Search

Babbitt metal cubes were prepared by means of selected laser melting (SLM) process using Sn—11% Sb—6% Cu alloy powders; their microstructures were studied using optical microscope (OM), SEM, EDS, XRD and DSC; and their mechanical properties were tested using Vickers hardness and tensile methods. The results show that fully dense Babbitt metal components can be prepared by using appropriate SLM process parameters and an interlayer staggered laser scanning strategy. Anisotropy characteristics appear in both the microstructure and mechanical properties of SLM-Babbitt cubic specimens. The average hardness was in the range of 32.5 HV0.05 to 35.3 HV0.05. Both tensile strength and elongation were somewhat higher in a direction parallel to the laser scanning speed (Y axis) than in a direction perpendicular to the laser scanning speed (X axis). The strengthening mechanism is suggested to include solid solution strengthening of oversaturated Sb in the Sn matrix, as well as dispersion strengthening of finely dispersed SnSb and Cu6Sn5 particles. The overgrown acicular Cu6Sn5 phases at low laser scanning speeds and void formation at higher laser scanning speeds seriously deteriorate the mechanical properties of the SLM-Babbitt specimens.

Knowledge Distillation (KD) transfers the knowledge from a cumbersome teacher model to a lightweight student network. Since a single image may reasonably relate to several categories, the one-hot label would inevitably introduce the encoding noise. From this perspective, we systematically analyze the distillation mechanism and demonstrate that the \(L_2\)-norm of the feature in penultimate layer would be too large under the influence of label noise, and the temperature T in KD could be regarded as a correction factor for \(L_2\)-norm to suppress the impact of noise. Noticing different samples suffer from varying intensities of label noise, we further propose a simple yet effective feature normalized knowledge distillation which introduces the sample specific correction factor to replace the unified temperature T for better reducing the impact of noise. Extensive experiments show that the proposed method surpasses standard KD as well as self-distillation significantly on Cifar-100, CUB-200-2011 and Stanford Cars datasets. The codes are in https://github.com/aztc/FNKD

Imaging non-uniformity of infrared focal plane array (IRFPA) behaves as fixed-pattern noise superimposed on the image, which affects the imaging quality of infrared system seriously. In scene-based non-uniformity correction methods, the drawbacks of ghosting artifacts and image blurring affect the sensitivity of the IRFPA imaging system seriously and decrease the image quality visibly. This paper proposes an improved neural network non-uniformity correction method with adaptive learning rate. On the one hand, using guided filter, the proposed algorithm decreases the effect of ghosting artifacts. On the other hand, due to the inappropriate learning rate is the main reason of image blurring, the proposed algorithm utilizes an adaptive learning rate with a temporal domain factor to eliminate the effect of image blurring. In short, the proposed algorithm combines the merits of the guided filter and the adaptive learning rate. Several real and simulated infrared image sequences are utilized to verify the performance of the proposed algorithm. The experiment results indicate that the proposed algorithm can not only reduce the non-uniformity with less ghosting artifacts but also overcome the problems of image blurring in static areas.

Iterative back-projection (IBP) technique is commonly utilized to minimize the error of the super resolution (SR) reconstruction and acquire a high spatial resolution image with appealing visual effect. However, the IBP-based SR method suffers from the chessboard and ringing effect around edges. In this paper, we employ the total variation as a part of the penalty function to minimize the reconstruction error, which helps to avoid across-edge smoothing and suppress the above-mentioned artifacts simultaneously. The experimental results validate that the proposed method leads to significant gain in terms of both the peak signal-to-noise ratio and the structural similarity index, and promotes the visual effect of the image considerably. © 2015 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

A background suppression method based on multi-scale Hidden Markov Mode(HMT) was proposed to remove the complex background clutter in the detection of dim and small targets.According to difference of distributed characteristics between target and background clutter in infrared image,the shearlet transform based multi-scale HMT was estimated by using the different scale and direction relation of inter-scale and cross-subband coefficients of decomposed images.Finally,the expectation-maximization(EM) algorithm was used to calculate the background,separate dim and small targets and background clutter of infrared image and to implement the suppression of background,and the preservation and enhancement of target signals.Compared with Max Median(MMed) and Local Means Remove(LMR) filter from subject inspection and value index,several groups of experimental results demonstrate that the proposed method can suppress the complicated background in dim and small target images effectively(SCR=1.6).

The scene adaptive nonuniformity correction (NUC) technique is commonly used to decrease the fixed pattern noise (FPN) in infrared focal plane arrays (IRFPA). However, the correction precision of existing scene adaptive NUC methods is reduced by the nonlinear response of IRFPA detectors seriously. In this paper, an improved scene adaptive NUC method that employs “S”-curve model to approximate the detector response is presented. The performance of the proposed method is tested with real infrared video sequence, and the experimental results validate that our method can promote the correction precision considerably.

In scene-based nonuniformity correction algorithms, artificial ghosting and image blurring degrade the correction quality severely. In this paper, an improved algorithm based on the diamond search block matching algorithm and the adaptive learning rate is proposed. First, accurate transform pairs between two adjacent frames are estimated by the diamond search block matching algorithm. Then, based on the error between the corresponding transform pairs, the gradient descent algorithm is applied to update correction parameters. During the process of gradient descent, the local standard deviation and a threshold are utilized to control the learning rate to avoid the accumulation of matching error. Finally, the nonuniformity correction would be realized by a linear model with updated correction parameters. The performance of the proposed algorithm is thoroughly studied with four real infrared image sequences. Experimental results indicate that the proposed algorithm can reduce the nonuniformity with less ghosting artifacts in moving areas and can also overcome the problem of image blurring in static areas.

An anomaly detection algorithm for hyperspectral images based on nonsubsampled Pyramid decomposition (NSPD) was proposed. Both spatial and spectral information have been used to locate and detect the anomaly under the condition of no prior knowledge about the anomaly and the background. Firstly, the hyper-spectral images was decomposed into a series of different scale sub-bands using NSPD; and then using the correlation of neighborhood coefficient of different scale space in a wave-band, the background data was optimally predicted by reducing the anomalous data using the improved kernel unsharp masking filter in different scale of each sub-band. Finally the anomaly targets could be detected by using the RX operator in the feature space. Numerical experiments were conducted on real and synthesized hyperspectral data to validate the effectiveness of the proposed algorithm. Compared with the classical RX algorithm, several experimental results show that the proposed algorithm has better detection performance and lower false alarm probability.

To improve the detection performance for weak and small targets signal in complex infrared background, such as the ground and the cloud, the small target background suppression algorithm based on vision perception model (VPM) is presented. Firstly, with simple cell receptive field model, original infrared image is decomposed to two images by different Gabor functions using convolution. And then, the nonlinear convergence function of complex cell response is utilized to fusion two images obtained by separation small target with background clutter in infrared image. Finally, the target image is obtained by using classical adaptive threshold method. Several groups of experimental results demonstrate that the proposed method can suppress the infrared background effectively, compared with several classical infrared dim and small target background suppression methods, such as local means remove and two-dimensional least means square filter methods.

A new image registration method for multimodal images is proposed in this paper. This method is a combination of the modified scale invariant feature transform (SIFT) feature extraction algorithm and the shape-context feature descriptor. Salient points of multimodal images are extracted by using the modified SIFT feature extraction algorithm. Then each salient point is described by using the shape-context descriptor that formed a feature vector from the orientation histograms of the subregion around each salient point. After salient points matching by using Euclidean distance, random sample consensus algorithm is used to eliminate wrong corresponding pairs. At last, multimodal images registration is achieved by affine transformation and bilinear interpolation. Experimental results for registration of IR images and electro-optical images show that this method has a good registration result.

An improved scene-adaptive nonuniformity correction (NUC) algorithm for infrared focal plane arrays (IRFPAs) is proposed. This method simultaneously estimates the infrared detectors' parameters and eliminates the nonuniformity causing fixed pattern noise (FPN) by using a neural network (NN) approach. In the learning process of neuron parameter estimation, the traditional LMS algorithm is substituted with the newly presented variable step size (VSS) normalized least-mean square (NLMS) based adaptive filtering algorithm, which yields faster convergence, smaller misadjustment, and lower computational cost. In addition, a new NN structure is designed to estimate the desired target value, which promotes the calibration precision considerably. The proposed NUC method reaches high correction performance, which is validated by the experimental results quantitatively tested with a simulative testing sequence and a real infrared image sequence.

The noncollinear SHG method has been used to measure the pulse width of passively mode-locked Nd: YAG laser. The nonlinear ADP crystal is oriented in such a way that phase matched SHG is produced only when both incident beams are present as ordinary rays. The angle between the two beams is 25°50′. Phase-matched SHG as an extra-ordinary occurs at an angle bisecting the two beams. The correlation width obtained with passively mode-locked Nd: YAG corresponds to a pulse width of 24 ps.

Ground state recovery time of organic dye BDN in the solvents dichlorethane and lodoethane was measured by means of the pump-probe technique. In this experiment, the integrated transmisson of a whole pulse train was measured as a function of time delay between pumping and probing pulses. The pulse train used in the experiment is generated by a passively mode-locked Nd : YAG laser. The results obtained agree with solutions of the rate equations for the three level model of bleachable dye.