
In the image super-resolution reconstruction, many methods based on deep learning mostly adopt the traditional mean squared error (MSE) as the loss function, and the reconstructed image is prone to the problem of fuzzy details and too smooth. In order to solve this problem, this paper improves the traditional mean square error loss function and proposes an image super-resolution reconstruction method based on multi-scale feature loss function. The whole network model consists of a DenseNet-based reconstruction model and a convolutional neural network which is used to optimize the multi-scale feature loss function. Taking the reconstructed image and the corresponding original HD image as the input of the convolved neural network in series, the mean square error of the different scale feature images obtained by convolution of the reconstructed image with the corresponding original HD image was calculated. Experimental results show that the method in this paper is improved in subjective vision, PSRN and SSIM.
Quantitative methods have been formed to measure the spatial resolution in spectral imaging field, but the results may differ with imaging position change when the resolving power of detector is deficient. Based on the spectral images of black-white lines under accurate shift, a new method for detecting the spatial resolution of area array spectral imaging equipment is proposed. This method presents the curve showing the gray level variation with the displacement of object for a single pixel, and can obtain all the results of gray level distribution among pixels theoretically. Through one type of curve division, the density value of the black-white lines which can be discerned on any imaging position will be obtained. This method has overcome the shortcoming of current methods, and its fea-sibility is validated by an experiment for one area array spectral imaging equipment.
Convolutional neural network (CNN) has recently achieved a great success for single image su-per-resolution (SISR). However, most deep CNN-based super-resolution models use chained stacking to build the network, which results in the fact that the relationship between layers is weak and does not make full use of hierar-chical features. In this paper, a multi-path recursive convolutional network (MRCN) is designed to address these problems in SISR. By using multi-path structure to strengthen the relationship between layers, our network can ef-fectively utilize features and extract rich high-frequency components. At the same time, we also use recursive structure to alleviate training difficulty. In addition, by introducing the operation of feature fusion into the model, our network can make full use of the features extracted from each layer in the reconstruction process and select the ef-fective features adaptively. Extensive experiments on benchmarks datasets have shown that MRCN has a significant performance improvement against existing methods.
For camera-basedimaging, low resolution and noise outliers are the major challenges. Here, we proposea novel super-resolution method-total generalized variation (TGV) super-resolution based on fast l1-norm dictionaryedge representations. First, anisotropic diffusion tensor (ADT) is utilized as high frequency edge information. The fast l1-norm dictionary representation method is used to create dictionaries of LR image and the corresponding high frequency edge information. This method can quickly build dictionaries on the same database, and avoid the influ-ence of outliers. Then we combine the edge information ADT and TGV model as the new regularization function. Finally, the super-resolution cost function is established. The results show that the algorithm has high feasibility and robustness to simulation data and SO12233 target data. It can effectively remove noise outliers and obtain high-quality clear images. Compared with other classical algorithms, the proposed algorithm can obtain higher PSNR and SSIM values.
When the electro-optic tracking system is used for space target tracking, it is difficult to extract the target from the field of view occasionally due to the impact of electromagnetic interference, cloud cover or earth shadow etc., and the closed-loop tracking system can barely work in severe cases. At this point the predicted orbit can be used to guide the system to ensure smooth scanning and tracking. In this paper, random sample consensus (RANSAC) algorithm is introduced, which has been widely used in feature extraction in computer vision, to achieve higher prediction accuracy. The loss function of RANSAC algorithm is improved and the WRANSAC algorithm is proposed according to the distribution of observed data, which is used to deal with the limited observation data in real time to track the space target. After the algorithm is adopted, the fault tolerance of observation data is improved and the sensitivity of the model is reduced. The accuracy and robustness of the prediction results are much better than that of the least squares method. The validity of the WRANSAC algorithm is proved by the comparison between the predicted trajectory and the actual trajectory.
In allusion of the video jitter problem caused by platform motion, a video stabilization technique based on optical flow sensor is presented. Firstly, the scheme improves the general optical flow sensor to output accurate motion vectors under rotational motion, then motion vectors between adjacent frames are obtained by using the optical flow sensor. The real-time translation and rotation information of the main camera are calculated through coordinate transformation. Secondly, the method compensates the motion of video sequences to attain stable video sequences, and finally realizes video stabilization. Experimental results indicate that, compared with the unstable image, the peak signal-to-noise ratio (PSNR) is increased by 11.86 dB. In the case of obvious video jitter, the scheme can significantly reduce the jitter between video sequences. The method which has the characteristics of salutary video stabilization can meet the performance requirements of video stabilization and improve the capacity of disturbance resistance for platform.
Due to the limitation of equipment, the resolution of depth map is low. Depth edges often become blurred when the low-resolution depth image is upsampled. In this paper, we present the pyramid dense residual network (PDRN) to efficiently reconstruct the high-resolution images. The network takes residual network as the main frame and adopts the cascaded pyramid structure for phased upsampling. At each pyramid level, the modified dense block is used to acquire high frequency residual, especially the edge features and the skip connection branch in the resi-dual structure is used to deal with the low frequency information. The network directly uses the low-resolution depth image as the initial input of the network and the subpixel convolution layers is used for upsampling. It reduces the computational complexity. The experiments indicate that the proposed method effectively solves the problem of blurred edge and obtains great results both in qualitative and quantitative.
For the problems of needing pre-training and poor robustness to rotation and illumination changes of various improved algorithms based on local binary pattern (LBP), this paper presents a new texture classification algorithm by integrating the completed local binary pattern (CLBP) and the local geometric invariant features of the image surface. In our algorithm, the local geometric invariant features are first computed. Then the computed results are further quantified and encoded to make combination with the CLBP histogram. The proposed algorithm can ex-tract image macroscopic and microscopic features simultaneously, and it has the properties of not significantly in-creasing feature dimension, without pre-training, and invariance to the rotation and illumination changes. Experi-mental verifications are conducted on two standard texture databases, and the results demonstrate that the pro-posed algorithm outperforms the comparative classification algorithms in classification accuracy and robustness.
In the seed breathing CO2 detection system, the traditional method cannot measure the concentration of CO2 in the seed breathing in real time. According to the characteristics of seed breathing CO2, a seed breathing detection system based on virtual instrument LabVIEW is designed based on tunable diode laser absorption spec-troscopy (TDLAS). The system mainly includes laser light source and its controller, seed breathing container based on multiple reflecting pool structure. The upper computer software is mainly set with data acquisition, signal processing, concentration inversion and other functional modules, in which the concentration inversion uses the or-thogonal vector phase-locked amplification algorithm to avoid the error caused by the phase difference between the reference signal and the signal to be measured. The experimental results show that the CO2 detection system for seed respiration implemented by virtual instrument software can effectively detect the change of seed respiration, and has good anti-interference and stability, which lays a foundation for the subsequent experimental research and development.
According to the Schupmann’s achromatic theory, a calculation method of off-axis four-mirror diffractive imaging optical system is introduced. By using the method, an optical system which has an aperture of 1 m, F-number of 8, full field of view of 0.12°, waveband of 582.8 nm~682.8 nm is designed. The results show that the chromatic aberration is corrected effectively. The modulation transfer function (MTF) is more than 0.53 in the range of 50 lp/mm, and the RMS radius of diffusion spot is less than the airy radius. It demonstrates that the image quality of this system is close to the diffraction limit. It is analyzed that the processing of diffractive primary lens and diffrac-tive correct mirror can be realized by traditional lithography and diamond turning, respectively. Monte-Carlo simula-tion of tolerance analysis is carried out, it determined that the tolerance error mainly originates from the tilt angle of relay mirror, which provides guidance for the process of assembling and adjusting. This system has the advantages of broadband, high image quality, which can provide references for the development of reflective diffractive imaging optical system.