation of apple harvesting robot is proposed, in order to improve the efficiency of harvesting. The apple's real-time night vision image contains lots of noise, which is captured by image processing system of apple harvesting robot. The noise will influence the operating efficiency and recognition precision, and then influence the harvesting efficiency. Under different artificial lights, apple night vision images are captured, the noises are analyzed through the difference image method, and the type of noise is determined to be mixed noise. The main part of mixed noise is Gaussian noise, accompanied by some salt-pepper noise. Aiming at the problem of Gaussian noise removal, the theory of independent component analysis (ICA) is introduced into the de-noising method for night vision image. The ICA algorithm mostly uses gradient iterative solver, so it has some defects, such as easily trapped in local minimum, slow convergence speed. All of these defects lead to the following phenomena easily, such as the unthoroughness in the de-noising and the long running time. In order to overcome these defects, particle swarm optimization (PSO) algorithm is used to optimize the ICA, further to establish an optimized ICA de-noising method based on PSO (PSO-ICA), applied in night vision image, hoping to minimize noise pollution and improving the operating efficiency of de-noising method. Using the standard Lenna image and apple image captured under nature light, by the simulation experiments, these 2 pictures are added with the Gaussian noise with the variance of 0.05 and the salt-pepper noise with the P value of 0.05, respectively. Compared with the average filtering method and ICA de-noising method, the results show that the de-noising effect of PSO-ICA algorithm is the most ideal. Using peak signal-to-noise ratio (PSNR) to do difference test, the result shows that, under 0.05 significant level, 3 de-noising methods show significant difference. Using different apple night vision images captured to do experiments, the results show that, from the visual evaluation, the low noise image is obtained by PSO-ICA de-noising method, and its noise decreased significantly. In order to evaluate the de-noising effect of night vision image more objectively, taking the natural light image as reference, the concept of relative peak signal-to-noise ratio (RPSNR) is proposed. From the RPSNR evaluation, compared with the original image, the image after average filtering de-noising and that after ICA de-noising, the image based on the method of PSO-ICA de-noising increased on average by 21.28%, 12.41% and 5.53%, respectively. From the run time evaluation, PSO algorithm has greatly improved the efficiency of ICA algorithm. Under incandescent lamp, the night vision image and its de-noised images have the highest RPSNR, so this type of light is suitable for artificial light source. Finally, under the natural light and 3 different artificial lights, 10 images of natural light and 30 night images are captured from 10 sample points. Using all of these images to do the repeated experiments, the trends of experimental results are consistent. In conclusion, PSO-ICA algorithm has unique advantage for night vision image de-noising, which provides a solid foundation for the night operation of apple picking robot. [ABSTRACT FROM AUTHOR]

In order to eliminate the noise in the image effectively and to protect the image detail better, this paper proposed a new method for image denosing based on Contourlet transform and improved NeighShink. It used the stein unbiased risk estimating in the directional subband of each scale for heuristic estimating after Contourlet transform of image, then according to neighboring window energy, set the low energy coefficient to 0. It used approximate maximum likelihood estimation method to estimate the variance of the high energy coefficient, after then used a minimum mean square error criterion to get real coefficient estimates. Finally it carried on the neighboring coefficient shrinkage. Experiment on image denoising shows that the method can eliminate the noise in image effectively and get a higher peak signal to noise ratio, and the image detail can be well protected. [ABSTRACT FROM AUTHOR]