In order to study the effect of different hole injection layers on the performance of flexible OLED devices, HAT-CN, PEDOT∶PSS and PEDOT∶PSS/HAT-CN were used to prepare flexible OLED devices and measure their performance parameters. The device structure is PET/ITO/HIL/TAPC (60 nm)/CBP∶Ir(ppy)3 (20 nm, 10%)/TmPyPB (45 nm)/Liq (2 nm)/Al (100 nm). PEDOT∶PSS was prepared by spin coating, and the other organic layers and cathode were prepared by vacuum evaporation. The results show that the flexible OLED device with PEDOT∶PSS/HAT-CN composite film as hole injection layer has the best performance. The maximum current efficiency and power efficiency can reach 84 cd/A and 76 lm/W. The research shows that the surface of the flexible substrate modified by PEDOT∶PSS has better continuity and smoothness, so that the phenomenon of leakage and short circuit is reduced; meanwhile, the composite film of PET/ITO/PEDOT∶PSS/HAT-CN has a higher transmittance in green light band, which can improve the extraction efficiency of the device. Moreover, the injection of carriers in this device is in balance, increasing the compound probability of electrons and holes.

The quality of the liquid crystal cell surface can affect the performance of the spatial light modulator. In order to accurately measure the liquid crystal cell surface, a low coherent carrier frequency interference method is proposed. Illuminating with a low coherent extended light source, forming a localized interference between the reference light and the test light, and adding a suitable carrier frequency by controlling the tilt angle between the reference surface and the surface of the liquid crystal cell, the carrier frequency interference fringe to be tested is obtained. The upper and lower surfaces of the air layer of the empty liquid crystal cell form a background interference fringe, so that a single carrier frequency interference pattern with a set of interference fringes is collected in the CCD camera. The frequency information of the background and the interference fringes is filtered by the spatial Fourier transform method, and the surface data of the fringes to be tested is extracted and restored. The surface of the liquid crystal cell made by dislocation bonding of two glass substrates was tested. The peak-to-valley value PV of the surface is 8.28λ, and the root mean square value RMS is 178λ. Comparative experimental results conducted by the method and ZYGO interferometer indicate that the differences of PV and RMS are 0.02λ and 0.004λ. The results obtained by the two measurement methods are consistent. The method can overcome the problem of multi-surface interference and measure the liquid crystal cell surface precisely.

There are two main types of bright dot defects in TFT LCD where liquid crystal is aligned by rubbing process, one is with particles and the other is without particles. The former can be reduced by enhancing cleaning, but the way to reduce the bright dots without particles is barely studied, so research and experiments are performed. According to macro and micro result, the alignment of liquid crystal is abnormal because of impurities as tall as 7.5~10 nm in the defect area. It can be deduced that the abnormally alignment is caused by the remaining of the back coating and the softener agent on the PI surface. Lastly, the rubbing cloth, in which the back coating and the softener agent are reduced, is tested in our factory. The results show that the bright dots w/o particles decrease from 06% to 0% when the back coating and the softener agent in the rubbing cloth are reduced, which proves the effectiveness of the laboratory results and the correction of the mechanism.

The dark corner Mura produced in the long-term process of THO reliability is studied. Because the Mura is visible on the TFT film surface, the capacitance change in TFT side during the process of reliability is mainly simulated and analyzed, and it is determined that it is a kind of electrical defect caused by the change of Cst. The mechanism is that water continuously enters into the PVX2 film and Cst increases at the same times, resulting in the reduction of the charging voltage and gray scale of TFT-LCD pixels. After improving the compactness of PVX2 film by decreasing pressure and increasing Si/N ratio in the deposition process, the change of capacitance decreases from 19.6% to 05% in the simulation reliability test because of the better capability of water resistance, which successfully solves the problem, avoids the risk in the reliability test and improves the product quality.

Uneven phenomenon will appeare when TFT-LCD (Thin Film Transistor-Liquid Crystal Display) TN curve panel used with WV (Wide View) polarizer. In the curve state, the mechanism of uneven color is speculated that there is non-uniform stress inside the panel when it is forced bending. 812.8 mm(32 in) TN curve panel is used as an example to discuss and research the uneven color phenomenon. For the problem of uneven phenomenon, the paper proposes two effective methods to improve uneven color phenomenon: First, increasing dark state brightness, the uneven color phenomenon is covered. The experimental results show that the adjustment circuit makes the display dark state brightness increase from 0.18 cd/m2 to 0.25 cd/m2, and the color uneven area is reduced from 30% to 10%. But the circuit adjustment makes the display response time increase to 2.5 times. The angle of the absorption axis is adjusted from 135° to 134.5°, and the area of color unevenness is reduced to 10%, and the adjustment of the absorption axis angle of 0.5° will reduce the contrast by 20%, which is detrimental to product performance influences. Homogenizing the stress in the TN curve panel is the second method, this paper improves the color unevenness by experiments such as increasing the cell gap of the panel, high temperature aging, and reducing the thickness of the glass substrate of the panel. The experimental results show that reducing the thickness of the glass to 0.2 mm and aging at high temperature for 72 h can make the color unevenness disappear without adversely affecting the product performance.

In order to solve the problem of multi-point data acquisition and intelligent monitoring in the family, an indoor data acquisition and monitoring system based on the internet of things is designed and implemented. The whole system takes ARM Cortex-M3 as the core processor, LORA module and WIFI module as the wireless communication equipment, LCD screen as data display terminal, and multi-channel wireless sensor nodes. Each node is connected with multi-channel sensors and control devices for data acquisition, control, display and upload. In order to increase the practicability of the system, the calendar function is added to provide time data for the calendar by acquiring network time or local time. The system is deployed indoors and tested. The experimental results show that the response time of the whole system is short, and the data acquisition and upload can be completed in less than 0.35 s. With the increase of the number of network nodes, the system can accurately identify and automatically complete the data acquisition, upload and equipment control without losing data packing. The whole system has high real-time, stability and security, and can better meet the needs of indoor data acquisition and intelligent control.

In order to solve the problem that the output signal collected in the IEPE sensor data acquisition system fluctuates greatly and is not suitable for direct acquisition, a conditioning circuit for IEPE sensor data acquisition system is designed. In the design of conditioning circuit, the micro-power supply circuit, low-power programmable front-end signal conditioning circuit and high-precision analog-to-digital conversion circuit are used, and FPGA is applied to the whole in the design of IEPE sensor data acquisition system, the convenience of operation and the stability of work are enhanced. The test results show that the IEPE sensor data acquisition system has higher data acquisition accuracy and lower electric power loss value. The data acquisition system has strong practical feasibility.

In order to better preserve the local contrast of the original color image and obtain the perceptual accurate grayscale image, a grayscale algorithm based on local contrast enhancement in contourlet transform domain is proposed. Firstly, in CIE Lab space, the conjugate gradient algorithm is used to optimize the objective function to get the global mapping function parameters, and preliminary grayscale image is obtained. Then, the multi-scale and multi-direction decomposition of color image and preliminary grayscale image are carried out by contourlet transform, local chromatic contrast ratio is used to enhance the contrast of directional detail image, and the enhanced detail image is obtained by inverse contourlet transform. Finally, the enhanced detail image is superimposed with the preliminary grayscale image to get the final grayscale image. Experiments on COLOR250 and adik database show that the proposed algorithm in this paper is superior to some typical algorithms in the existing literature, which can effectively preserve the contrast and structure information of the original image. The visual perception is natural for the grayscale image, whose subjectivity and objective evaluations are optimal.

For the problem of the high computational complexity of the superposed sparse representation based classification (SSRC), the advantages of the collaborative representation based classification (CRC) that has significantly less complexity than SRC is used and its recognition rate is similar to SRC. The face recognition based on superposed collaborative representation based classification (SCRC) is proposed. Based on a prototype plus variation model, the dictionary is assembled by the class centroids and the sample-to-centroid differences in the SCRC, which can leads to a substantial improvement on CRC. The experimental results show that, if the proposed prototype plus variation representation model is applied, the collaborative representation plays a crucial role in face recognition, and performs well even when the dictionary bases are collected under uncontrolled conditions and only a single sample per classes is available. Compared with the other algorithms, SCRC greatly reduces the computational complexity and ensures the recognition rates．

In order to protect the image edge details while filtering out the image salt and pepper noise, an image salt and pepper noise filtering algorithm based on connectivity detection is proposed. Since the gray value of the salt and pepper noise point tends to be different from the gray value of the normal pixel point, the algorithm compares the pixel point gray value with the gray value of the neighborhood pixel point to make the pixel point with larger difference. It is listed as a suspected noise point, and then it is judged whether the point is a noise point by detecting whether the suspected noise point is a part of the image connected area. The noise point is finally filtered out by the median filter. The algorithm can effectively distinguish the edge of the image area from the salt and pepper noise. The experimental results show that the algorithm can effectively remove the salt and pepper noise with the density ranging from 0 to 0.9. The peak signal-to-noise ratio of the algorithm can still reach 30 dB with the noise density of 0.9. It satisfies the requirement to effectively remove the salt and pepper noise of different density ranges while protecting the image details.

For the mass image clustering problem, the improved k-means algorithm is proposed and applied to the color image clustering. The algorithm consists of intraclass-interclass distance weighted k-means algorithm and nearest neighbor propagation clustering algorithm. In the experiment, the LBP map of the luminance component of the color image is reconstructed into a row vector and then constitutes a sample set. The improved k-means algorithm proposed in this paper is used to cluster the sample set. The experimental results show that the proposed method achieves higher clustering accuracy than the traditional methods in the evaluation indicators commonly used in multiple clustering methods. At the same time, the method is more efficient than traditional methods.

A discrete wavelet transform domain video watermarking approach based on extreme learning machine algorithm is designed. The approach includes watermark embedding and watermark extraction. In the watermark embedding process, the scene switching detection algorithm is used to realize non-overlapping frame extraction, and then the fifth-order discrete wavelet transform is applied to the luminance component of the non-overlapping frame to extract the fifth-order low-frequency sub-band coefficient matrix. The training data set is constructed by the coefficient matrix and the regression training is performed by the extreme learning machine. The output vector of the regression model and the watermark sub-block are used to correct the coefficient matrix. Finally, the sequence of video frames embedded in watermark is obtained by inverse discrete wavelet transform. In the watermark extraction process, a 5-level discrete wavelet transform is performed on the luminance component of the watermarked video frame sequence and the luminance component of the original video frame sequence, respectively. The watermark sub-block is obtained by extracting the difference portion of the two low-frequency sub-band coefficient matrices. A complete watermark can be obtained by reorganizing all the sub-blocks. A series of experiments show that the proposed approach is robust and extremely efficient.