In order to realize a large field, high resolution, and dynamic measurement of phase defects of optical components, a Reflective Shearing Point Diffraction Interferometer (RSPDI) was proposed. A lateral shear amount was introduced between the test and reference light to generate a high-density linear carrier frequency, and the information of the wavefront under the test was extracted from a single interferogram by an fast Fourier transform algorithm to determine the simultaneous measurement of the defects. The lensless imaging algorithm suppressed the diffraction effect of the defect, and the effective discrimination method of the defect type was summarized. In the experiment, an optical flat in a high-power laser system was detected by the RSPDI, and the correctness of the defect type criterion was verified. In addition, a laser-damaged optical plate was measured by the RSPDI and a Veeco NT9100 white light interferometer, with a relative error of 2.1%. The results indicate that the interferometer can be used to detect phase defects of large-aperture optical components effectively.

In order to realize the real-time measurement of haze and light transmittance for large-surface transparent samples, a haze and transmittance tester using LED as the light source and a double integrating sphere as the signal measurement part was designed. First, the theoretical derivation of the light transmittance and haze was completed on the basis of the system scheme. Second, simulation analysis of the light source spectral distribution affecting measurement, optical design, integrating sphere parameter design, and support adjustment mechanism design were completed, increasing the test accuracy and efficiency. Finally, calibration and sample testing of the haze and transmittance real-time tester were conducted. The test results show that the transmittance absolute error of the standard haze sample is less than 0.4% and its haze absolute error is less than 0.3%. Moreover, the transmittance deviation between the measured value and theoretically calculated value of the samples from the subscriber is also less than 0.4%, which confirms the claim of the subscriber.

In order to improve the signal-to-noise ratio for laser applications underwater, a grating resonator with an intra-cavity etalon was designed to compress the spectrum of XeF blue-green laser. First, a grating resonator was used to obtain a sub-nanometer laser linewidth, and then, a further spectral compression was performed by inserting an etalon into the grating resonator. The experimental results show that a linewidth of 0.14 nm can be obtained by using a 1 200 line/mm grating, and the linewidth can be further compressed to 39 pm by inserting an etalon, with a reflectivity of 50%, in the grating resonator. The final laser pulse energy is about 65 mJ and the pulse duration is 1 μs. Although the gain coefficient of the XeF (C-A) laser is low, the optical loss of the resonator can be reasonably controlled by optimizing the working parameters of the grating and the etalon, thereby a 10pm-level laser linewidth can be realized.

To analyze the relationship between the spatial light modulation characteristics of a liquid crystal phased array and the driving voltage using the liquid crystal continuous elastic deformation theory, in this paper, a method for solving the spatial distribution of a differential iteration liquid crystal director was proposed based on nonlinear least squares. Based on the electro-optical properties of the liquid crystal material, the distribution of the liquid crystal director at the threshold voltage was used as the initial value. For the direct calculation of the component of the potential shift vector along the Z axis with the driving voltage, the director space distribution of the liquid crystal was deduced under the action of an applied electric field. The phase modulation characteristic curve of the liquid crystal phase delay versus the applied voltage in the steady state was plotted. Finally, the theory was verified experimentally. Far-field beam pointing deflection control was achieved by controlling the driving voltage of the liquid crystal phased array. Using the driving voltage to directly solve the electric displacement vector, the initial value of the coupling between the director and potential shift was provided. This not only simplified the calculation but also accords with the motion process of the liquid crystal molecule under the action of the actual electric field as well as reduced the model error. The simulation results show that when the system error is 1.0×10-8, the mean time of iteration is 0.33 s under different voltages. When the driving voltage is 5 V, the angular error precision of the tilt angle of the liquid crystal molecules is improved by 0.09 rad compared to that achieved with the direct difference iteration method.

In order to study the mass transfer between the free gas and oil in a fluid power system and realize bubble mass measurement in oil, the radius limit of the bubble during measurement was derived from steady state and zero temperature gradient conditions. A high-pressure gas-oil mass transfer optical measurement system was designed, and the key algorithms for bubble identification and tracking were studied. First, the radius limit of a bubble during measurement was derived from the Biot number and Newton's Law of Cooling, and a high-pressure gas-oil mass transfer optical measurement system was designed. Then, a method to identify bubble elements from a video was demonstrated, based on the roundness of a circle and a set threshold. Finally, a minimum vector length method was used to track the center of a bubble between frames. Experimental results indicate that the system can achieve optical measurements under a pressure of 14 MPa, and the error for measuring the radius of the bubble is within 4%. The experimental method for gas-oil mass transport study can satisfy the system requirements of optical measurement of gas-oil mass transfer in a fluid power system.

In this paper, an optical imaging interferometry system was proposed for the accurate detection of the temperature, pressure, and concentration of gas hydrates. As part of this system, laser light with a wavelength of 1.65 μm, was transmitted across the sea surface through a fiber to excite gas hydrates on the sea bed. Ten spectral lines close to the 1.65 μm waves emitted by CH4 were collected and transmitted to the sea surface through another fiber. Five image interference fringes for CH4 were obtained on a Charge-Coupled Device (CCD) camera. Using the "rotation line temperature measurement method" and the Lorentzian line profile, and by selecting the gray values of two adjacent interference fringes, the temperature, pressure, and concentration of the CH4 gas hydrates on the sea bed could be obtained. The forward results of the imaging interferogram for CH4 show that the electron count on the CCD is 1.68×105, much larger than which results in the noise value of 400e. The maximum signal-to-noise ratio of the system is 291. The narrowband interference filter could distinguish 10 spectral lines in the waves emitted by CH4 within the 16° field of view. The measuring precision of the CH4 temperature and concentration values are 1 K and 3%, respectively. This study shows that an imaging interference system can be used in the remote sensing of seabed gas hydrates.

A metalens consisting of GaN nanopillars as basic cells was designed for subwavelength focusing at visible wavelengths, which provided an improvement over bulky and low-efficiency conventional imaging systems, and the possibility of its application to micro imaging. The metalens is composed of a GaN nanopillar array, which maintains the same height but gradually varying width. An analysis of the capacity and reasons for GaN nanopillars controlling the phase of incoming light was performed. In addition, based on FDTD methods, a simulation of 460 nm wavelength focusing in the transmission mode was performed. Subsequently, full width at half maximum (FWHM) are shown when the sizes of metalens are 3.75 μm× 3.75 m, 6.75 μm×6.75 m, 8.75 μm× 8.75 m, and 10.75 μm× 10.75 m, and the results are 1, 0.8, 0.5, and 0.3 m, respectively. The other focusing results, such as intensity distribution and focus spots, were also discussed. As a result, there are differences between designed focal length and simulated focal length, which changes under the different sizes of metalens. In conclusion, the designed GaN based metalens can focus in micron dimension effectively, which reduces the complexity of traditional imaging systems.

To improve the rapid response of a spacecraft and ensure that its trajectory-orbit unified design meets the requirements of speed, accuracy, and reliability, a rapid trajectory-orbit unified design method for spacecrafts was studied. First, the dynamic model of a spacecraft was constructed. The process and terminal constraints required for a spacecraft's trajectory-orbit unified design were analyzed and its flight procedure was given. Then, the traditional trajectory design numerical solution was improved. Through application of the improved Newtonian, Broyden rank-1, and steepest descent methods, the convergence of the methods was ensured when the initial value wad incorect. Finally, a multi-step and multi-layer method was considered, and the algorithm was further improved. The simulation result reveals that the proposed rapid trajectory-orbit unified design method for a spacecraft can be executed in 30 s, and the design orbital precision is 0.5 m. The algorithm is reliable, rapid, and accurate, and can be applied to trajectory-orbit unified designs of rapid-response spacecrafts.

A Position-Sensitive Detector (PSD) was used to measure two-dimensional displacement with high precision. A combination of three two-dimensional PSDs can implement measurements of six-degree-of-freedom relative displacement and angle. The measurement system mainly comprises three PSD sensors (including PSD photosensitive surface and light-emitting tube), low noise signal conditioning, and an Analog to Digital(AD) acquisition circuit. Three PSD orthogonal layout schemes were used to calculate the displacement and angle of relative motion through the spot position of three PSD photosensitive surfaces. A six-degree-of-freedom PSD calibration and test system was designed for the calibration of the PSD measurement system center offset and light tube installation error. The test results show that the measurement range of the PSD measurement system is better than ±10 mm and ±2.5°. The noise error of the PSD measurement system after calibration is 0.1 mm and the angle is 0.02°, while the absolute displacement error of the measuring system is less than 0.5 mm and the angle error is less than 0.14°. This proves that the system meets the requirements of 0.5 mm and 0.5°. In addition, the environmental adaptability of the PSD sensor was evaluated. It can satisfy the system requirements of noncontact, online, real-time, higher precision, and rapid speed, as well as strong anti-jamming and stabilization. The system is successfully used in the space microgravity active vibration isolation system of the Tian Zhou 1 cargo ship for relative motion measurement.

To solve the problems of low accuracy and folding rate of the main reflective surface of a traditional solid reflector mechanism, a deployable solid reflector mechanism was designed in this paper. First, the configurations of four types of deployable units were proposed, and their degrees of freedom were analyzed. After comparing and analyzing the advantages and disadvantages of the configuration schemes, configuration scheme 4 was defined as the basic configuration of the deployable unit, and its structure and deployment principle were designed. Then, based on the D-H coordinate transformation, a basic deployable unit kinematic model was established. According to the geometric constraints, the position vector equation was established; the relationship between rotation angles was established by the vector closed-form projection method; and the position, velocity, and acceleration were analyzed. Simultaneously, ADAMS software was used to establish the mechanism simulation model and verify the correctness of the kinematics analysis. Finally, based on the relationship between the rotation angle and the structure size of the mechanism, the effects of rotation angle and structure size on the folding rate were analyzed, and the rotation angles and of a 2.2 m deployable solid reflector mechanism are determined to be 78° and 30°, respectively. The center disk radius, the longitudinal dimension of hinge support hole, the transverse dimension, the longitudinal dimension of the switching link, and the folding rate are 245 mm, 8 mm, 40 mm, 18 mm, and 0.326 respectively, which provide a reference for the design of the deployable solid reflector mechanism.

A neural network-based nonlinear intelligent control method was proposed for electro-optical (EO) tracking systems to overcome the performance reduction caused by the complex nonlinearity existent in real systems. A radial basis function neural network supervisory control structure was employed, and the associated advantages and characteristics were expatiated in the proposed study. Furthermore, a tracking experiment was conducted for performance evaluation. The obtained experimental results demonstrate that the disturbance attenuation performance can vary from -28 dB to -51 dB within the disturbance frequency of 1 Hz and amplitude of 3°, which indicates an improvement of 15 dB over the PID control method. The results also indicate that EO tracking technology based on neural network control possesses the advantage of intelligent optimized tracking by learning a system's nonlinear information without human intervention. Hence, compared to conventional tracking algorithms, neural network-based EO tracking technology can be incorporated more effectively in complex application environments.

A small amount of defocus is generated in the optical system of high-resolution space cameras owing to the harsh mechanical environment of the launch section and the complex temperature environment of the orbit. To ensure the image quality of the camera and correct the defocus, a new high-precision focusing mechanism, consisting of a worm gear and a cam slider, was designed. The first stage of deceleration was achieved through the worm gear in this design. The drive's rotary axis was orthogonal to the optical axis, hence, the mechanism was more compact. Further deceleration was performed by the cam slider mechanism, and the rotary motion was converted into a linear motion; the linear motion accuracy of the slider was provided by a linear guide. The mechanism is small, has high precision, and is capable of self-locking in the case of a power failure. The mechanism motion pair was coated with a solid lubricant to prevent vacuum cold welding. According to the characteristics of the mechanism, the resolution and error of the focusing mechanism were analyzed. Mechanical and thermal vacuum cycle tests were carried out on the mechanism. Various precision indicators of the mechanism were compared and tested before and after the experiments. The results show that the mechanism performance is normal after the environmental test, the focus shift error is better than 5.8 μm, the linearity is better than 0.3%, and the yaw angle is better than 14″. In the thermo-optical test, the focal plane was adjusted by the focusing mechanism, and the image quality was significantly improved. The focusing mechanism can meet the accuracy and environmental adaptability requirements of the optical system.

During the automatic docking of cylindrical workpieces, it may be necessary to measure the orientation of workpieces accurately without targets. For this reason, a precision scanning method without targets was utilized. The point cloud data was obtained by a laser profile sensor, and the postures were determined by a comprehensive method. Based on Monte-Carlo simulation, the accuracies of the axis fitting method and bus section fitting method were compared, and a new comprehensive method was proposed further. In this study, the pitch angle was obtained by the axis fitting method, the deflection angle was obtained by the bus section fitting method, and the measurement accuracy was significantly improved. In addition, to balance the accuracy and efficiency, the relationship among the number of intercepted ellipses, the number of contour points on each ellipse, and the precision were discussed. Finally, a prototype experiment was conducted to prove the validity and accuracy of the comprehensive method. It is proved that the absolute accuracy of the axis attitude is better than 0.02°, and the standard deviation is less than 0.01°.

Trajectory accuracy is an important performance indicator for precision turntables. To improve the motion accuracy, this study improves the traditional control algorithm from the perspective of trajectory planning and motion control. For trajectory planning, the S-curve trajectory planning equations are deduced, and the trajectory planning parameter values are given in combination with the dynamic constraints of the turntable. This provides the motion control algorithm with trajectory commands that meet the dynamic requirements. For motion control, on the basis of traditional double closed-loop feedback control, DOB disturbance compensation and feedforward compensation are added to improve the servo performance and motion accuracy of the turntable. After describing the design process of the trajectory planning and motion control algorithms in detail, the two algorithms are integrated, and the specific implementation steps are provided. The performance tests of the multiple groups of algorithms are performed using the harmonic turntable and the RV turntable. The experimental results show that, compared with the traditional control methods, the proposed method can significantly improve the dynamic accuracy of the turntable, thus verifying the effectiveness of the algorithm in improving the motion accuracy.

Forward kinematics of the near singular position of the Stewart parallel manipulator based on the Newton method or quasi-Newton method are not converge; Newton downhill method is time-consuming, sometimes. To resolve the situation, a method of applying the step-adjusting Newton method to a parallel manipulator is proposed. Firstly, the process of forward kinematics of the Stewart parallel manipulator based on the step-adjusting Newton method was designed. Then, the fewest iterative steps in the forward kinematics of sixty-four kinds of utmost poses was used as the objective function. The initial values of the step matrix and geometric parameters were taken as design variables by genetic algorithm to obtain the optimal values. Numerical examples show that when the absolute error of the rod length was set as 0.01 mm, in solving the forward kinematics of sixty-four kinds of utmost poses, the Newton method or quasi-Newton method did not converge on six kinds of utmost poses. The Newton downhill method takes longer than 2.0 ms on ten kinds of utmost poses, while the time taken by the step-adjusting Newton method was less than 2.0 ms. The step-adjusting Newton method provides theoretical guidance for the forward kinematics of the Stewart parallel manipulator in real-time occasions.

Texture is the embodiment of the distribution of surface microstructure. Tactile texture is a crucial factor to consider for improvement of the perception and exerting control over the environment of the material. In this study, a highly accurate and responsive tactile sensor was designed and fabricated using the inverse magnetostrictive effect of Galfenol to detect the surface microstructure of different objects and determine their roughness and fine density. Based on the Euler-Bernoulli beam dynamics theory, linear constitutive equations of magnetostrictive materials, and Faraday's law of electromagnetic induction, a relationship was established between the microstructure of the textured surface and the output voltage. The experimental results showed that the sensor had high sensitivity to object roughness recognition for roughness greater than 6.5. For fineness greater than 6, the method of extracting harmonic frequency was highly sensitive in identifying fineness. However, when the fineness was less than 6, the method of extracting the center of gravity of the power spectrum had high sensitivity for the identification of fineness. These results showed that the signal obtained by the sensor could be used to characterize the rough-smooth and sparse-fine attributes of different objects by eigenvalue extraction.

In order to improve the surface quality of micro machined parts, micro milling experiments were designed to study the variations of surface roughness and sidewall residual burrs of micro grooves. First, from a theoretical point of view, the model of processed surface formation and micro machined surface roughness theory were introduced for the purpose of proposing changes to the sidewall for tool run-out. Then, typical parts were selected to process the micro straight grooves by the developed 3-axis linkage micro milling machine tool; the spindle speed, axial cutting depth, feed speed, tool run-out, and material organization structure on surface roughness were studied in depth. Finally, the milling surface roughness of the grooves was researched by adopting multiple factor orthogonal tests and analysis of extreme difference. The best surface roughness values for the three kinds of micro grooved bottoms made of aluminum alloy, steel, and titanium alloy are determined as 1.073-1.481 μm, 0.485-0.883 μm, and 0.235-0.267 μm, respectively. Without tool run out, the maximum height of the micro groove wall burr is found as 7.637 μm, and the maximum height of the micro-groove wall burr is found to be 21.79 μm when the tool has a radial composite runout of 0.3 μm. The experimental results show that the parameters are listed in the effective order given below: feed rate, spindle speed, and axial cutting depth. With the gradually increasing feed speed and axial cutting depth, the surface roughness Ra of the groove bottom gradually increased. However, increasing the spindle speed caused Ra to decrease and then increase. Under the same conditions, if the arc blades are without wear, the tool run-out on the micro straight groove sidewall surface quality has a greater impact. At the same time, different metal material properties are also potential factors affecting the surface quality of the micro channels.

The modulation transfer function (MTF) is an important indicator that describes the performance of an optical imaging system. To avoid the effect of atmosphere on the measurement readings obtained by the on-orbit MTF on a meteorological satellite, the calculation method considering the lunar limb is investigated. First, we select the moon image obtained by FY-2G/VISSR as the input for the Knife-edge method. Then, we separate the image into sub-images according to different pixels of the detectors. Finally, based on different sets of sub-images, we obtain different results, which can improve the comprehensiveness of the MTF results. The experiments indicate that the result calculated by the Moon image is better than the result from the Earth image, which are 0.320 1 and 0.216 4, respectively. The standard deviation of MTF results from continuous sub-image is 0.013 31, which indicates the stability of performance of the system.

The navigation camera is a sensitive component that is crucial for deep space exploration. This report describes a method for the comprehensive improvement of the performance of such a camera by increasing their sensitivity. In particular, the approach improves the temporal resolution of this instrument and simultaneously solves the target detection problem under high dynamic conditions. Firstly, the imaging quality of a navigation camera is analyzed according to the operating mode and the performance characteristics of an Electron Multiplying Charge-Coupled Device (EMCCD). In addition, a theoretical model of a target's signal-to-noise ratio (SNR) is established. Secondly, the study focused on the design method for an EMCCD navigation camera. Key technologies are illustrated such as the high-frequency high-amplitude drive circuit of an EMCCD, analog front circuitry, TEC vacuum cooling, timing control, and data processing. Finally, the experimental results show that the optimal target SNR is 68.6 dB at M = 10 and that the moon can be imaged using an integration time of 1 ms with a 13 mm aperture. The navigation camera satisfies the requirements of short integration time imaging under the high dynamic conditions associated with deep space exploration.

While removing noise, the loss of useful information is inevitable, especially the edge information of linear images. To maximize the image and minimize the loss of useful information according to the edge features of the testing image, a window with appropriate size and shape was selected to perform median filtering and other operations. It is innovatively proposed to apply a Hough transform to the filtering window shape selection of the wheel crack CT image, aiming at the image of the single direction contour. The Hough transform was used to detect the direction of the contour, and the corresponding shape window was used to filter the crack. This method was compared with the traditional method, and the image with better visual effect can be obtained after pre-processing by Hough transform. In this paper, to improve the filtering effect for images with multi-direction edge contours, an oblique filter is designed according to the pixel gradient. The data shows that the peak signal-to-noise ratio (PSNR) of the image after proposed filtering is improved by 4-6 compared with the traditional median filter. The structure similarity (SSIM) is increased by approximately 1%-2%. Three-dimensional images are obtained using the stack of CT images of buckwheat slices. The filtering result of the proposed method is favorable by the contrast of before and after.

Objects intruding railway clearance pose great threat to normal railway operations. Identifying intruding pedestrians within the railway clearance limit was of great significance to ensure the safety of railway operations. The existing railway intrusion detection system only detected the intrusion, but did not distinguish whether it was a true alarm of pedestrian intrusion or false alarm caused by light interferences. To reduce false alarms, a training and test set of the alarm image samples were established. A pedestrian classification algorithm based on improved deep convolutional network, trained with combined features of HOG and high-level Alex was then proposed. First, an improved AlexNet deep convolutional neural network model was introduced to extract high-level Alex features by automatic learning; the extracted features were then combined with HOG features to form the combined features of Alex-HOG. Finally, the combined features were used to train the classification network. Experiments on the test set show that the proposed method has a high recognition accuracy of 98.46% in 3.78 s for 1 498 test image samples. The improvements in both accuracy and real-time performance will greatly reduce the false alarm rate of the railway intrusion detection system.

High dynamic range (HDR) scene reproduction has been widely used in consumer electronics, virtual reality, photography, and computer vision. An HDR scene reproduction algorithm based on hybrid mapping was proposed to address the problem of poor overall appearance and local details of an existing work. First, HDR scenes were transformed from the RGB to HSV color space, which converts the color information to hue, saturation, and luminance value components. Second, the luminance values were transformed into a logarithm domain using a special logarithm function, and an adaptive arctangent function was used to compress the global dynamic range, which then enables achieving good overall appearance. Then, the improved single-scale Retinex algorithm was applied to local detail adjustment, which enhanced the details of the bright and dark areas. Finally, the saturation component was adjusted, and the color information was restored from HSV to RGB color space. Ten types of HDR scenes were tested in the experiments, and the obtained results were compared both subjectively and objectively. The experimental results demonstrate that the proposed method enables good visual appearance and preserves more details. Moreover, the proposed algorithm has a low computation cost. The proposed algorithm is better than five traditional methods and can produce desirable images in most HDR scenes.

To solve the problems of storage and transmission of elemental image arrays in the integrated imaging system, an elemental image array coding and decoding algorithm that combines imaging geometry features was proposed. First, the offset of the corresponding image points in different elemental images were determined according to relevant physical parameters in the process of collecting the elemental image arrays. In addition, adjacent elemental images of each row in the element image arrays were grouped, and their coding sequences were determined. Then, the prediction image of the elemental image to be encoded was determined, and the residual between the elemental image and its prediction map was calculated. Finally, the residual was coded using High Efficiency Video Coding (HEVC). The traditional HEVC intra-frame prediction coding algorithm, which was formed by all elemental images in the elemental image array, was compared with the proposed HEVC coding algorithm. The experimental results show that the quality of decoded images using the proposed algorithm improved from 10 dB to 25 dB under the same bit rate, demonstrating the efficient coding performance of the proposed algorithm.

To prevent over-deformation and to solve occlusion problems that are difficult to solve for correlated filtering tracking algorithms in the tracking field, an improved multiscale target tracking algorithm based on PSR and objective similarity was proposed in this paper. The proposed method combined a traditional correlation operation, peak side lobe ratio, with a perceptual hashing algorithm to tackle problems such as target occlusion, over deformation, and other complex scene judgments. Experimental results using the OTB-2015 demonstrate proposed algorithm's reliability and integrity of the target trajectory. The accuracy and robustness of our algorithm is better than that of Kernelized Correlation Filter (KCF) tracking algorithms. This paper presents a novel idea for occlusion detection in the target tracking field.

To realize a fast and high-precision angular measurement method, this study focused on the signal processing algorithm of a single-track absolute encoder, which can be easily applied to photoelectrical theodolites. Using this algorithm, the grating line center and pulse width could be quickly estimated with only two sampling points of the rising and falling edges, by setting a threshold for the observed grating data. To improve the existing subdivision method, a high-resolution subdivision algorithm for angular measurement was provided by calculating multiple grating line centers. We also analyzed the algorithm's time complexity and anti-noise ability by simulation. Experiments were performed to evaluate the measurement precision, output stability, and running speed. The results showed that the standard deviation of error was within 2.5 seconds, and the rate of angular output was 25 times per second. This algorithm is easy to implement and has merits of both limited computation and high-precision. Thus, a new angular measurement signal-processing algorithm, which can be applied to a single-track absolute encoder for a photoelectrical theodolite or total station instrument in the future, can be developed with the results of this research.

A general automatic fault recognition algorithm based on sparse-coding-based spatial pyramid matching and Genetic Algorithm Optimized Support Vector Machine (GA-SVM) was proposed for fault detection of the bogie block key, dust collector, and fastening bolt in the Trouble of moving Freight car Detection System (TFDS). First, the image of a sample was divided into patch areas in different scale spaces, and the Scale-Invariant Feature Transforms (SIFT) of each patch area was extracted. Sparse coding was then performed by iteratively learning dictionaries using the SIFT features of randomly extracted samples. Second, principal component analysis was used to define the contribution of the encoded features towards fault recognition accuracy and reduce the dimensionality of the coding features. Then, the SVM classifier was trained using the reduced dimension features after coding and optimization with the genetic algorithm. Finally, the trained classifier was used to detect the bogie block key, dust collector, and fastening bolt faults from their images. The experimental results show that the algorithm can adaptively recognize the three different kinds of faults. The fault recognition rates were 97.25%, 99.00%, and 97.50% for bogie block key, dust collector, and fastening bolts, respectively. This technique is robust to noise and illumination changes and can meet the actual detection requirements of a vehicle's structural faults.

To solve the low accuracy problem of remote sensing image scene classification due to small sample sizes, a classification method was proposed based on Multiscale Features Fusion (MSFF). First, the remote sensing images were scaled to obtain several different scale images of the same remote sensing image. Thereafter, they were inputted into a Deep Convolutional Neural Network (DCNN) for convolutional operation, and the different scale features of the convolutional and the fully connected layers were reduced and coded or average pooled. Finally, the different scale features were coded and fused, and a multikernel support vector machine was used to classify the scenes. In the two public remote sensing image data sets UCM Land-Use and NWPU-RESISC45, the highest classification accuracy of the experiment are 98.91% and 99.33%, respectively. This method can use image features of different scales and low, middle and high-level semantic representations combined, thus the fusion feature is more recognizable. Furthermore, the use of a multikernel support vector machine improves the generalization of the deep network learning ability, so the classification effect is better.

To detect micro defects in thick-section carbon fiber reinforced composite (CFRP), the recurrence analysis method was used to analyze ultrasonic signals of tested CFRP. First, small holes were made to simulate micro defects, and an ultrasonic pulse echo method was adopted to test these simulated defects of different sizes. Then, the signal segments around the defect position were selected, and recurrence analysis was performed after proper parameters like embedding dimension(m), time delay(τ), and threshold(ε) were chosen. The recurrence plots (RPs) of defect-free signals were compared with those of defective ones and, according to the physical meanings of recurrence quantification analysis (RQA) variables, the changes that appeared in RPs were explained. Finally, ultrasonic transducers with different frequencies were evaluated to determine which one has the best performance. The results show that a 7.5 MHz resolution series transducer is the best choice in our experiment, and while m=7, τ=2, and ε=2 , the defects may cause dark areas, white bands, and longer diagnosis structure in RPs and correspondingly larger RQA variables. More concretely, the ultrasonic signal of defect-free thick-section CFRP appears in a chaotic state, while defects may break this state and lead to another one. The results will lay a foundation for the quantitative identification and classification of real micro defects.